diff --git a/aft_wave_v2/coin_real_4x__agreement/training/ARTIFACT_MANIFEST.local.json b/aft_wave_v2/coin_real_4x__agreement/training/ARTIFACT_MANIFEST.local.json
index 1e67193dc245193ec703f55fc97a91f45946bbd6..f98cd23cae19565d89a468a48e3f98fad018c72a 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/ARTIFACT_MANIFEST.local.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/ARTIFACT_MANIFEST.local.json
@@ -3,13 +3,9 @@
   "remote_prefix": "aft_wave_v2/coin_real_4x__agreement/training",
   "local_folder": "/workspace/wave/training",
   "files": {
-    "ARTIFACT_MANIFEST.local.json": {
-      "size": 35054,
-      "sha256": "358d03f66da22d79c9aeaef103fd13ac6757c23ca03bc8a8e1ce8eaa088edc61"
-    },
     "TRAINED.json": {
       "size": 964,
-      "sha256": "63c842682fc4bf7d162bfb0e12a0cedf0b85eacb3c5f9794acba00eb0f918ea8"
+      "sha256": "edf924a6e4b875e87894052b80f255fdc708da5b2e770dda415cb89eb082e7ca"
     },
     "axolotl.yaml": {
       "size": 1211,
@@ -25,11 +21,11 @@
     },
     "checkpoints/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "16cc65acfba87debf950e49fc05f5dce80ea16745d793c826ed96faac31a47a1"
+      "sha256": "daeafd5d686b4b39c5b3dd656eca36514ea93ef225c5afab298c74bbfab4b632"
     },
     "checkpoints/chat_template.jinja": {
       "size": 1532,
@@ -41,11 +37,11 @@
     },
     "checkpoints/checkpoint-128/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-128/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "e7202ce5ed3e0cfd2be9bb413acb17a214475b81453c800a2e3c69d4ff5d7959"
+      "sha256": "67174eba71aaa201dd7aff62c7fbd3596ad8b4afeba399abae18cc3a0dd00888"
     },
     "checkpoints/checkpoint-128/chat_template.jinja": {
       "size": 1532,
@@ -53,7 +49,7 @@
     },
     "checkpoints/checkpoint-128/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "eb131e3cb1d8065ea88ab03c3129c1bdf7f01361df81ae0569e4c2b6008fa036"
+      "sha256": "c48bb4d7fcaca8cbcee6c332057a50925e65b15baae813555c4b32ddf641b01b"
     },
     "checkpoints/checkpoint-128/rng_state.pth": {
       "size": 14645,
@@ -77,7 +73,7 @@
     },
     "checkpoints/checkpoint-128/trainer_state.json": {
       "size": 56227,
-      "sha256": "03387942f4c2a7b6ad45075f88a38b298c38a135b13eb87e922d7f07579218dc"
+      "sha256": "42baeabcb054047def81fc174ed053695ed3fa2f56b0d6a652cd2b7ea11addde"
     },
     "checkpoints/checkpoint-128/training_args.bin": {
       "size": 8273,
@@ -89,11 +85,11 @@
     },
     "checkpoints/checkpoint-160/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-160/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "50621401dbfe149cde6d4a622a952bc693f8c1c46da9363b866907fa002871d4"
+      "sha256": "4f45bebc7b5c77c792dbf05c98745f04359a2b3462e61a009e38cbe69159480d"
     },
     "checkpoints/checkpoint-160/chat_template.jinja": {
       "size": 1532,
@@ -101,7 +97,7 @@
     },
     "checkpoints/checkpoint-160/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "5dafd1343477b7fcb5e94b4e4a0059e8fabb1e49d85a0acc3c63a5a789736f11"
+      "sha256": "5ac494318568079c3e19c022c29f18c065bb4ba6884c43ff2ef297462392b14f"
     },
     "checkpoints/checkpoint-160/rng_state.pth": {
       "size": 14645,
@@ -124,8 +120,8 @@
       "sha256": "a9c33caced7d456d62f444f1f50d8ffdf898b5a0e31c26d17c2f3edd05e45f64"
     },
     "checkpoints/checkpoint-160/trainer_state.json": {
-      "size": 70212,
-      "sha256": "2a6e915d2b03c5ef72eba13329c450186c30fb8516a29f8cc04316eb8365021a"
+      "size": 70210,
+      "sha256": "837c450cb4078aa28309f4999542aedfa831db547cb0d7b227f3a37c393d0355"
     },
     "checkpoints/checkpoint-160/training_args.bin": {
       "size": 8273,
@@ -137,11 +133,11 @@
     },
     "checkpoints/checkpoint-192/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-192/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "1dd775d8b35d84d9d817a4a74f08f039adb13a2320077366a52bad07c43ebf8b"
+      "sha256": "dd65c9138a23ee4b709df1f879e1ce2a71a67a3461f4f7ed0fae00ba0eb853d5"
     },
     "checkpoints/checkpoint-192/chat_template.jinja": {
       "size": 1532,
@@ -149,7 +145,7 @@
     },
     "checkpoints/checkpoint-192/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "1d796a53e0037fe564b92a554d12c2094e42094df256517710edaa58b7d48d31"
+      "sha256": "0e74bd30f32259d424a0528a8fc03e88152c9448882265722a55166fb622af3a"
     },
     "checkpoints/checkpoint-192/rng_state.pth": {
       "size": 14645,
@@ -172,8 +168,8 @@
       "sha256": "6dcb03db82c34201f9966e9d9c444c1957b95cd53c5ea9ed3b950f223ea74fbe"
     },
     "checkpoints/checkpoint-192/trainer_state.json": {
-      "size": 84203,
-      "sha256": "3544cc06d91f290b9129f48d31697c72cbf5c7bfccb06427072589859f519397"
+      "size": 84216,
+      "sha256": "76ffe843ba65218315ddfe5c2695fb2b03f1cf28ed59a4af495c5f42f439624f"
     },
     "checkpoints/checkpoint-192/training_args.bin": {
       "size": 8273,
@@ -185,11 +181,11 @@
     },
     "checkpoints/checkpoint-224/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-224/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "b8eecfc00fed67bfc79ccbe2c4580025f3a5eb09de1478ed4d237cdce0a676c0"
+      "sha256": "16e7d16f507266e4de0618e13529f0f6852f7dfacc533b3d9beb61399405123d"
     },
     "checkpoints/checkpoint-224/chat_template.jinja": {
       "size": 1532,
@@ -197,7 +193,7 @@
     },
     "checkpoints/checkpoint-224/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "ab103e3591277e8cd1e5fa934a5bc9882dafbe277490d1a7a13579f389795dd1"
+      "sha256": "f456a3b4b0b5cfe5bb7061718fafa69fb3393a4a4055872f85b783cf6135b255"
     },
     "checkpoints/checkpoint-224/rng_state.pth": {
       "size": 14645,
@@ -220,8 +216,8 @@
       "sha256": "14c3d2181ef61e0cf34f6fe59f885a012f56477195e0bfcf74f965a091355c99"
     },
     "checkpoints/checkpoint-224/trainer_state.json": {
-      "size": 98212,
-      "sha256": "70be8685e7805f1ac444171a50c1c168a34bb409e22af079d8183c9aed8c41db"
+      "size": 98236,
+      "sha256": "57eb4a9f8711b9449eb9c8e699bb3ce11e013cb7c5ed53b4ec205fdb2524b50c"
     },
     "checkpoints/checkpoint-224/training_args.bin": {
       "size": 8273,
@@ -233,11 +229,11 @@
     },
     "checkpoints/checkpoint-256/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-256/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "1c74a0c1890d16768995e589354d66ad0aaf896432a0f8708036dcf170656173"
+      "sha256": "de74ce97413c0093f4f76198ae64237c0da203ba9a293aa2b281e66d918ca952"
     },
     "checkpoints/checkpoint-256/chat_template.jinja": {
       "size": 1532,
@@ -245,7 +241,7 @@
     },
     "checkpoints/checkpoint-256/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "143ebc082a4dcdfe72a9b44639352173fc591f7676a6655b84d556a3b293a9d7"
+      "sha256": "a2e59a33e5d93b9218821342e940feec395f412661f80e94be666601cd30c6c3"
     },
     "checkpoints/checkpoint-256/rng_state.pth": {
       "size": 14645,
@@ -268,8 +264,8 @@
       "sha256": "a1db2747ccfd098f33eee7a195bbd033988a2849ce30491603e0d31cbe794a14"
     },
     "checkpoints/checkpoint-256/trainer_state.json": {
-      "size": 112250,
-      "sha256": "ee7da18d6f8a75f21c96263765db9e618129ea823a617c6ad49a8b8e490a99b7"
+      "size": 112286,
+      "sha256": "2437fc722ed7350b2570a2bdb13ab9f6fc9d680ad29c3ad87cb5ac95dba68124"
     },
     "checkpoints/checkpoint-256/training_args.bin": {
       "size": 8273,
@@ -281,11 +277,11 @@
     },
     "checkpoints/checkpoint-288/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-288/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "f742ce5e3dc5e92cf1d1b5194d55616562458401c38397bb3b1f150a3613ed30"
+      "sha256": "da2bdb4f81462625bb226aa42eb809bea433a88b8d19660ad0cb52a8f25dfb1f"
     },
     "checkpoints/checkpoint-288/chat_template.jinja": {
       "size": 1532,
@@ -293,7 +289,7 @@
     },
     "checkpoints/checkpoint-288/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "9c6a83d5d9751f22dbe1bd2713e7e188432fb354065c57c9d49029f1c114f05b"
+      "sha256": "c144b815c2b7bcbf81221112a3f40c21c4e01914d789b90a94c663d34e6977b2"
     },
     "checkpoints/checkpoint-288/rng_state.pth": {
       "size": 14645,
@@ -316,8 +312,8 @@
       "sha256": "f1eed36b4e578b166623815e5accad0aac99aa3b265255a7471d4f0a7c22b297"
     },
     "checkpoints/checkpoint-288/trainer_state.json": {
-      "size": 126335,
-      "sha256": "8a07ce94ffd53a7217a9731c5040e5dc755a8f4efdc057c6854e4fb191dec158"
+      "size": 126385,
+      "sha256": "d34f36262ee1658dbadffb114672a386ee3cc513554f05b57c3f7b9c9d453ed7"
     },
     "checkpoints/checkpoint-288/training_args.bin": {
       "size": 8273,
@@ -329,11 +325,11 @@
     },
     "checkpoints/checkpoint-32/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-32/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "ee7003d2997d35b1193b3e3463a4d9c2b6d209eee799d44f21ec131ffe7ff39a"
+      "sha256": "32aa7a1bdc40de6c271cdaf4ba2e825940729a8400f15293192a0ab0fd0be3c4"
     },
     "checkpoints/checkpoint-32/chat_template.jinja": {
       "size": 1532,
@@ -341,7 +337,7 @@
     },
     "checkpoints/checkpoint-32/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "1f5489845cc180566855dfb58a67b01f34eb2d8fa011a26887196570356e40f0"
+      "sha256": "d2c09fd208676c30f1a17a813e8f50e6c1234fdc5a8449106274744a5bc40156"
     },
     "checkpoints/checkpoint-32/rng_state.pth": {
       "size": 14645,
@@ -364,8 +360,8 @@
       "sha256": "d9025748b8b157d8490d18576a010318e827c122bd7c9d4da1001a4e203d8a9b"
     },
     "checkpoints/checkpoint-32/trainer_state.json": {
-      "size": 14414,
-      "sha256": "dedbe24bdbea1ce6561cb65724d0bdea4ca32569be518904d97085b21573b6d6"
+      "size": 14406,
+      "sha256": "12d755108b595d32441e426440b368f10f87ac530ad04e3b58bd04d5896495d0"
     },
     "checkpoints/checkpoint-32/training_args.bin": {
       "size": 8273,
@@ -377,11 +373,11 @@
     },
     "checkpoints/checkpoint-320/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-320/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "a6e8332b819be957190c74414784873a8ade3712d142e98f560d80f243855b7d"
+      "sha256": "ecfb48f90e7121908ec1d3cf02765c1c9138a340de7d19e8645f0c7691611d51"
     },
     "checkpoints/checkpoint-320/chat_template.jinja": {
       "size": 1532,
@@ -389,7 +385,7 @@
     },
     "checkpoints/checkpoint-320/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "5c58b89c38a9a89883bd5892349c9624cd88bb7db622f03f3ec205d7355f5b0f"
+      "sha256": "4b35a07c34ef08caa48bca5c42b10295e72e8b1793c7dd8a269e93a8b9766920"
     },
     "checkpoints/checkpoint-320/rng_state.pth": {
       "size": 14645,
@@ -412,8 +408,8 @@
       "sha256": "fa2e57cf6dee598825b9bba06159bc70e457d9e8315c9d1028a34b835eec8770"
     },
     "checkpoints/checkpoint-320/trainer_state.json": {
-      "size": 140424,
-      "sha256": "0585bc36ab82793ef7b34896d945d4fc476eb36e1c8c932fce9ce735565c06d5"
+      "size": 140501,
+      "sha256": "4c560090a8acf9f23947b1325483710f17afc68529ecd1eb3df8e3ff6401e1ed"
     },
     "checkpoints/checkpoint-320/training_args.bin": {
       "size": 8273,
@@ -425,11 +421,11 @@
     },
     "checkpoints/checkpoint-352/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-352/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "c0c753b89913681f39e2253cb3fd4c99db2072d6402e52c9a6eb911c9939f7ca"
+      "sha256": "f21e802c0d2fcc1d790922454cc55a3257b76beae21bbf54ea77c6151be66e6e"
     },
     "checkpoints/checkpoint-352/chat_template.jinja": {
       "size": 1532,
@@ -437,7 +433,7 @@
     },
     "checkpoints/checkpoint-352/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "da67a87dafe48008ff80730e49da5dc2be6bac75e3671c7d772f8b936a730e84"
+      "sha256": "6ff65aaa33bf6a907d3faa52032f129bb11b519ac761adc2816aec8951f53b4e"
     },
     "checkpoints/checkpoint-352/rng_state.pth": {
       "size": 14645,
@@ -460,8 +456,8 @@
       "sha256": "b9bc28c20f274eed19841a293251e42a3a606798b019918fd3fd29a96ca3043c"
     },
     "checkpoints/checkpoint-352/trainer_state.json": {
-      "size": 154550,
-      "sha256": "d0c9abdce7185cb21098e9b774e5d1ffb410bbf9eb726d5d72c241ba87cd3994"
+      "size": 154629,
+      "sha256": "dd3348f05d98c22b8b7f6bdce1fdebf5bd2f43e34e4868f4f627cb95387429aa"
     },
     "checkpoints/checkpoint-352/training_args.bin": {
       "size": 8273,
@@ -473,11 +469,11 @@
     },
     "checkpoints/checkpoint-384/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-384/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "2f7dcd0f7d14c350cd56309aaef62d28c523be064d87d9feae5141fe2890924d"
+      "sha256": "535e88052cb6afb16717578e8ca20d23dd4e86de845c837f78a8af4881d745cf"
     },
     "checkpoints/checkpoint-384/chat_template.jinja": {
       "size": 1532,
@@ -485,7 +481,7 @@
     },
     "checkpoints/checkpoint-384/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "8155a0ce1ff150d4770b9daab3658e3e6adb736967f6f914797107c25fd6c8b0"
+      "sha256": "1e2785d9c6ff17bc108c22ed9e79c9bbf116cbeae3109e3b396835a887d1ed2d"
     },
     "checkpoints/checkpoint-384/rng_state.pth": {
       "size": 14645,
@@ -508,8 +504,8 @@
       "sha256": "3075bc59309b4b7b0fefd61404619b2d7b2fe67268f884f30ea4319fd447e21e"
     },
     "checkpoints/checkpoint-384/trainer_state.json": {
-      "size": 168694,
-      "sha256": "4fac0c89f0943b720753cda5ab7ae711c8966c9ea600d107b9bab46f910a2807"
+      "size": 168780,
+      "sha256": "af01e24ff12e70b4ef31c9bf5b86c1e7bd14855cea703883da4d76fe3105c551"
     },
     "checkpoints/checkpoint-384/training_args.bin": {
       "size": 8273,
@@ -521,11 +517,11 @@
     },
     "checkpoints/checkpoint-416/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-416/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "ee180b0dcc20c9c141d258bba075090f4b87eda8f6b515908a03572f6fc4ca76"
+      "sha256": "0d560dd0eb4e8597dfc939225d96feb31798b32c08203167541c630616fa2492"
     },
     "checkpoints/checkpoint-416/chat_template.jinja": {
       "size": 1532,
@@ -533,7 +529,7 @@
     },
     "checkpoints/checkpoint-416/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "a82de99afc36ad2c4890b24e750b8c58940441e1006803dd32e087423bd37e67"
+      "sha256": "cd9783ec9609f3730f73bfcc5cfd64cc123f08f492bbf069f1c6f47c16144b5d"
     },
     "checkpoints/checkpoint-416/rng_state.pth": {
       "size": 14645,
@@ -556,8 +552,8 @@
       "sha256": "033256f14efaa1c491d2eeb0fe5a1f206222c187594393c6791fa23f37aacb5e"
     },
     "checkpoints/checkpoint-416/trainer_state.json": {
-      "size": 182841,
-      "sha256": "31f80e20dc8f66f2f3e66ab4b51cf798ca041899d6c3408c7924bd3382b2a3dd"
+      "size": 182932,
+      "sha256": "997563a587d83c14e55f23a0000c2acd92899991625a2831dedefcc6d621267d"
     },
     "checkpoints/checkpoint-416/training_args.bin": {
       "size": 8273,
@@ -569,11 +565,11 @@
     },
     "checkpoints/checkpoint-448/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-448/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "f25f15c1bc008ea9820a869ec237898029c46e71dbb17a16b7c540d0ab55950f"
+      "sha256": "3bacb4d3f901db3a71785ec9bbf0c989eb190275b59257a46ebe7cda04a4211b"
     },
     "checkpoints/checkpoint-448/chat_template.jinja": {
       "size": 1532,
@@ -581,7 +577,7 @@
     },
     "checkpoints/checkpoint-448/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "5d6cd80b908abae800035b7b84556d6b95bdc1afef25c3cd05b9d5a4d951f6c7"
+      "sha256": "33cfd5b5b7300a5e59a22e211b622aa8250b83b1e42d9aa245e82b799848bf3e"
     },
     "checkpoints/checkpoint-448/rng_state.pth": {
       "size": 14645,
@@ -604,8 +600,8 @@
       "sha256": "87ae99c7d9168aee7dd04df6332dedc5884f11b6c9ce320edc6d2b867c048531"
     },
     "checkpoints/checkpoint-448/trainer_state.json": {
-      "size": 196995,
-      "sha256": "c794bd88f8a7af8e83714d0df996bcf2e23dd3fe3e9b7e034f5484bcd606470b"
+      "size": 197087,
+      "sha256": "dd453a8c241e9adba0f2155ac5b899e149e25772266ea7f4644cd0217d343d4f"
     },
     "checkpoints/checkpoint-448/training_args.bin": {
       "size": 8273,
@@ -617,11 +613,11 @@
     },
     "checkpoints/checkpoint-480/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-480/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "8574c62e5993d868e285944d01ea5c0f530e591f2154d08a0c57b6eadfedd47d"
+      "sha256": "759e2f47a5e976dcfbe11dde9491135d96d1b05bada169151f4a8f939269acc5"
     },
     "checkpoints/checkpoint-480/chat_template.jinja": {
       "size": 1532,
@@ -629,7 +625,7 @@
     },
     "checkpoints/checkpoint-480/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "d83b575649c6ab2f7828c29ac5ae5dcb50824f042fa45adedc229558f4ef1923"
+      "sha256": "cd01ae9ec3eb43f4befc813946bc00ce45c82d3237dfa7f15ddccfe7878ea01d"
     },
     "checkpoints/checkpoint-480/rng_state.pth": {
       "size": 14645,
@@ -652,8 +648,8 @@
       "sha256": "6c547605fe97848a0748ea1ac9bef1baa4a78902cfb711aedb151a69c5994f49"
     },
     "checkpoints/checkpoint-480/trainer_state.json": {
-      "size": 211173,
-      "sha256": "8d0359199a7fd15b57527a4903abb09766e2682d61df31282ebfa2ca9cfd7e70"
+      "size": 211284,
+      "sha256": "aead414e3a072327e5746caf31d43a0b98ec71e9b30ca1897ce90790862804a6"
     },
     "checkpoints/checkpoint-480/training_args.bin": {
       "size": 8273,
@@ -665,11 +661,11 @@
     },
     "checkpoints/checkpoint-512/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-512/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "16cc65acfba87debf950e49fc05f5dce80ea16745d793c826ed96faac31a47a1"
+      "sha256": "daeafd5d686b4b39c5b3dd656eca36514ea93ef225c5afab298c74bbfab4b632"
     },
     "checkpoints/checkpoint-512/chat_template.jinja": {
       "size": 1532,
@@ -677,7 +673,7 @@
     },
     "checkpoints/checkpoint-512/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "fe1efd539e57db6375dbfc409cf1e2a434b05b88c1f3bd7e58c16a8d2075a365"
+      "sha256": "05bb512f0593d6339b6dd5bd7fd23b00fbb5ccbe33fcfaa642269827ce75bdc6"
     },
     "checkpoints/checkpoint-512/rng_state.pth": {
       "size": 14645,
@@ -700,8 +696,8 @@
       "sha256": "c0fcac61392efcd924c1610ad304635be7ae41b33c0ad3526b109f7b1f8f7fe9"
     },
     "checkpoints/checkpoint-512/trainer_state.json": {
-      "size": 225350,
-      "sha256": "2dbfcbe1a03c6951df9b24ec896edee522b63548a54faf52ac159f28161dac6a"
+      "size": 225479,
+      "sha256": "02f9a66b02bb552b9fb5b0fe2cf523809d3b2ee4475683870014d1b13ad7a8a2"
     },
     "checkpoints/checkpoint-512/training_args.bin": {
       "size": 8273,
@@ -713,11 +709,11 @@
     },
     "checkpoints/checkpoint-64/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-64/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "051830de7f1d163f4de9749f244e6981723b34f0246ab1c4576cc93cf5c97034"
+      "sha256": "e859c9377bfb05045990aad9622ecbcea4ec1f51ee262c524148710f93c608f1"
     },
     "checkpoints/checkpoint-64/chat_template.jinja": {
       "size": 1532,
@@ -725,7 +721,7 @@
     },
     "checkpoints/checkpoint-64/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "b96a1d411d21f029342790380398e4a7970a8f0e26d27c412cfd73e290cced3a"
+      "sha256": "f0bb5b0e8d378704b8006d08d74c2f903a1422f4e828a3f1e606a4b8edbfa7a3"
     },
     "checkpoints/checkpoint-64/rng_state.pth": {
       "size": 14645,
@@ -748,8 +744,8 @@
       "sha256": "9c3a6e3ed0949f0de9f9b7c1d666e376e3d3626dfe0a6da2a265ef135b18e10b"
     },
     "checkpoints/checkpoint-64/trainer_state.json": {
-      "size": 28326,
-      "sha256": "6b8def4d7ff9d58d7a52fab48646972203a27e4ea3b984ad785edc54873a7aac"
+      "size": 28330,
+      "sha256": "82a0433edb2739b866083caf137728dc3d96ccd93e5eb2fd39a66d8d6f7bd5ac"
     },
     "checkpoints/checkpoint-64/training_args.bin": {
       "size": 8273,
@@ -761,11 +757,11 @@
     },
     "checkpoints/checkpoint-96/adapter_config.json": {
       "size": 1098,
-      "sha256": "3b321f93ae273f917611377cbfbf74459bad326b4811326bdf528471050a99f2"
+      "sha256": "7c54ee9b6e682028e1421326f0a1cbbeb36cdbcc1ad907238e6d932c196e293d"
     },
     "checkpoints/checkpoint-96/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "a0e403bf65449bdaf7ebc6919f111acf4c4c5995e60d3ee7136848a492f809e9"
+      "sha256": "0e85942ea7b1d777c6e874fcfe57e08e0752f1e581a02a35bb62b7dc1942bc5a"
     },
     "checkpoints/checkpoint-96/chat_template.jinja": {
       "size": 1532,
@@ -773,7 +769,7 @@
     },
     "checkpoints/checkpoint-96/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "d4bf3da6e8fbdc8a5ebb3ed971650bce7f765b5da49d443bde267183074a0ff5"
+      "sha256": "e09d86b7ef11b8b51a819ec69e40b2c45e5f32072380fdac56e42f42d36fcb81"
     },
     "checkpoints/checkpoint-96/rng_state.pth": {
       "size": 14645,
@@ -796,8 +792,8 @@
       "sha256": "f276353ce1fa0f6362678bebc659fb55eeca16e731ebeb9a62916aa8efaa3b8b"
     },
     "checkpoints/checkpoint-96/trainer_state.json": {
-      "size": 42266,
-      "sha256": "105f8c3fef16db19a68e043f6103028151c5228515d58a359bec7b71a0197e33"
+      "size": 42261,
+      "sha256": "06d1244ce26f0b9ac73872fde6867ebfff2239569a3be1aa30c8b754d5f626c0"
     },
     "checkpoints/checkpoint-96/training_args.bin": {
       "size": 8273,
@@ -808,8 +804,8 @@
       "sha256": "7c5b66498629a75e7fe3e4219bc41040b8106735e598f0837d60656025390e1a"
     },
     "checkpoints/debug.log": {
-      "size": 269339,
-      "sha256": "c828d311a21606c89225305f9859467a52bc95f4f748bf4e67dab1c5bc244fa2"
+      "size": 267536,
+      "sha256": "5096497579321697536666777d8b80e318847cf367750b87cfd3891265c1880c"
     },
     "checkpoints/processor_config.json": {
       "size": 519,
@@ -841,19 +837,19 @@
     },
     "health/training_started.json": {
       "size": 137,
-      "sha256": "623e0673869f0f9ea285385f2a5afe654d2aacaae12874f184a8a12834c77f69"
+      "sha256": "cc75570db76b03c03df3dcaac3381a89a9e92ff2b39719f3d00ac41bb89147d1"
     },
     "run.json": {
       "size": 407,
-      "sha256": "91082e603138a13b994458847e377c12969878bae8910ad8f9c4d8481b44c9e2"
+      "sha256": "d72e7f12f02ca389af664a6e6a90f9061050132100b32c737e18c5207f5c9ef3"
     },
     "train.log": {
-      "size": 276767,
-      "sha256": "732dda99c9e3cfed30f1924e8fc601b9431274287070d601d9c61b373b8c67c7"
+      "size": 274978,
+      "sha256": "b37791314be7b268280e0005a698396839dc9b72aa3ab293e0b4ad7c345ef6fa"
     },
     "trainer_state.final.json": {
-      "size": 225350,
-      "sha256": "2dbfcbe1a03c6951df9b24ec896edee522b63548a54faf52ac159f28161dac6a"
+      "size": 225479,
+      "sha256": "02f9a66b02bb552b9fb5b0fe2cf523809d3b2ee4475683870014d1b13ad7a8a2"
     },
     "training_examples.jsonl": {
       "size": 3159083,
@@ -861,11 +857,11 @@
     },
     "training_provenance.json": {
       "size": 4090,
-      "sha256": "1950335dceebdbab7947003c065cb933c19e7a4d6f513fba0440cfe0b6a1555f"
+      "sha256": "d4d5407f6abf83d471b752bd6b1e74777b935dc03d9dce25dd4ec0c1d32d3a6a"
     },
     "training_trace.jsonl": {
-      "size": 182082,
-      "sha256": "be93695ba28788aeb6f0fd98000bb9b59fb86cba22501505a16c1bc809d19b7a"
+      "size": 182211,
+      "sha256": "3162f25b3bb9dbc2b6e99501728763c5a9acec2b98a1dd4988902fa8ee0b3a2a"
     }
   }
 }
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/TRAINED.json b/aft_wave_v2/coin_real_4x__agreement/training/TRAINED.json
index b480cf4b4704cc738550967fe05281a265cb466f..2dc03f9f0e4ded9bb518128a7ba2a048173e2551 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/TRAINED.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/TRAINED.json
@@ -8,7 +8,7 @@
   "training_rows": 8192,
   "stage": "aft_dispatch_v4_wide",
   "seed": 42,
-  "minutes": 59.59,
+  "minutes": 58.83,
   "lora": {
     "r": 32,
     "alpha": 64,
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_model.safetensors
index ea3cea0a354d5b9ee2785a0c48f7ebf1c6877a9d..39b53dcd92648efb307c35556c5bb95fa36cebce 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16cc65acfba87debf950e49fc05f5dce80ea16745d793c826ed96faac31a47a1
+oid sha256:daeafd5d686b4b39c5b3dd656eca36514ea93ef225c5afab298c74bbfab4b632
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_model.safetensors
index 1196939fdd4f259a2a753c562732fe835341c68e..9fcde1482dfcf42534c09306522136a1be6f96c7 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7202ce5ed3e0cfd2be9bb413acb17a214475b81453c800a2e3c69d4ff5d7959
+oid sha256:67174eba71aaa201dd7aff62c7fbd3596ad8b4afeba399abae18cc3a0dd00888
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/optimizer.pt
index 091e23a55968e1cce84119d146759baa1dd6d631..0ed34c15cc1a38a200f5666d2606294417ce517d 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb131e3cb1d8065ea88ab03c3129c1bdf7f01361df81ae0569e4c2b6008fa036
+oid sha256:c48bb4d7fcaca8cbcee6c332057a50925e65b15baae813555c4b32ddf641b01b
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/trainer_state.json
index 75ef897a6f4efafa438761abe2549f73e6dbcee7..4665ac0220af5f5c85217ed5a18e528a00932c57 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-128/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,870 +935,870 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_model.safetensors
index a18dd2aa9a72081895b9f7d6fb86af2528d02607..132a4643cf8eb23eae7dc0d93b5c93a5b2df43ea 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50621401dbfe149cde6d4a622a952bc693f8c1c46da9363b866907fa002871d4
+oid sha256:4f45bebc7b5c77c792dbf05c98745f04359a2b3462e61a009e38cbe69159480d
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/optimizer.pt
index 3ae0cf54c1cf4a1d398a1382554b062376096eb3..4147753107e21f300cbcda9d8972f46573dc992b 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5dafd1343477b7fcb5e94b4e4a0059e8fabb1e49d85a0acc3c63a5a789736f11
+oid sha256:5ac494318568079c3e19c022c29f18c065bb4ba6884c43ff2ef297462392b14f
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/trainer_state.json
index e0f22f278529f77f9aed72075d589b3fc4616bdd..6399400efff7f7a714c7de0615662fa18ee674e3 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-160/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,170 +2083,170 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_model.safetensors
index d349b7848ac2218c2aac91952a7ec51e23a3dc82..7c38f277b2988ecef75093b81f9cabff2849787b 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1dd775d8b35d84d9d817a4a74f08f039adb13a2320077366a52bad07c43ebf8b
+oid sha256:dd65c9138a23ee4b709df1f879e1ce2a71a67a3461f4f7ed0fae00ba0eb853d5
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/optimizer.pt
index fb550349a84748dec227e783a6ce16473e1a25a7..1ef7ab51d70b8404ebd220923a03e88fc3521d69 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d796a53e0037fe564b92a554d12c2094e42094df256517710edaa58b7d48d31
+oid sha256:0e74bd30f32259d424a0528a8fc03e88152c9448882265722a55166fb622af3a
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/trainer_state.json
index e942851c0c229619f1194aaa0f3635b5743e750d..acb7fe0eda7bd95d784e91c34f2884e62c9ec089 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-192/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,100 +2601,100 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_model.safetensors
index dd5fdd3b9ecf8b0a792569e10abae8e54885060e..96e80f4222c3c67e9b1b9ffe5a99a13a6dcd4072 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8eecfc00fed67bfc79ccbe2c4580025f3a5eb09de1478ed4d237cdce0a676c0
+oid sha256:16e7d16f507266e4de0618e13529f0f6852f7dfacc533b3d9beb61399405123d
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/optimizer.pt
index 54fc5c50e0805afb57921ff4f2314cce8b37fa5b..2aaf0ecddb13d41fc730e6683f9095486cb85d7f 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab103e3591277e8cd1e5fa934a5bc9882dafbe277490d1a7a13579f389795dd1
+oid sha256:f456a3b4b0b5cfe5bb7061718fafa69fb3393a4a4055872f85b783cf6135b255
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/trainer_state.json
index bd3135e06ad912ced53c33d801156a43506fd52a..0feb4ffee97b487d4672637f7360abb1e20f0f5c 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-224/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,324 +2825,324 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_model.safetensors
index 82e5f42ace709546e1e2ee91462b7e27e19e841d..8e2802a2181a88fbac99f3f07ff03fe88d0beb04 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c74a0c1890d16768995e589354d66ad0aaf896432a0f8708036dcf170656173
+oid sha256:de74ce97413c0093f4f76198ae64237c0da203ba9a293aa2b281e66d918ca952
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/optimizer.pt
index a288ea150c6c7b6237836fb2ed193a422ef53e79..9a689f19f6473352ea644b780c5567cff17aeeaa 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:143ebc082a4dcdfe72a9b44639352173fc591f7676a6655b84d556a3b293a9d7
+oid sha256:a2e59a33e5d93b9218821342e940feec395f412661f80e94be666601cd30c6c3
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/trainer_state.json
index 15b2d78be9198713922da94b4f1d1676093fbcc4..10a35f34b15499e380132d3d9aabc1ba225ef864 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-256/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,86 +3511,86 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_model.safetensors
index cc7bbc81182f3c492152eae7e1fc70df55cbd67b..e0222031d1128af85a57672f7c84b60bb4db7c8c 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f742ce5e3dc5e92cf1d1b5194d55616562458401c38397bb3b1f150a3613ed30
+oid sha256:da2bdb4f81462625bb226aa42eb809bea433a88b8d19660ad0cb52a8f25dfb1f
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/optimizer.pt
index 5c9933e13a1b990bb7bc4575041df8023f064da0..7731a1257683cf67e7e373e62d62c6cd8aee31fb 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c6a83d5d9751f22dbe1bd2713e7e188432fb354065c57c9d49029f1c114f05b
+oid sha256:c144b815c2b7bcbf81221112a3f40c21c4e01914d789b90a94c663d34e6977b2
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/trainer_state.json
index d2ddd8c32fea7953fe13f14378c5712068de6c99..f1779fa2c2d9302f894b5f4095ff951e84426c7a 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-288/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,534 +3511,534 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_model.safetensors
index 0899b954d9cc7852a0e6951ca59694025ffa3287..06ac4f80038ef19ef6731ea63a99bca7098061c1 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee7003d2997d35b1193b3e3463a4d9c2b6d209eee799d44f21ec131ffe7ff39a
+oid sha256:32aa7a1bdc40de6c271cdaf4ba2e825940729a8400f15293192a0ab0fd0be3c4
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/optimizer.pt
index 60cd82e6a0b7d9ccf47b7e91208fcecea879705a..d326fdf1c6a2001ff067e9fc8b55a9cc0bd886e0 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f5489845cc180566855dfb58a67b01f34eb2d8fa011a26887196570356e40f0
+oid sha256:d2c09fd208676c30f1a17a813e8f50e6c1234fdc5a8449106274744a5bc40156
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/trainer_state.json
index b523a0d7fa2d7b26c2753d19e956e5513f5784c3..8619d66cc1630228247a828c3fb9c963f4da017a 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-32/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,427 +34,427 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_model.safetensors
index f0b4099d4b89f6efe2f851679023ae3656b1a130..cd974f03d222d0f7c8c32998480b7bfe5b4ac656 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6e8332b819be957190c74414784873a8ade3712d142e98f560d80f243855b7d
+oid sha256:ecfb48f90e7121908ec1d3cf02765c1c9138a340de7d19e8645f0c7691611d51
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/optimizer.pt
index 7df0aa43fef050b943c7d646915721a67b54e228..df981cc0831118b1a8b02418aa156cbe182fbc82 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c58b89c38a9a89883bd5892349c9624cd88bb7db622f03f3ec205d7355f5b0f
+oid sha256:4b35a07c34ef08caa48bca5c42b10295e72e8b1793c7dd8a269e93a8b9766920
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/trainer_state.json
index de3de3770fc1f0d568915edf1abeb961538a1b17..a8b0ea771fac23ebf49a47d5d7f03ad97334cb74 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-320/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,436 +4057,436 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_model.safetensors
index 283e0a045f8f186fb0adc2aeeddfb53a60d98dfb..7f1df589785d8c99e02fd1a14f567ef2ab23521a 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0c753b89913681f39e2253cb3fd4c99db2072d6402e52c9a6eb911c9939f7ca
+oid sha256:f21e802c0d2fcc1d790922454cc55a3257b76beae21bbf54ea77c6151be66e6e
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/optimizer.pt
index 359e4d7f1ff9bbf95e15ca2856413ff580c22f7b..3d0be747174a54adbe2286abb9f340217468635a 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da67a87dafe48008ff80730e49da5dc2be6bac75e3671c7d772f8b936a730e84
+oid sha256:6ff65aaa33bf6a907d3faa52032f129bb11b519ac761adc2816aec8951f53b4e
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/trainer_state.json
index d7879a65f9e762b1a09cfff0402c8a227447b261..33732546ee645e48509f2a22e6a96098c89c6521 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-352/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,226 +4715,226 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_model.safetensors
index f88d0a8fae2c49e25dc49efba3fc838ee15b53e5..014aed184a25dce1b15610cdd110c8ea65fc7f5f 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f7dcd0f7d14c350cd56309aaef62d28c523be064d87d9feae5141fe2890924d
+oid sha256:535e88052cb6afb16717578e8ca20d23dd4e86de845c837f78a8af4881d745cf
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/optimizer.pt
index 93aeb321d7b80b05b8969cae64d4210f7c88628e..039ea7adf3f4cb048b24848d210dde5e24e90044 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8155a0ce1ff150d4770b9daab3658e3e6adb736967f6f914797107c25fd6c8b0
+oid sha256:1e2785d9c6ff17bc108c22ed9e79c9bbf116cbeae3109e3b396835a887d1ed2d
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/trainer_state.json
index 44425253ba29e52e87d4901a2aa1e79fd61e77cb..c2f5c88b41809802f91a2f1fbd15b1ef0b291546 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-384/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,674 +4715,674 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.370597779750824,
+      "grad_norm": 0.03195308893918991,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 0.007550997193902731,
+      "loss": 0.0002666166110429913,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00758,
+      "ppl": 1.00027,
       "step": 353,
       "tokens/total": 10684176,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 161420
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0031376827973872423,
+      "grad_norm": 0.07012991607189178,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 2.3453332687495276e-05,
+      "loss": 0.000741704716347158,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00074,
       "step": 354,
       "tokens/total": 10714352,
-      "tokens/train_per_sec_per_gpu": 34.02,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 161868
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.0021399864926934242,
+      "grad_norm": 0.0005822654929943383,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 2.5832894607447088e-05,
+      "loss": 1.0378402294008993e-05,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 355,
       "tokens/total": 10744464,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 162318
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0028148347046226263,
+      "grad_norm": 0.012435230426490307,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 3.7286932638380677e-05,
+      "loss": 7.89838086348027e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00008,
       "step": 356,
       "tokens/total": 10774976,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.54,
       "tokens/trainable": 162789
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.1642119586467743,
+      "grad_norm": 0.0045598647557199,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 0.0027845175936818123,
+      "loss": 2.48450869548833e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00279,
+      "ppl": 1.00002,
       "step": 357,
       "tokens/total": 10805600,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 163254
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.066535085439682,
+      "grad_norm": 0.3044262230396271,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.0002488417667336762,
+      "loss": 0.002298796083778143,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00025,
+      "ppl": 1.0023,
       "step": 358,
       "tokens/total": 10835920,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.62,
       "tokens/trainable": 163715
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.006617655046284199,
+      "grad_norm": 0.002458421979099512,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 7.841112528694794e-05,
+      "loss": 1.9634167983895168e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 359,
       "tokens/total": 10866048,
-      "tokens/train_per_sec_per_gpu": 34.4,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 164197
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.010409035719931126,
+      "grad_norm": 0.0024773837067186832,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010489403939573094,
+      "loss": 2.744927041931078e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 360,
       "tokens/total": 10896288,
-      "tokens/train_per_sec_per_gpu": 31.6,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 164661
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.002426267135888338,
+      "grad_norm": 0.00026717092259787023,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 4.0343060391023755e-05,
+      "loss": 6.422977094189264e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 361,
       "tokens/total": 10926816,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 165111
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.015666797757148743,
+      "grad_norm": 0.0008353493758477271,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 0.00014991794887464494,
+      "loss": 1.3549893083109055e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 362,
       "tokens/total": 10954928,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 165552
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.2695651650428772,
+      "grad_norm": 0.006071125157177448,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0014983330620452762,
+      "loss": 6.428411870729178e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.0015,
+      "ppl": 1.00006,
       "step": 363,
       "tokens/total": 10985328,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 166026
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.03586212173104286,
+      "grad_norm": 0.25424960255622864,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.00030509717180393636,
+      "loss": 0.004243595991283655,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00031,
+      "ppl": 1.00425,
       "step": 364,
       "tokens/total": 11015664,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 166486
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.0019308789633214474,
+      "grad_norm": 0.0005575277027674019,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 2.4129443772835657e-05,
+      "loss": 8.403902938880492e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 365,
       "tokens/total": 11046256,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 166959
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005940837785601616,
+      "grad_norm": 0.0014335239538922906,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 2.8238933737156913e-05,
+      "loss": 1.4738036043127067e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 366,
       "tokens/total": 11076288,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 167394
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.002125627128407359,
+      "grad_norm": 0.0010152696631848812,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 1.602789416210726e-05,
+      "loss": 1.3278609912958927e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 367,
       "tokens/total": 11106416,
-      "tokens/train_per_sec_per_gpu": 28.5,
+      "tokens/train_per_sec_per_gpu": 28.58,
       "tokens/trainable": 167832
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.2510211765766144,
+      "grad_norm": 0.11941836029291153,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 0.00010395953722763807,
+      "loss": 0.0006476733833551407,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00065,
       "step": 368,
       "tokens/total": 11136848,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 31.02,
       "tokens/trainable": 168242
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.001904280623421073,
+      "grad_norm": 0.006877629552036524,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 2.6418363631819375e-05,
+      "loss": 7.449048280250281e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 369,
       "tokens/total": 11167376,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 168715
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.0028152521699666977,
+      "grad_norm": 0.0007151139434427023,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 2.9974533390486613e-05,
+      "loss": 1.1437590728746727e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 370,
       "tokens/total": 11197776,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.86,
       "tokens/trainable": 169147
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.0010822078911587596,
+      "grad_norm": 0.009094453416764736,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 1.34217716549756e-05,
+      "loss": 5.9242345741949975e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11227840,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 169628
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.005367911420762539,
+      "grad_norm": 0.000247267191298306,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 6.65646803099662e-05,
+      "loss": 7.548428584414069e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 372,
       "tokens/total": 11258176,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 170085
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.047570567578077316,
+      "grad_norm": 0.0003029497747775167,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.0003471036907285452,
+      "loss": 5.4013939916330855e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00035,
+      "ppl": 1.00001,
       "step": 373,
       "tokens/total": 11288336,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 170584
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.040886107832193375,
+      "grad_norm": 0.001942179980687797,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.0002880148240365088,
+      "loss": 2.380511250521522e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00029,
+      "ppl": 1.00002,
       "step": 374,
       "tokens/total": 11318848,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 171058
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.0031216361094266176,
+      "grad_norm": 0.0019308892078697681,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 2.690855944820214e-05,
+      "loss": 1.5233906196954194e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 375,
       "tokens/total": 11349168,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 171518
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.012176645919680595,
+      "grad_norm": 0.6044301986694336,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 8.402287494391203e-05,
+      "loss": 0.006350134499371052,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00008,
+      "ppl": 1.00637,
       "step": 376,
       "tokens/total": 11379328,
-      "tokens/train_per_sec_per_gpu": 40.48,
+      "tokens/train_per_sec_per_gpu": 40.62,
       "tokens/trainable": 172007
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.24136756360530853,
+      "grad_norm": 0.25446587800979614,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 0.0013303000014275312,
+      "loss": 0.0018724403344094753,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00133,
+      "ppl": 1.00187,
       "step": 377,
       "tokens/total": 11409568,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 172468
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.021714258939027786,
+      "grad_norm": 0.0019207323202863336,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 0.00020160498388577253,
+      "loss": 3.049923907383345e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 378,
       "tokens/total": 11439968,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 172923
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.35412073135375977,
+      "grad_norm": 0.33491548895835876,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 0.015761105343699455,
+      "loss": 0.02673855796456337,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01589,
+      "ppl": 1.0271,
       "step": 379,
       "tokens/total": 11470496,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 173370
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005766173824667931,
+      "grad_norm": 0.006177723873406649,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 2.5790239305933937e-05,
+      "loss": 5.195035191718489e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 380,
       "tokens/total": 11500720,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 173837
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.005898744333535433,
+      "grad_norm": 0.06973031163215637,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 5.821501690661535e-05,
+      "loss": 0.0006310560274869204,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00006,
+      "ppl": 1.00063,
       "step": 381,
       "tokens/total": 11531280,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 174294
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005451703444123268,
+      "grad_norm": 0.0039021887350827456,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 4.0181505028158426e-05,
+      "loss": 4.849781544180587e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 382,
       "tokens/total": 11561264,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.84,
       "tokens/trainable": 174701
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0038302047178149223,
+      "grad_norm": 0.0007975143962539732,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 3.403786467970349e-05,
+      "loss": 2.3489263185183518e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 383,
       "tokens/total": 11591616,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 175131
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.004227485507726669,
+      "grad_norm": 0.012541128322482109,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 3.562243364285678e-05,
+      "loss": 0.00011495217040646821,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00011,
       "step": 384,
       "tokens/total": 11622000,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 175586
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_model.safetensors
index 1456a02749858800eb116defcb5ca35f531da046..3c266dc766268e163d19266d3054c932fed78a74 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee180b0dcc20c9c141d258bba075090f4b87eda8f6b515908a03572f6fc4ca76
+oid sha256:0d560dd0eb4e8597dfc939225d96feb31798b32c08203167541c630616fa2492
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/optimizer.pt
index 950115dc3d2f093c1d3c5fe496c4b4b22a8e65cc..9e91de2a2949e863d20d36465bd9c2af57c81513 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a82de99afc36ad2c4890b24e750b8c58940441e1006803dd32e087423bd37e67
+oid sha256:cd9783ec9609f3730f73bfcc5cfd64cc123f08f492bbf069f1c6f47c16144b5d
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/trainer_state.json
index 214d6bf6da324da9008c9edc7ff77b3dba9213ae..81d723027848e0ac359b04236d221e710d21e369 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-416/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,1105 +4715,1105 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.370597779750824,
+      "grad_norm": 0.03195308893918991,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 0.007550997193902731,
+      "loss": 0.0002666166110429913,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00758,
+      "ppl": 1.00027,
       "step": 353,
       "tokens/total": 10684176,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 161420
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0031376827973872423,
+      "grad_norm": 0.07012991607189178,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 2.3453332687495276e-05,
+      "loss": 0.000741704716347158,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00074,
       "step": 354,
       "tokens/total": 10714352,
-      "tokens/train_per_sec_per_gpu": 34.02,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 161868
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.0021399864926934242,
+      "grad_norm": 0.0005822654929943383,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 2.5832894607447088e-05,
+      "loss": 1.0378402294008993e-05,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 355,
       "tokens/total": 10744464,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 162318
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0028148347046226263,
+      "grad_norm": 0.012435230426490307,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 3.7286932638380677e-05,
+      "loss": 7.89838086348027e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00008,
       "step": 356,
       "tokens/total": 10774976,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.54,
       "tokens/trainable": 162789
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.1642119586467743,
+      "grad_norm": 0.0045598647557199,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 0.0027845175936818123,
+      "loss": 2.48450869548833e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00279,
+      "ppl": 1.00002,
       "step": 357,
       "tokens/total": 10805600,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 163254
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.066535085439682,
+      "grad_norm": 0.3044262230396271,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.0002488417667336762,
+      "loss": 0.002298796083778143,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00025,
+      "ppl": 1.0023,
       "step": 358,
       "tokens/total": 10835920,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.62,
       "tokens/trainable": 163715
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.006617655046284199,
+      "grad_norm": 0.002458421979099512,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 7.841112528694794e-05,
+      "loss": 1.9634167983895168e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 359,
       "tokens/total": 10866048,
-      "tokens/train_per_sec_per_gpu": 34.4,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 164197
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.010409035719931126,
+      "grad_norm": 0.0024773837067186832,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010489403939573094,
+      "loss": 2.744927041931078e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 360,
       "tokens/total": 10896288,
-      "tokens/train_per_sec_per_gpu": 31.6,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 164661
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.002426267135888338,
+      "grad_norm": 0.00026717092259787023,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 4.0343060391023755e-05,
+      "loss": 6.422977094189264e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 361,
       "tokens/total": 10926816,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 165111
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.015666797757148743,
+      "grad_norm": 0.0008353493758477271,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 0.00014991794887464494,
+      "loss": 1.3549893083109055e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 362,
       "tokens/total": 10954928,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 165552
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.2695651650428772,
+      "grad_norm": 0.006071125157177448,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0014983330620452762,
+      "loss": 6.428411870729178e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.0015,
+      "ppl": 1.00006,
       "step": 363,
       "tokens/total": 10985328,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 166026
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.03586212173104286,
+      "grad_norm": 0.25424960255622864,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.00030509717180393636,
+      "loss": 0.004243595991283655,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00031,
+      "ppl": 1.00425,
       "step": 364,
       "tokens/total": 11015664,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 166486
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.0019308789633214474,
+      "grad_norm": 0.0005575277027674019,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 2.4129443772835657e-05,
+      "loss": 8.403902938880492e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 365,
       "tokens/total": 11046256,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 166959
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005940837785601616,
+      "grad_norm": 0.0014335239538922906,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 2.8238933737156913e-05,
+      "loss": 1.4738036043127067e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 366,
       "tokens/total": 11076288,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 167394
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.002125627128407359,
+      "grad_norm": 0.0010152696631848812,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 1.602789416210726e-05,
+      "loss": 1.3278609912958927e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 367,
       "tokens/total": 11106416,
-      "tokens/train_per_sec_per_gpu": 28.5,
+      "tokens/train_per_sec_per_gpu": 28.58,
       "tokens/trainable": 167832
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.2510211765766144,
+      "grad_norm": 0.11941836029291153,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 0.00010395953722763807,
+      "loss": 0.0006476733833551407,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00065,
       "step": 368,
       "tokens/total": 11136848,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 31.02,
       "tokens/trainable": 168242
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.001904280623421073,
+      "grad_norm": 0.006877629552036524,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 2.6418363631819375e-05,
+      "loss": 7.449048280250281e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 369,
       "tokens/total": 11167376,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 168715
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.0028152521699666977,
+      "grad_norm": 0.0007151139434427023,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 2.9974533390486613e-05,
+      "loss": 1.1437590728746727e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 370,
       "tokens/total": 11197776,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.86,
       "tokens/trainable": 169147
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.0010822078911587596,
+      "grad_norm": 0.009094453416764736,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 1.34217716549756e-05,
+      "loss": 5.9242345741949975e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11227840,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 169628
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.005367911420762539,
+      "grad_norm": 0.000247267191298306,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 6.65646803099662e-05,
+      "loss": 7.548428584414069e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 372,
       "tokens/total": 11258176,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 170085
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.047570567578077316,
+      "grad_norm": 0.0003029497747775167,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.0003471036907285452,
+      "loss": 5.4013939916330855e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00035,
+      "ppl": 1.00001,
       "step": 373,
       "tokens/total": 11288336,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 170584
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.040886107832193375,
+      "grad_norm": 0.001942179980687797,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.0002880148240365088,
+      "loss": 2.380511250521522e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00029,
+      "ppl": 1.00002,
       "step": 374,
       "tokens/total": 11318848,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 171058
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.0031216361094266176,
+      "grad_norm": 0.0019308892078697681,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 2.690855944820214e-05,
+      "loss": 1.5233906196954194e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 375,
       "tokens/total": 11349168,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 171518
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.012176645919680595,
+      "grad_norm": 0.6044301986694336,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 8.402287494391203e-05,
+      "loss": 0.006350134499371052,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00008,
+      "ppl": 1.00637,
       "step": 376,
       "tokens/total": 11379328,
-      "tokens/train_per_sec_per_gpu": 40.48,
+      "tokens/train_per_sec_per_gpu": 40.62,
       "tokens/trainable": 172007
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.24136756360530853,
+      "grad_norm": 0.25446587800979614,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 0.0013303000014275312,
+      "loss": 0.0018724403344094753,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00133,
+      "ppl": 1.00187,
       "step": 377,
       "tokens/total": 11409568,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 172468
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.021714258939027786,
+      "grad_norm": 0.0019207323202863336,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 0.00020160498388577253,
+      "loss": 3.049923907383345e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 378,
       "tokens/total": 11439968,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 172923
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.35412073135375977,
+      "grad_norm": 0.33491548895835876,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 0.015761105343699455,
+      "loss": 0.02673855796456337,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01589,
+      "ppl": 1.0271,
       "step": 379,
       "tokens/total": 11470496,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 173370
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005766173824667931,
+      "grad_norm": 0.006177723873406649,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 2.5790239305933937e-05,
+      "loss": 5.195035191718489e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 380,
       "tokens/total": 11500720,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 173837
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.005898744333535433,
+      "grad_norm": 0.06973031163215637,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 5.821501690661535e-05,
+      "loss": 0.0006310560274869204,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00006,
+      "ppl": 1.00063,
       "step": 381,
       "tokens/total": 11531280,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 174294
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005451703444123268,
+      "grad_norm": 0.0039021887350827456,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 4.0181505028158426e-05,
+      "loss": 4.849781544180587e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 382,
       "tokens/total": 11561264,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.84,
       "tokens/trainable": 174701
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0038302047178149223,
+      "grad_norm": 0.0007975143962539732,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 3.403786467970349e-05,
+      "loss": 2.3489263185183518e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 383,
       "tokens/total": 11591616,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 175131
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.004227485507726669,
+      "grad_norm": 0.012541128322482109,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 3.562243364285678e-05,
+      "loss": 0.00011495217040646821,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00011,
       "step": 384,
       "tokens/total": 11622000,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 175586
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.013361346907913685,
+      "grad_norm": 0.004752015229314566,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0001464220113120973,
+      "loss": 6.502695032395422e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00007,
       "step": 385,
       "tokens/total": 11652352,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176026
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.008831475861370564,
+      "grad_norm": 0.014192809350788593,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 8.2662510976661e-05,
+      "loss": 8.075163350440562e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00008,
       "step": 386,
       "tokens/total": 11682736,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.91,
       "tokens/trainable": 176512
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.009311008267104626,
+      "grad_norm": 0.007288333959877491,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 0.0001226613821927458,
+      "loss": 8.845872798701748e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00009,
       "step": 387,
       "tokens/total": 11713136,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.45,
       "tokens/trainable": 176943
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.0017816838808357716,
+      "grad_norm": 0.10162956267595291,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 2.3221660740091465e-05,
+      "loss": 0.0007253064541146159,
       "memory/device_reserved (GiB)": 35.23,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00073,
       "step": 388,
       "tokens/total": 11743808,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 177370
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.023602057248353958,
+      "grad_norm": 0.005729300435632467,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 0.00012638044427148998,
+      "loss": 8.652975520817563e-05,
       "memory/device_reserved (GiB)": 35.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00013,
+      "ppl": 1.00009,
       "step": 389,
       "tokens/total": 11774144,
-      "tokens/train_per_sec_per_gpu": 35.14,
+      "tokens/train_per_sec_per_gpu": 35.2,
       "tokens/trainable": 177851
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.04383797571063042,
+      "grad_norm": 0.0715414360165596,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 0.0004640861588995904,
+      "loss": 0.0007950748549774289,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00046,
+      "ppl": 1.0008,
       "step": 390,
       "tokens/total": 11804272,
-      "tokens/train_per_sec_per_gpu": 34.68,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 178310
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.061932601034641266,
+      "grad_norm": 0.009775097481906414,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.00033243370126001537,
+      "loss": 0.00010615254723234102,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00033,
+      "ppl": 1.00011,
       "step": 391,
       "tokens/total": 11832432,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 178736
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.04983547702431679,
+      "grad_norm": 0.0015552763361483812,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 0.000599355495069176,
+      "loss": 2.954876617877744e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0006,
+      "ppl": 1.00003,
       "step": 392,
       "tokens/total": 11862960,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 179233
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 0.012036106549203396,
+      "grad_norm": 0.026106838136911392,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0001237639953615144,
+      "loss": 0.00021109850786160678,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00021,
       "step": 393,
       "tokens/total": 11893232,
-      "tokens/train_per_sec_per_gpu": 38.49,
+      "tokens/train_per_sec_per_gpu": 38.57,
       "tokens/trainable": 179730
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.008992817252874374,
+      "grad_norm": 0.020347680896520615,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 0.00010385089262854308,
+      "loss": 0.0002601295418571681,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0001,
+      "ppl": 1.00026,
       "step": 394,
       "tokens/total": 11923664,
-      "tokens/train_per_sec_per_gpu": 37.81,
+      "tokens/train_per_sec_per_gpu": 37.87,
       "tokens/trainable": 180227
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.24631057679653168,
+      "grad_norm": 0.005419979803264141,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.0031685903668403625,
+      "loss": 4.929217175231315e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00317,
+      "ppl": 1.00005,
       "step": 395,
       "tokens/total": 11954032,
-      "tokens/train_per_sec_per_gpu": 32.85,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 180687
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.06871633976697922,
+      "grad_norm": 0.0031002764590084553,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 0.0005168755305930972,
+      "loss": 6.144218787085265e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00052,
+      "ppl": 1.00006,
       "step": 396,
       "tokens/total": 11984512,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 181141
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.011674679815769196,
+      "grad_norm": 0.006748533807694912,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 0.00017496946384198964,
+      "loss": 9.171784040518105e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 397,
       "tokens/total": 12014880,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 36.02,
       "tokens/trainable": 181648
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.14610664546489716,
+      "grad_norm": 0.08927815407514572,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.0014660547021776438,
+      "loss": 0.001000746968202293,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00147,
+      "ppl": 1.001,
       "step": 398,
       "tokens/total": 12044928,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 182109
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.016359565779566765,
+      "grad_norm": 0.004651801194995642,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 7.928608101792634e-05,
+      "loss": 5.3439554903889075e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00008,
+      "ppl": 1.00005,
       "step": 399,
       "tokens/total": 12075344,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 182559
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0010484450031071901,
+      "grad_norm": 0.0017514342907816172,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 1.526270352769643e-05,
+      "loss": 4.570486271404661e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00005,
       "step": 400,
       "tokens/total": 12105616,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 182985
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.0005900752730667591,
+      "grad_norm": 0.01574699766933918,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 1.2166316082584672e-05,
+      "loss": 0.0001242965809069574,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 401,
       "tokens/total": 12136080,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 183440
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.013194791041314602,
+      "grad_norm": 0.007152364123612642,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 0.0001970212033484131,
+      "loss": 0.0001222842838615179,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0002,
+      "ppl": 1.00012,
       "step": 402,
       "tokens/total": 12166240,
-      "tokens/train_per_sec_per_gpu": 34.52,
+      "tokens/train_per_sec_per_gpu": 34.51,
       "tokens/trainable": 183917
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.02566305734217167,
+      "grad_norm": 0.0015342400874942541,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.00022389904188457876,
+      "loss": 3.621872019721195e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00022,
+      "ppl": 1.00004,
       "step": 403,
       "tokens/total": 12196544,
-      "tokens/train_per_sec_per_gpu": 30.39,
+      "tokens/train_per_sec_per_gpu": 30.36,
       "tokens/trainable": 184348
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.008275284431874752,
+      "grad_norm": 0.0035214691888540983,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 8.35009923321195e-05,
+      "loss": 6.367819150909781e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00006,
       "step": 404,
       "tokens/total": 12226784,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 184786
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.038177311420440674,
+      "grad_norm": 0.0007211520569398999,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 0.00018527230713516474,
+      "loss": 2.0325338482507505e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00019,
+      "ppl": 1.00002,
       "step": 405,
       "tokens/total": 12257264,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 185249
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.007672510575503111,
+      "grad_norm": 0.0014804014936089516,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 0.00011439670925028622,
+      "loss": 3.379736153874546e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00003,
       "step": 406,
       "tokens/total": 12287584,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 185680
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.003159885760396719,
+      "grad_norm": 0.004397980403155088,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 4.649449692806229e-05,
+      "loss": 5.1796458137687296e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00005,
       "step": 407,
       "tokens/total": 12317984,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 186100
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.32744285464286804,
+      "grad_norm": 0.0026473007164895535,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 0.002739987801760435,
+      "loss": 3.441090666456148e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00274,
+      "ppl": 1.00003,
       "step": 408,
       "tokens/total": 12348144,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.84,
       "tokens/trainable": 186565
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.0006675662589259446,
+      "grad_norm": 0.0289909727871418,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 1.2970660463906825e-05,
+      "loss": 0.00021602815832011402,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00022,
       "step": 409,
       "tokens/total": 12378544,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 187017
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.003976705949753523,
+      "grad_norm": 0.007092812564224005,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 3.857718547806144e-05,
+      "loss": 9.646185935707763e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.0001,
       "step": 410,
       "tokens/total": 12408784,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 187468
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.0024577646981924772,
+      "grad_norm": 0.02710823155939579,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 2.431379834888503e-05,
+      "loss": 0.0003458422143012285,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00035,
       "step": 411,
       "tokens/total": 12438832,
-      "tokens/train_per_sec_per_gpu": 37.61,
+      "tokens/train_per_sec_per_gpu": 37.57,
       "tokens/trainable": 187961
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003754909848794341,
+      "grad_norm": 0.29485711455345154,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 4.119630830246024e-05,
+      "loss": 0.0032923028338700533,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.0033,
       "step": 412,
       "tokens/total": 12469296,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 188447
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.3583323657512665,
+      "grad_norm": 0.22881586849689484,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.004735157359391451,
+      "loss": 0.0035809341352432966,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00475,
+      "ppl": 1.00359,
       "step": 413,
       "tokens/total": 12499392,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 188903
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.049485232681035995,
+      "grad_norm": 0.0026347371749579906,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 0.00019404500199016184,
+      "loss": 6.0475373174995184e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.00006,
       "step": 414,
       "tokens/total": 12529952,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 189398
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0030447980388998985,
+      "grad_norm": 0.0017223359318450093,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 2.0038012735312805e-05,
+      "loss": 4.137849100516178e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 415,
       "tokens/total": 12560240,
       "tokens/train_per_sec_per_gpu": 34.16,
@@ -5821,16 +5821,16 @@
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.25415608286857605,
+      "grad_norm": 0.011837545782327652,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 0.0016201161779463291,
+      "loss": 0.00011106643069069833,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00162,
+      "ppl": 1.00011,
       "step": 416,
       "tokens/total": 12590736,
-      "tokens/train_per_sec_per_gpu": 32.64,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 190306
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_model.safetensors
index 902aa5c1e90c93f7b38732ae8340a6a9d842849f..9d34729a9d5937ebe54d58f659473585e20e1fa8 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f25f15c1bc008ea9820a869ec237898029c46e71dbb17a16b7c540d0ab55950f
+oid sha256:3bacb4d3f901db3a71785ec9bbf0c989eb190275b59257a46ebe7cda04a4211b
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/optimizer.pt
index 5001b0d8c909aa2b5d5f2413fa091164318cd167..497d794ad140949d944c8eb319b826137c1d62eb 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d6cd80b908abae800035b7b84556d6b95bdc1afef25c3cd05b9d5a4d951f6c7
+oid sha256:33cfd5b5b7300a5e59a22e211b622aa8250b83b1e42d9aa245e82b799848bf3e
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/trainer_state.json
index f42d5a4f4e04184412fda09cfbf1d28fd1afa3d6..941befa302375529037b0bba27f07f2004fe3690 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-448/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,1105 +4715,1105 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.370597779750824,
+      "grad_norm": 0.03195308893918991,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 0.007550997193902731,
+      "loss": 0.0002666166110429913,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00758,
+      "ppl": 1.00027,
       "step": 353,
       "tokens/total": 10684176,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 161420
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0031376827973872423,
+      "grad_norm": 0.07012991607189178,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 2.3453332687495276e-05,
+      "loss": 0.000741704716347158,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00074,
       "step": 354,
       "tokens/total": 10714352,
-      "tokens/train_per_sec_per_gpu": 34.02,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 161868
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.0021399864926934242,
+      "grad_norm": 0.0005822654929943383,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 2.5832894607447088e-05,
+      "loss": 1.0378402294008993e-05,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 355,
       "tokens/total": 10744464,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 162318
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0028148347046226263,
+      "grad_norm": 0.012435230426490307,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 3.7286932638380677e-05,
+      "loss": 7.89838086348027e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00008,
       "step": 356,
       "tokens/total": 10774976,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.54,
       "tokens/trainable": 162789
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.1642119586467743,
+      "grad_norm": 0.0045598647557199,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 0.0027845175936818123,
+      "loss": 2.48450869548833e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00279,
+      "ppl": 1.00002,
       "step": 357,
       "tokens/total": 10805600,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 163254
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.066535085439682,
+      "grad_norm": 0.3044262230396271,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.0002488417667336762,
+      "loss": 0.002298796083778143,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00025,
+      "ppl": 1.0023,
       "step": 358,
       "tokens/total": 10835920,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.62,
       "tokens/trainable": 163715
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.006617655046284199,
+      "grad_norm": 0.002458421979099512,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 7.841112528694794e-05,
+      "loss": 1.9634167983895168e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 359,
       "tokens/total": 10866048,
-      "tokens/train_per_sec_per_gpu": 34.4,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 164197
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.010409035719931126,
+      "grad_norm": 0.0024773837067186832,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010489403939573094,
+      "loss": 2.744927041931078e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 360,
       "tokens/total": 10896288,
-      "tokens/train_per_sec_per_gpu": 31.6,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 164661
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.002426267135888338,
+      "grad_norm": 0.00026717092259787023,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 4.0343060391023755e-05,
+      "loss": 6.422977094189264e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 361,
       "tokens/total": 10926816,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 165111
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.015666797757148743,
+      "grad_norm": 0.0008353493758477271,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 0.00014991794887464494,
+      "loss": 1.3549893083109055e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 362,
       "tokens/total": 10954928,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 165552
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.2695651650428772,
+      "grad_norm": 0.006071125157177448,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0014983330620452762,
+      "loss": 6.428411870729178e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.0015,
+      "ppl": 1.00006,
       "step": 363,
       "tokens/total": 10985328,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 166026
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.03586212173104286,
+      "grad_norm": 0.25424960255622864,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.00030509717180393636,
+      "loss": 0.004243595991283655,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00031,
+      "ppl": 1.00425,
       "step": 364,
       "tokens/total": 11015664,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 166486
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.0019308789633214474,
+      "grad_norm": 0.0005575277027674019,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 2.4129443772835657e-05,
+      "loss": 8.403902938880492e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 365,
       "tokens/total": 11046256,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 166959
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005940837785601616,
+      "grad_norm": 0.0014335239538922906,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 2.8238933737156913e-05,
+      "loss": 1.4738036043127067e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 366,
       "tokens/total": 11076288,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 167394
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.002125627128407359,
+      "grad_norm": 0.0010152696631848812,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 1.602789416210726e-05,
+      "loss": 1.3278609912958927e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 367,
       "tokens/total": 11106416,
-      "tokens/train_per_sec_per_gpu": 28.5,
+      "tokens/train_per_sec_per_gpu": 28.58,
       "tokens/trainable": 167832
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.2510211765766144,
+      "grad_norm": 0.11941836029291153,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 0.00010395953722763807,
+      "loss": 0.0006476733833551407,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00065,
       "step": 368,
       "tokens/total": 11136848,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 31.02,
       "tokens/trainable": 168242
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.001904280623421073,
+      "grad_norm": 0.006877629552036524,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 2.6418363631819375e-05,
+      "loss": 7.449048280250281e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 369,
       "tokens/total": 11167376,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 168715
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.0028152521699666977,
+      "grad_norm": 0.0007151139434427023,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 2.9974533390486613e-05,
+      "loss": 1.1437590728746727e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 370,
       "tokens/total": 11197776,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.86,
       "tokens/trainable": 169147
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.0010822078911587596,
+      "grad_norm": 0.009094453416764736,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 1.34217716549756e-05,
+      "loss": 5.9242345741949975e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11227840,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 169628
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.005367911420762539,
+      "grad_norm": 0.000247267191298306,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 6.65646803099662e-05,
+      "loss": 7.548428584414069e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 372,
       "tokens/total": 11258176,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 170085
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.047570567578077316,
+      "grad_norm": 0.0003029497747775167,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.0003471036907285452,
+      "loss": 5.4013939916330855e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00035,
+      "ppl": 1.00001,
       "step": 373,
       "tokens/total": 11288336,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 170584
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.040886107832193375,
+      "grad_norm": 0.001942179980687797,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.0002880148240365088,
+      "loss": 2.380511250521522e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00029,
+      "ppl": 1.00002,
       "step": 374,
       "tokens/total": 11318848,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 171058
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.0031216361094266176,
+      "grad_norm": 0.0019308892078697681,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 2.690855944820214e-05,
+      "loss": 1.5233906196954194e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 375,
       "tokens/total": 11349168,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 171518
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.012176645919680595,
+      "grad_norm": 0.6044301986694336,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 8.402287494391203e-05,
+      "loss": 0.006350134499371052,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00008,
+      "ppl": 1.00637,
       "step": 376,
       "tokens/total": 11379328,
-      "tokens/train_per_sec_per_gpu": 40.48,
+      "tokens/train_per_sec_per_gpu": 40.62,
       "tokens/trainable": 172007
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.24136756360530853,
+      "grad_norm": 0.25446587800979614,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 0.0013303000014275312,
+      "loss": 0.0018724403344094753,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00133,
+      "ppl": 1.00187,
       "step": 377,
       "tokens/total": 11409568,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 172468
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.021714258939027786,
+      "grad_norm": 0.0019207323202863336,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 0.00020160498388577253,
+      "loss": 3.049923907383345e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 378,
       "tokens/total": 11439968,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 172923
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.35412073135375977,
+      "grad_norm": 0.33491548895835876,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 0.015761105343699455,
+      "loss": 0.02673855796456337,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01589,
+      "ppl": 1.0271,
       "step": 379,
       "tokens/total": 11470496,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 173370
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005766173824667931,
+      "grad_norm": 0.006177723873406649,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 2.5790239305933937e-05,
+      "loss": 5.195035191718489e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 380,
       "tokens/total": 11500720,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 173837
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.005898744333535433,
+      "grad_norm": 0.06973031163215637,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 5.821501690661535e-05,
+      "loss": 0.0006310560274869204,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00006,
+      "ppl": 1.00063,
       "step": 381,
       "tokens/total": 11531280,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 174294
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005451703444123268,
+      "grad_norm": 0.0039021887350827456,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 4.0181505028158426e-05,
+      "loss": 4.849781544180587e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 382,
       "tokens/total": 11561264,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.84,
       "tokens/trainable": 174701
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0038302047178149223,
+      "grad_norm": 0.0007975143962539732,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 3.403786467970349e-05,
+      "loss": 2.3489263185183518e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 383,
       "tokens/total": 11591616,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 175131
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.004227485507726669,
+      "grad_norm": 0.012541128322482109,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 3.562243364285678e-05,
+      "loss": 0.00011495217040646821,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00011,
       "step": 384,
       "tokens/total": 11622000,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 175586
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.013361346907913685,
+      "grad_norm": 0.004752015229314566,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0001464220113120973,
+      "loss": 6.502695032395422e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00007,
       "step": 385,
       "tokens/total": 11652352,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176026
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.008831475861370564,
+      "grad_norm": 0.014192809350788593,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 8.2662510976661e-05,
+      "loss": 8.075163350440562e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00008,
       "step": 386,
       "tokens/total": 11682736,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.91,
       "tokens/trainable": 176512
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.009311008267104626,
+      "grad_norm": 0.007288333959877491,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 0.0001226613821927458,
+      "loss": 8.845872798701748e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00009,
       "step": 387,
       "tokens/total": 11713136,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.45,
       "tokens/trainable": 176943
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.0017816838808357716,
+      "grad_norm": 0.10162956267595291,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 2.3221660740091465e-05,
+      "loss": 0.0007253064541146159,
       "memory/device_reserved (GiB)": 35.23,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00073,
       "step": 388,
       "tokens/total": 11743808,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 177370
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.023602057248353958,
+      "grad_norm": 0.005729300435632467,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 0.00012638044427148998,
+      "loss": 8.652975520817563e-05,
       "memory/device_reserved (GiB)": 35.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00013,
+      "ppl": 1.00009,
       "step": 389,
       "tokens/total": 11774144,
-      "tokens/train_per_sec_per_gpu": 35.14,
+      "tokens/train_per_sec_per_gpu": 35.2,
       "tokens/trainable": 177851
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.04383797571063042,
+      "grad_norm": 0.0715414360165596,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 0.0004640861588995904,
+      "loss": 0.0007950748549774289,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00046,
+      "ppl": 1.0008,
       "step": 390,
       "tokens/total": 11804272,
-      "tokens/train_per_sec_per_gpu": 34.68,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 178310
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.061932601034641266,
+      "grad_norm": 0.009775097481906414,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.00033243370126001537,
+      "loss": 0.00010615254723234102,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00033,
+      "ppl": 1.00011,
       "step": 391,
       "tokens/total": 11832432,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 178736
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.04983547702431679,
+      "grad_norm": 0.0015552763361483812,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 0.000599355495069176,
+      "loss": 2.954876617877744e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0006,
+      "ppl": 1.00003,
       "step": 392,
       "tokens/total": 11862960,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 179233
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 0.012036106549203396,
+      "grad_norm": 0.026106838136911392,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0001237639953615144,
+      "loss": 0.00021109850786160678,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00021,
       "step": 393,
       "tokens/total": 11893232,
-      "tokens/train_per_sec_per_gpu": 38.49,
+      "tokens/train_per_sec_per_gpu": 38.57,
       "tokens/trainable": 179730
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.008992817252874374,
+      "grad_norm": 0.020347680896520615,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 0.00010385089262854308,
+      "loss": 0.0002601295418571681,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0001,
+      "ppl": 1.00026,
       "step": 394,
       "tokens/total": 11923664,
-      "tokens/train_per_sec_per_gpu": 37.81,
+      "tokens/train_per_sec_per_gpu": 37.87,
       "tokens/trainable": 180227
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.24631057679653168,
+      "grad_norm": 0.005419979803264141,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.0031685903668403625,
+      "loss": 4.929217175231315e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00317,
+      "ppl": 1.00005,
       "step": 395,
       "tokens/total": 11954032,
-      "tokens/train_per_sec_per_gpu": 32.85,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 180687
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.06871633976697922,
+      "grad_norm": 0.0031002764590084553,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 0.0005168755305930972,
+      "loss": 6.144218787085265e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00052,
+      "ppl": 1.00006,
       "step": 396,
       "tokens/total": 11984512,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 181141
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.011674679815769196,
+      "grad_norm": 0.006748533807694912,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 0.00017496946384198964,
+      "loss": 9.171784040518105e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 397,
       "tokens/total": 12014880,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 36.02,
       "tokens/trainable": 181648
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.14610664546489716,
+      "grad_norm": 0.08927815407514572,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.0014660547021776438,
+      "loss": 0.001000746968202293,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00147,
+      "ppl": 1.001,
       "step": 398,
       "tokens/total": 12044928,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 182109
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.016359565779566765,
+      "grad_norm": 0.004651801194995642,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 7.928608101792634e-05,
+      "loss": 5.3439554903889075e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00008,
+      "ppl": 1.00005,
       "step": 399,
       "tokens/total": 12075344,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 182559
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0010484450031071901,
+      "grad_norm": 0.0017514342907816172,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 1.526270352769643e-05,
+      "loss": 4.570486271404661e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00005,
       "step": 400,
       "tokens/total": 12105616,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 182985
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.0005900752730667591,
+      "grad_norm": 0.01574699766933918,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 1.2166316082584672e-05,
+      "loss": 0.0001242965809069574,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 401,
       "tokens/total": 12136080,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 183440
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.013194791041314602,
+      "grad_norm": 0.007152364123612642,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 0.0001970212033484131,
+      "loss": 0.0001222842838615179,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0002,
+      "ppl": 1.00012,
       "step": 402,
       "tokens/total": 12166240,
-      "tokens/train_per_sec_per_gpu": 34.52,
+      "tokens/train_per_sec_per_gpu": 34.51,
       "tokens/trainable": 183917
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.02566305734217167,
+      "grad_norm": 0.0015342400874942541,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.00022389904188457876,
+      "loss": 3.621872019721195e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00022,
+      "ppl": 1.00004,
       "step": 403,
       "tokens/total": 12196544,
-      "tokens/train_per_sec_per_gpu": 30.39,
+      "tokens/train_per_sec_per_gpu": 30.36,
       "tokens/trainable": 184348
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.008275284431874752,
+      "grad_norm": 0.0035214691888540983,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 8.35009923321195e-05,
+      "loss": 6.367819150909781e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00006,
       "step": 404,
       "tokens/total": 12226784,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 184786
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.038177311420440674,
+      "grad_norm": 0.0007211520569398999,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 0.00018527230713516474,
+      "loss": 2.0325338482507505e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00019,
+      "ppl": 1.00002,
       "step": 405,
       "tokens/total": 12257264,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 185249
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.007672510575503111,
+      "grad_norm": 0.0014804014936089516,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 0.00011439670925028622,
+      "loss": 3.379736153874546e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00003,
       "step": 406,
       "tokens/total": 12287584,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 185680
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.003159885760396719,
+      "grad_norm": 0.004397980403155088,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 4.649449692806229e-05,
+      "loss": 5.1796458137687296e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00005,
       "step": 407,
       "tokens/total": 12317984,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 186100
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.32744285464286804,
+      "grad_norm": 0.0026473007164895535,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 0.002739987801760435,
+      "loss": 3.441090666456148e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00274,
+      "ppl": 1.00003,
       "step": 408,
       "tokens/total": 12348144,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.84,
       "tokens/trainable": 186565
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.0006675662589259446,
+      "grad_norm": 0.0289909727871418,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 1.2970660463906825e-05,
+      "loss": 0.00021602815832011402,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00022,
       "step": 409,
       "tokens/total": 12378544,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 187017
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.003976705949753523,
+      "grad_norm": 0.007092812564224005,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 3.857718547806144e-05,
+      "loss": 9.646185935707763e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.0001,
       "step": 410,
       "tokens/total": 12408784,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 187468
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.0024577646981924772,
+      "grad_norm": 0.02710823155939579,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 2.431379834888503e-05,
+      "loss": 0.0003458422143012285,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00035,
       "step": 411,
       "tokens/total": 12438832,
-      "tokens/train_per_sec_per_gpu": 37.61,
+      "tokens/train_per_sec_per_gpu": 37.57,
       "tokens/trainable": 187961
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003754909848794341,
+      "grad_norm": 0.29485711455345154,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 4.119630830246024e-05,
+      "loss": 0.0032923028338700533,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.0033,
       "step": 412,
       "tokens/total": 12469296,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 188447
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.3583323657512665,
+      "grad_norm": 0.22881586849689484,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.004735157359391451,
+      "loss": 0.0035809341352432966,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00475,
+      "ppl": 1.00359,
       "step": 413,
       "tokens/total": 12499392,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 188903
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.049485232681035995,
+      "grad_norm": 0.0026347371749579906,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 0.00019404500199016184,
+      "loss": 6.0475373174995184e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.00006,
       "step": 414,
       "tokens/total": 12529952,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 189398
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0030447980388998985,
+      "grad_norm": 0.0017223359318450093,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 2.0038012735312805e-05,
+      "loss": 4.137849100516178e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 415,
       "tokens/total": 12560240,
       "tokens/train_per_sec_per_gpu": 34.16,
@@ -5821,293 +5821,293 @@
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.25415608286857605,
+      "grad_norm": 0.011837545782327652,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 0.0016201161779463291,
+      "loss": 0.00011106643069069833,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00162,
+      "ppl": 1.00011,
       "step": 416,
       "tokens/total": 12590736,
-      "tokens/train_per_sec_per_gpu": 32.64,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 190306
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.006711115129292011,
+      "grad_norm": 0.01777251996099949,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 4.321872620494105e-05,
+      "loss": 0.0001219596597366035,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 417,
       "tokens/total": 12621168,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 190746
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.004931971430778503,
+      "grad_norm": 0.003218573285266757,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 4.740363510791212e-05,
+      "loss": 5.22282425663434e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00005,
       "step": 418,
       "tokens/total": 12651456,
-      "tokens/train_per_sec_per_gpu": 31.07,
+      "tokens/train_per_sec_per_gpu": 31.15,
       "tokens/trainable": 191156
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.007677357643842697,
+      "grad_norm": 0.013744015246629715,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 7.545409607701004e-05,
+      "loss": 0.0001364837517030537,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00014,
       "step": 419,
       "tokens/total": 12681760,
-      "tokens/train_per_sec_per_gpu": 29.47,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 191573
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.015445503406226635,
+      "grad_norm": 0.25013473629951477,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 0.0001045453900587745,
+      "loss": 0.002094803610816598,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0001,
+      "ppl": 1.0021,
       "step": 420,
       "tokens/total": 12712144,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.16882061958312988,
+      "grad_norm": 0.14290185272693634,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 0.0024146074429154396,
+      "loss": 0.001426510512828827,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00242,
+      "ppl": 1.00143,
       "step": 421,
       "tokens/total": 12742400,
-      "tokens/train_per_sec_per_gpu": 32.82,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 192453
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.002461223630234599,
+      "grad_norm": 0.0009390547638759017,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 3.336594454594888e-05,
+      "loss": 1.822916055971291e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12772288,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 192886
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.0005388484569266438,
+      "grad_norm": 0.004544281866401434,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 8.696397344465367e-06,
+      "loss": 6.211431173142046e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 423,
       "tokens/total": 12802880,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.87,
       "tokens/trainable": 193323
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.005009706597775221,
+      "grad_norm": 0.039859045296907425,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 4.8607362259645015e-05,
+      "loss": 0.0005113891093060374,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00051,
       "step": 424,
       "tokens/total": 12833456,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 193753
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.001982118235900998,
+      "grad_norm": 0.00408409908413887,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 3.262238169554621e-05,
+      "loss": 7.249199552461505e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 425,
       "tokens/total": 12863952,
-      "tokens/train_per_sec_per_gpu": 36.33,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 194240
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.05305991321802139,
+      "grad_norm": 0.13116440176963806,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 0.0005229170201346278,
+      "loss": 0.0015061571029946208,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00052,
+      "ppl": 1.00151,
       "step": 426,
       "tokens/total": 12894352,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 194680
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.001265498693101108,
+      "grad_norm": 0.010440024547278881,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 1.5427456673933193e-05,
+      "loss": 8.445358253084123e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00008,
       "step": 427,
       "tokens/total": 12924688,
-      "tokens/train_per_sec_per_gpu": 38.73,
+      "tokens/train_per_sec_per_gpu": 38.77,
       "tokens/trainable": 195186
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.0016263640718534589,
+      "grad_norm": 0.007594608701765537,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 2.7778178264270537e-05,
+      "loss": 9.946282807504758e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 428,
       "tokens/total": 12955072,
-      "tokens/train_per_sec_per_gpu": 36.51,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 195693
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.013330154120922089,
+      "grad_norm": 0.006465950049459934,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 6.972272240091115e-05,
+      "loss": 0.00013557568308897316,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 429,
       "tokens/total": 12985328,
-      "tokens/train_per_sec_per_gpu": 36.93,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 196130
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.0034615101758390665,
+      "grad_norm": 0.00230405549518764,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 4.500148497754708e-05,
+      "loss": 3.7114587030373514e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 430,
       "tokens/total": 13015600,
-      "tokens/train_per_sec_per_gpu": 29.58,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 196530
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.03424045443534851,
+      "grad_norm": 0.0041900877840816975,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0002949235204141587,
+      "loss": 6.116987788118422e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00029,
+      "ppl": 1.00006,
       "step": 431,
       "tokens/total": 13045728,
-      "tokens/train_per_sec_per_gpu": 29.88,
+      "tokens/train_per_sec_per_gpu": 29.94,
       "tokens/trainable": 196974
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0064499350264668465,
+      "grad_norm": 0.02951621450483799,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 7.787663344061002e-05,
+      "loss": 0.0004236411186866462,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00008,
+      "ppl": 1.00042,
       "step": 432,
       "tokens/total": 13075920,
-      "tokens/train_per_sec_per_gpu": 35.25,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 197454
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.00021908426424488425,
+      "grad_norm": 0.0003202867810614407,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 6.603059773624409e-06,
+      "loss": 1.1190046279807575e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 433,
       "tokens/total": 13106128,
-      "tokens/train_per_sec_per_gpu": 31.4,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 197902
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.0002612156968098134,
+      "grad_norm": 0.0006868013297207654,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 6.425582796509843e-06,
+      "loss": 2.1185776859056205e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 434,
       "tokens/total": 13136368,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.06,
       "tokens/trainable": 198374
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.001639618189074099,
+      "grad_norm": 0.0025407401844859123,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 2.5485322112217546e-05,
+      "loss": 4.928320413455367e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 435,
       "tokens/total": 13166704,
-      "tokens/train_per_sec_per_gpu": 37.54,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 198846
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.01887959986925125,
+      "grad_norm": 0.0013901089550927281,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0002141120348824188,
+      "loss": 3.577578900149092e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00021,
+      "ppl": 1.00004,
       "step": 436,
       "tokens/total": 13197344,
       "tokens/train_per_sec_per_gpu": 37.67,
@@ -6115,41 +6115,41 @@
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.0012294527841731906,
+      "grad_norm": 0.003646081080660224,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.316079735464882e-05,
+      "loss": 6.294051127042621e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 437,
       "tokens/total": 13227664,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 199777
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.005585651844739914,
+      "grad_norm": 0.030430182814598083,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 4.5935248635942116e-05,
+      "loss": 0.0001770013477653265,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00018,
       "step": 438,
       "tokens/total": 13257632,
-      "tokens/train_per_sec_per_gpu": 30.82,
+      "tokens/train_per_sec_per_gpu": 30.93,
       "tokens/trainable": 200194
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.001553745474666357,
+      "grad_norm": 0.0018230377463623881,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 2.2851421817904338e-05,
+      "loss": 3.7613608583342284e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13288240,
       "tokens/train_per_sec_per_gpu": 33.88,
@@ -6157,128 +6157,128 @@
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0029791900888085365,
+      "grad_norm": 0.0029376039747148752,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 4.093274037586525e-05,
+      "loss": 4.779352093464695e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 440,
       "tokens/total": 13318992,
-      "tokens/train_per_sec_per_gpu": 33.79,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 201121
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0017392379231750965,
+      "grad_norm": 0.0018373411148786545,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 7.29740804672474e-06,
+      "loss": 3.149824624415487e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 441,
       "tokens/total": 13349392,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.81,
       "tokens/trainable": 201598
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0009943305049091578,
+      "grad_norm": 0.0015428521437570453,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 1.7162077710963786e-05,
+      "loss": 2.0810390196857043e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 442,
       "tokens/total": 13379648,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 202068
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0011473585618659854,
+      "grad_norm": 0.0010770867811515927,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 2.3819740817998536e-05,
+      "loss": 2.303836117789615e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 443,
       "tokens/total": 13409936,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 202524
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.006799118127673864,
+      "grad_norm": 0.006372133269906044,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 6.556943117175251e-05,
+      "loss": 7.449327677022666e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
       "ppl": 1.00007,
       "step": 444,
       "tokens/total": 13440288,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 202996
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.0011645558988675475,
+      "grad_norm": 0.007094603031873703,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 1.8389995602774434e-05,
+      "loss": 4.211071063764393e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 445,
       "tokens/total": 13468880,
-      "tokens/train_per_sec_per_gpu": 39.14,
+      "tokens/train_per_sec_per_gpu": 39.19,
       "tokens/trainable": 203470
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.020891210064291954,
+      "grad_norm": 0.006931444630026817,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 0.00016399261949118227,
+      "loss": 8.66219779709354e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00016,
+      "ppl": 1.00009,
       "step": 446,
       "tokens/total": 13499184,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 203919
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.09615519642829895,
+      "grad_norm": 0.0041664596647024155,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 0.0010675137164071202,
+      "loss": 4.786982390214689e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00107,
+      "ppl": 1.00005,
       "step": 447,
       "tokens/total": 13529632,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 204376
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.09744399785995483,
+      "grad_norm": 0.042264848947525024,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.001196134602651,
+      "loss": 0.00027092613163404167,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0012,
+      "ppl": 1.00027,
       "step": 448,
       "tokens/total": 13560080,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 204821
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_model.safetensors
index 8e3393e0d9d830da40019eea8976c5200652c786..5ba7adf7daa2c6973e8b767c3878e00a270cd473 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8574c62e5993d868e285944d01ea5c0f530e591f2154d08a0c57b6eadfedd47d
+oid sha256:759e2f47a5e976dcfbe11dde9491135d96d1b05bada169151f4a8f939269acc5
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/optimizer.pt
index 8c06eb3a25c2ac81edc1245d8a86c290ba00e0dc..f9173dd0543c7e5e03c5a94bc3c36e1ee50c22f9 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d83b575649c6ab2f7828c29ac5ae5dcb50824f042fa45adedc229558f4ef1923
+oid sha256:cd01ae9ec3eb43f4befc813946bc00ce45c82d3237dfa7f15ddccfe7878ea01d
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/trainer_state.json
index d02d36e752100f5efb4a3d138ed927d6e4adeb13..02c043ff803a1bb9f1e3315e32af7dc8b64659cd 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-480/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,1105 +4715,1105 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.370597779750824,
+      "grad_norm": 0.03195308893918991,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 0.007550997193902731,
+      "loss": 0.0002666166110429913,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00758,
+      "ppl": 1.00027,
       "step": 353,
       "tokens/total": 10684176,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 161420
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0031376827973872423,
+      "grad_norm": 0.07012991607189178,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 2.3453332687495276e-05,
+      "loss": 0.000741704716347158,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00074,
       "step": 354,
       "tokens/total": 10714352,
-      "tokens/train_per_sec_per_gpu": 34.02,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 161868
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.0021399864926934242,
+      "grad_norm": 0.0005822654929943383,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 2.5832894607447088e-05,
+      "loss": 1.0378402294008993e-05,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 355,
       "tokens/total": 10744464,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 162318
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0028148347046226263,
+      "grad_norm": 0.012435230426490307,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 3.7286932638380677e-05,
+      "loss": 7.89838086348027e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00008,
       "step": 356,
       "tokens/total": 10774976,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.54,
       "tokens/trainable": 162789
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.1642119586467743,
+      "grad_norm": 0.0045598647557199,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 0.0027845175936818123,
+      "loss": 2.48450869548833e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00279,
+      "ppl": 1.00002,
       "step": 357,
       "tokens/total": 10805600,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 163254
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.066535085439682,
+      "grad_norm": 0.3044262230396271,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.0002488417667336762,
+      "loss": 0.002298796083778143,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00025,
+      "ppl": 1.0023,
       "step": 358,
       "tokens/total": 10835920,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.62,
       "tokens/trainable": 163715
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.006617655046284199,
+      "grad_norm": 0.002458421979099512,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 7.841112528694794e-05,
+      "loss": 1.9634167983895168e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 359,
       "tokens/total": 10866048,
-      "tokens/train_per_sec_per_gpu": 34.4,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 164197
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.010409035719931126,
+      "grad_norm": 0.0024773837067186832,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010489403939573094,
+      "loss": 2.744927041931078e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 360,
       "tokens/total": 10896288,
-      "tokens/train_per_sec_per_gpu": 31.6,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 164661
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.002426267135888338,
+      "grad_norm": 0.00026717092259787023,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 4.0343060391023755e-05,
+      "loss": 6.422977094189264e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 361,
       "tokens/total": 10926816,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 165111
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.015666797757148743,
+      "grad_norm": 0.0008353493758477271,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 0.00014991794887464494,
+      "loss": 1.3549893083109055e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 362,
       "tokens/total": 10954928,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 165552
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.2695651650428772,
+      "grad_norm": 0.006071125157177448,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0014983330620452762,
+      "loss": 6.428411870729178e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.0015,
+      "ppl": 1.00006,
       "step": 363,
       "tokens/total": 10985328,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 166026
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.03586212173104286,
+      "grad_norm": 0.25424960255622864,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.00030509717180393636,
+      "loss": 0.004243595991283655,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00031,
+      "ppl": 1.00425,
       "step": 364,
       "tokens/total": 11015664,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 166486
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.0019308789633214474,
+      "grad_norm": 0.0005575277027674019,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 2.4129443772835657e-05,
+      "loss": 8.403902938880492e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 365,
       "tokens/total": 11046256,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 166959
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005940837785601616,
+      "grad_norm": 0.0014335239538922906,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 2.8238933737156913e-05,
+      "loss": 1.4738036043127067e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 366,
       "tokens/total": 11076288,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 167394
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.002125627128407359,
+      "grad_norm": 0.0010152696631848812,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 1.602789416210726e-05,
+      "loss": 1.3278609912958927e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 367,
       "tokens/total": 11106416,
-      "tokens/train_per_sec_per_gpu": 28.5,
+      "tokens/train_per_sec_per_gpu": 28.58,
       "tokens/trainable": 167832
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.2510211765766144,
+      "grad_norm": 0.11941836029291153,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 0.00010395953722763807,
+      "loss": 0.0006476733833551407,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00065,
       "step": 368,
       "tokens/total": 11136848,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 31.02,
       "tokens/trainable": 168242
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.001904280623421073,
+      "grad_norm": 0.006877629552036524,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 2.6418363631819375e-05,
+      "loss": 7.449048280250281e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 369,
       "tokens/total": 11167376,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 168715
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.0028152521699666977,
+      "grad_norm": 0.0007151139434427023,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 2.9974533390486613e-05,
+      "loss": 1.1437590728746727e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 370,
       "tokens/total": 11197776,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.86,
       "tokens/trainable": 169147
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.0010822078911587596,
+      "grad_norm": 0.009094453416764736,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 1.34217716549756e-05,
+      "loss": 5.9242345741949975e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11227840,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 169628
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.005367911420762539,
+      "grad_norm": 0.000247267191298306,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 6.65646803099662e-05,
+      "loss": 7.548428584414069e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 372,
       "tokens/total": 11258176,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 170085
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.047570567578077316,
+      "grad_norm": 0.0003029497747775167,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.0003471036907285452,
+      "loss": 5.4013939916330855e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00035,
+      "ppl": 1.00001,
       "step": 373,
       "tokens/total": 11288336,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 170584
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.040886107832193375,
+      "grad_norm": 0.001942179980687797,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.0002880148240365088,
+      "loss": 2.380511250521522e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00029,
+      "ppl": 1.00002,
       "step": 374,
       "tokens/total": 11318848,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 171058
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.0031216361094266176,
+      "grad_norm": 0.0019308892078697681,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 2.690855944820214e-05,
+      "loss": 1.5233906196954194e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 375,
       "tokens/total": 11349168,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 171518
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.012176645919680595,
+      "grad_norm": 0.6044301986694336,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 8.402287494391203e-05,
+      "loss": 0.006350134499371052,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00008,
+      "ppl": 1.00637,
       "step": 376,
       "tokens/total": 11379328,
-      "tokens/train_per_sec_per_gpu": 40.48,
+      "tokens/train_per_sec_per_gpu": 40.62,
       "tokens/trainable": 172007
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.24136756360530853,
+      "grad_norm": 0.25446587800979614,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 0.0013303000014275312,
+      "loss": 0.0018724403344094753,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00133,
+      "ppl": 1.00187,
       "step": 377,
       "tokens/total": 11409568,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 172468
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.021714258939027786,
+      "grad_norm": 0.0019207323202863336,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 0.00020160498388577253,
+      "loss": 3.049923907383345e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 378,
       "tokens/total": 11439968,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 172923
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.35412073135375977,
+      "grad_norm": 0.33491548895835876,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 0.015761105343699455,
+      "loss": 0.02673855796456337,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01589,
+      "ppl": 1.0271,
       "step": 379,
       "tokens/total": 11470496,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 173370
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005766173824667931,
+      "grad_norm": 0.006177723873406649,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 2.5790239305933937e-05,
+      "loss": 5.195035191718489e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 380,
       "tokens/total": 11500720,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 173837
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.005898744333535433,
+      "grad_norm": 0.06973031163215637,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 5.821501690661535e-05,
+      "loss": 0.0006310560274869204,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00006,
+      "ppl": 1.00063,
       "step": 381,
       "tokens/total": 11531280,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 174294
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005451703444123268,
+      "grad_norm": 0.0039021887350827456,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 4.0181505028158426e-05,
+      "loss": 4.849781544180587e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 382,
       "tokens/total": 11561264,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.84,
       "tokens/trainable": 174701
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0038302047178149223,
+      "grad_norm": 0.0007975143962539732,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 3.403786467970349e-05,
+      "loss": 2.3489263185183518e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 383,
       "tokens/total": 11591616,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 175131
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.004227485507726669,
+      "grad_norm": 0.012541128322482109,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 3.562243364285678e-05,
+      "loss": 0.00011495217040646821,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00011,
       "step": 384,
       "tokens/total": 11622000,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 175586
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.013361346907913685,
+      "grad_norm": 0.004752015229314566,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0001464220113120973,
+      "loss": 6.502695032395422e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00007,
       "step": 385,
       "tokens/total": 11652352,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176026
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.008831475861370564,
+      "grad_norm": 0.014192809350788593,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 8.2662510976661e-05,
+      "loss": 8.075163350440562e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00008,
       "step": 386,
       "tokens/total": 11682736,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.91,
       "tokens/trainable": 176512
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.009311008267104626,
+      "grad_norm": 0.007288333959877491,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 0.0001226613821927458,
+      "loss": 8.845872798701748e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00009,
       "step": 387,
       "tokens/total": 11713136,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.45,
       "tokens/trainable": 176943
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.0017816838808357716,
+      "grad_norm": 0.10162956267595291,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 2.3221660740091465e-05,
+      "loss": 0.0007253064541146159,
       "memory/device_reserved (GiB)": 35.23,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00073,
       "step": 388,
       "tokens/total": 11743808,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 177370
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.023602057248353958,
+      "grad_norm": 0.005729300435632467,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 0.00012638044427148998,
+      "loss": 8.652975520817563e-05,
       "memory/device_reserved (GiB)": 35.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00013,
+      "ppl": 1.00009,
       "step": 389,
       "tokens/total": 11774144,
-      "tokens/train_per_sec_per_gpu": 35.14,
+      "tokens/train_per_sec_per_gpu": 35.2,
       "tokens/trainable": 177851
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.04383797571063042,
+      "grad_norm": 0.0715414360165596,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 0.0004640861588995904,
+      "loss": 0.0007950748549774289,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00046,
+      "ppl": 1.0008,
       "step": 390,
       "tokens/total": 11804272,
-      "tokens/train_per_sec_per_gpu": 34.68,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 178310
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.061932601034641266,
+      "grad_norm": 0.009775097481906414,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.00033243370126001537,
+      "loss": 0.00010615254723234102,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00033,
+      "ppl": 1.00011,
       "step": 391,
       "tokens/total": 11832432,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 178736
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.04983547702431679,
+      "grad_norm": 0.0015552763361483812,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 0.000599355495069176,
+      "loss": 2.954876617877744e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0006,
+      "ppl": 1.00003,
       "step": 392,
       "tokens/total": 11862960,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 179233
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 0.012036106549203396,
+      "grad_norm": 0.026106838136911392,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0001237639953615144,
+      "loss": 0.00021109850786160678,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00021,
       "step": 393,
       "tokens/total": 11893232,
-      "tokens/train_per_sec_per_gpu": 38.49,
+      "tokens/train_per_sec_per_gpu": 38.57,
       "tokens/trainable": 179730
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.008992817252874374,
+      "grad_norm": 0.020347680896520615,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 0.00010385089262854308,
+      "loss": 0.0002601295418571681,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0001,
+      "ppl": 1.00026,
       "step": 394,
       "tokens/total": 11923664,
-      "tokens/train_per_sec_per_gpu": 37.81,
+      "tokens/train_per_sec_per_gpu": 37.87,
       "tokens/trainable": 180227
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.24631057679653168,
+      "grad_norm": 0.005419979803264141,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.0031685903668403625,
+      "loss": 4.929217175231315e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00317,
+      "ppl": 1.00005,
       "step": 395,
       "tokens/total": 11954032,
-      "tokens/train_per_sec_per_gpu": 32.85,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 180687
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.06871633976697922,
+      "grad_norm": 0.0031002764590084553,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 0.0005168755305930972,
+      "loss": 6.144218787085265e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00052,
+      "ppl": 1.00006,
       "step": 396,
       "tokens/total": 11984512,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 181141
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.011674679815769196,
+      "grad_norm": 0.006748533807694912,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 0.00017496946384198964,
+      "loss": 9.171784040518105e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 397,
       "tokens/total": 12014880,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 36.02,
       "tokens/trainable": 181648
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.14610664546489716,
+      "grad_norm": 0.08927815407514572,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.0014660547021776438,
+      "loss": 0.001000746968202293,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00147,
+      "ppl": 1.001,
       "step": 398,
       "tokens/total": 12044928,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 182109
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.016359565779566765,
+      "grad_norm": 0.004651801194995642,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 7.928608101792634e-05,
+      "loss": 5.3439554903889075e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00008,
+      "ppl": 1.00005,
       "step": 399,
       "tokens/total": 12075344,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 182559
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0010484450031071901,
+      "grad_norm": 0.0017514342907816172,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 1.526270352769643e-05,
+      "loss": 4.570486271404661e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00005,
       "step": 400,
       "tokens/total": 12105616,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 182985
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.0005900752730667591,
+      "grad_norm": 0.01574699766933918,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 1.2166316082584672e-05,
+      "loss": 0.0001242965809069574,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 401,
       "tokens/total": 12136080,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 183440
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.013194791041314602,
+      "grad_norm": 0.007152364123612642,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 0.0001970212033484131,
+      "loss": 0.0001222842838615179,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0002,
+      "ppl": 1.00012,
       "step": 402,
       "tokens/total": 12166240,
-      "tokens/train_per_sec_per_gpu": 34.52,
+      "tokens/train_per_sec_per_gpu": 34.51,
       "tokens/trainable": 183917
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.02566305734217167,
+      "grad_norm": 0.0015342400874942541,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.00022389904188457876,
+      "loss": 3.621872019721195e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00022,
+      "ppl": 1.00004,
       "step": 403,
       "tokens/total": 12196544,
-      "tokens/train_per_sec_per_gpu": 30.39,
+      "tokens/train_per_sec_per_gpu": 30.36,
       "tokens/trainable": 184348
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.008275284431874752,
+      "grad_norm": 0.0035214691888540983,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 8.35009923321195e-05,
+      "loss": 6.367819150909781e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00006,
       "step": 404,
       "tokens/total": 12226784,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 184786
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.038177311420440674,
+      "grad_norm": 0.0007211520569398999,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 0.00018527230713516474,
+      "loss": 2.0325338482507505e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00019,
+      "ppl": 1.00002,
       "step": 405,
       "tokens/total": 12257264,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 185249
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.007672510575503111,
+      "grad_norm": 0.0014804014936089516,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 0.00011439670925028622,
+      "loss": 3.379736153874546e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00003,
       "step": 406,
       "tokens/total": 12287584,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 185680
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.003159885760396719,
+      "grad_norm": 0.004397980403155088,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 4.649449692806229e-05,
+      "loss": 5.1796458137687296e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00005,
       "step": 407,
       "tokens/total": 12317984,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 186100
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.32744285464286804,
+      "grad_norm": 0.0026473007164895535,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 0.002739987801760435,
+      "loss": 3.441090666456148e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00274,
+      "ppl": 1.00003,
       "step": 408,
       "tokens/total": 12348144,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.84,
       "tokens/trainable": 186565
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.0006675662589259446,
+      "grad_norm": 0.0289909727871418,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 1.2970660463906825e-05,
+      "loss": 0.00021602815832011402,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00022,
       "step": 409,
       "tokens/total": 12378544,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 187017
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.003976705949753523,
+      "grad_norm": 0.007092812564224005,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 3.857718547806144e-05,
+      "loss": 9.646185935707763e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.0001,
       "step": 410,
       "tokens/total": 12408784,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 187468
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.0024577646981924772,
+      "grad_norm": 0.02710823155939579,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 2.431379834888503e-05,
+      "loss": 0.0003458422143012285,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00035,
       "step": 411,
       "tokens/total": 12438832,
-      "tokens/train_per_sec_per_gpu": 37.61,
+      "tokens/train_per_sec_per_gpu": 37.57,
       "tokens/trainable": 187961
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003754909848794341,
+      "grad_norm": 0.29485711455345154,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 4.119630830246024e-05,
+      "loss": 0.0032923028338700533,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.0033,
       "step": 412,
       "tokens/total": 12469296,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 188447
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.3583323657512665,
+      "grad_norm": 0.22881586849689484,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.004735157359391451,
+      "loss": 0.0035809341352432966,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00475,
+      "ppl": 1.00359,
       "step": 413,
       "tokens/total": 12499392,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 188903
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.049485232681035995,
+      "grad_norm": 0.0026347371749579906,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 0.00019404500199016184,
+      "loss": 6.0475373174995184e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.00006,
       "step": 414,
       "tokens/total": 12529952,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 189398
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0030447980388998985,
+      "grad_norm": 0.0017223359318450093,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 2.0038012735312805e-05,
+      "loss": 4.137849100516178e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 415,
       "tokens/total": 12560240,
       "tokens/train_per_sec_per_gpu": 34.16,
@@ -5821,293 +5821,293 @@
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.25415608286857605,
+      "grad_norm": 0.011837545782327652,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 0.0016201161779463291,
+      "loss": 0.00011106643069069833,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00162,
+      "ppl": 1.00011,
       "step": 416,
       "tokens/total": 12590736,
-      "tokens/train_per_sec_per_gpu": 32.64,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 190306
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.006711115129292011,
+      "grad_norm": 0.01777251996099949,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 4.321872620494105e-05,
+      "loss": 0.0001219596597366035,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 417,
       "tokens/total": 12621168,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 190746
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.004931971430778503,
+      "grad_norm": 0.003218573285266757,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 4.740363510791212e-05,
+      "loss": 5.22282425663434e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00005,
       "step": 418,
       "tokens/total": 12651456,
-      "tokens/train_per_sec_per_gpu": 31.07,
+      "tokens/train_per_sec_per_gpu": 31.15,
       "tokens/trainable": 191156
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.007677357643842697,
+      "grad_norm": 0.013744015246629715,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 7.545409607701004e-05,
+      "loss": 0.0001364837517030537,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00014,
       "step": 419,
       "tokens/total": 12681760,
-      "tokens/train_per_sec_per_gpu": 29.47,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 191573
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.015445503406226635,
+      "grad_norm": 0.25013473629951477,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 0.0001045453900587745,
+      "loss": 0.002094803610816598,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0001,
+      "ppl": 1.0021,
       "step": 420,
       "tokens/total": 12712144,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.16882061958312988,
+      "grad_norm": 0.14290185272693634,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 0.0024146074429154396,
+      "loss": 0.001426510512828827,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00242,
+      "ppl": 1.00143,
       "step": 421,
       "tokens/total": 12742400,
-      "tokens/train_per_sec_per_gpu": 32.82,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 192453
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.002461223630234599,
+      "grad_norm": 0.0009390547638759017,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 3.336594454594888e-05,
+      "loss": 1.822916055971291e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12772288,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 192886
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.0005388484569266438,
+      "grad_norm": 0.004544281866401434,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 8.696397344465367e-06,
+      "loss": 6.211431173142046e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 423,
       "tokens/total": 12802880,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.87,
       "tokens/trainable": 193323
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.005009706597775221,
+      "grad_norm": 0.039859045296907425,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 4.8607362259645015e-05,
+      "loss": 0.0005113891093060374,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00051,
       "step": 424,
       "tokens/total": 12833456,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 193753
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.001982118235900998,
+      "grad_norm": 0.00408409908413887,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 3.262238169554621e-05,
+      "loss": 7.249199552461505e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 425,
       "tokens/total": 12863952,
-      "tokens/train_per_sec_per_gpu": 36.33,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 194240
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.05305991321802139,
+      "grad_norm": 0.13116440176963806,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 0.0005229170201346278,
+      "loss": 0.0015061571029946208,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00052,
+      "ppl": 1.00151,
       "step": 426,
       "tokens/total": 12894352,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 194680
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.001265498693101108,
+      "grad_norm": 0.010440024547278881,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 1.5427456673933193e-05,
+      "loss": 8.445358253084123e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00008,
       "step": 427,
       "tokens/total": 12924688,
-      "tokens/train_per_sec_per_gpu": 38.73,
+      "tokens/train_per_sec_per_gpu": 38.77,
       "tokens/trainable": 195186
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.0016263640718534589,
+      "grad_norm": 0.007594608701765537,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 2.7778178264270537e-05,
+      "loss": 9.946282807504758e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 428,
       "tokens/total": 12955072,
-      "tokens/train_per_sec_per_gpu": 36.51,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 195693
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.013330154120922089,
+      "grad_norm": 0.006465950049459934,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 6.972272240091115e-05,
+      "loss": 0.00013557568308897316,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 429,
       "tokens/total": 12985328,
-      "tokens/train_per_sec_per_gpu": 36.93,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 196130
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.0034615101758390665,
+      "grad_norm": 0.00230405549518764,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 4.500148497754708e-05,
+      "loss": 3.7114587030373514e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 430,
       "tokens/total": 13015600,
-      "tokens/train_per_sec_per_gpu": 29.58,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 196530
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.03424045443534851,
+      "grad_norm": 0.0041900877840816975,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0002949235204141587,
+      "loss": 6.116987788118422e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00029,
+      "ppl": 1.00006,
       "step": 431,
       "tokens/total": 13045728,
-      "tokens/train_per_sec_per_gpu": 29.88,
+      "tokens/train_per_sec_per_gpu": 29.94,
       "tokens/trainable": 196974
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0064499350264668465,
+      "grad_norm": 0.02951621450483799,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 7.787663344061002e-05,
+      "loss": 0.0004236411186866462,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00008,
+      "ppl": 1.00042,
       "step": 432,
       "tokens/total": 13075920,
-      "tokens/train_per_sec_per_gpu": 35.25,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 197454
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.00021908426424488425,
+      "grad_norm": 0.0003202867810614407,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 6.603059773624409e-06,
+      "loss": 1.1190046279807575e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 433,
       "tokens/total": 13106128,
-      "tokens/train_per_sec_per_gpu": 31.4,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 197902
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.0002612156968098134,
+      "grad_norm": 0.0006868013297207654,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 6.425582796509843e-06,
+      "loss": 2.1185776859056205e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 434,
       "tokens/total": 13136368,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.06,
       "tokens/trainable": 198374
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.001639618189074099,
+      "grad_norm": 0.0025407401844859123,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 2.5485322112217546e-05,
+      "loss": 4.928320413455367e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 435,
       "tokens/total": 13166704,
-      "tokens/train_per_sec_per_gpu": 37.54,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 198846
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.01887959986925125,
+      "grad_norm": 0.0013901089550927281,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0002141120348824188,
+      "loss": 3.577578900149092e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00021,
+      "ppl": 1.00004,
       "step": 436,
       "tokens/total": 13197344,
       "tokens/train_per_sec_per_gpu": 37.67,
@@ -6115,41 +6115,41 @@
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.0012294527841731906,
+      "grad_norm": 0.003646081080660224,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.316079735464882e-05,
+      "loss": 6.294051127042621e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 437,
       "tokens/total": 13227664,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 199777
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.005585651844739914,
+      "grad_norm": 0.030430182814598083,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 4.5935248635942116e-05,
+      "loss": 0.0001770013477653265,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00018,
       "step": 438,
       "tokens/total": 13257632,
-      "tokens/train_per_sec_per_gpu": 30.82,
+      "tokens/train_per_sec_per_gpu": 30.93,
       "tokens/trainable": 200194
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.001553745474666357,
+      "grad_norm": 0.0018230377463623881,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 2.2851421817904338e-05,
+      "loss": 3.7613608583342284e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13288240,
       "tokens/train_per_sec_per_gpu": 33.88,
@@ -6157,559 +6157,559 @@
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0029791900888085365,
+      "grad_norm": 0.0029376039747148752,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 4.093274037586525e-05,
+      "loss": 4.779352093464695e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 440,
       "tokens/total": 13318992,
-      "tokens/train_per_sec_per_gpu": 33.79,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 201121
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0017392379231750965,
+      "grad_norm": 0.0018373411148786545,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 7.29740804672474e-06,
+      "loss": 3.149824624415487e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 441,
       "tokens/total": 13349392,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.81,
       "tokens/trainable": 201598
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0009943305049091578,
+      "grad_norm": 0.0015428521437570453,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 1.7162077710963786e-05,
+      "loss": 2.0810390196857043e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 442,
       "tokens/total": 13379648,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 202068
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0011473585618659854,
+      "grad_norm": 0.0010770867811515927,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 2.3819740817998536e-05,
+      "loss": 2.303836117789615e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 443,
       "tokens/total": 13409936,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 202524
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.006799118127673864,
+      "grad_norm": 0.006372133269906044,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 6.556943117175251e-05,
+      "loss": 7.449327677022666e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
       "ppl": 1.00007,
       "step": 444,
       "tokens/total": 13440288,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 202996
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.0011645558988675475,
+      "grad_norm": 0.007094603031873703,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 1.8389995602774434e-05,
+      "loss": 4.211071063764393e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 445,
       "tokens/total": 13468880,
-      "tokens/train_per_sec_per_gpu": 39.14,
+      "tokens/train_per_sec_per_gpu": 39.19,
       "tokens/trainable": 203470
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.020891210064291954,
+      "grad_norm": 0.006931444630026817,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 0.00016399261949118227,
+      "loss": 8.66219779709354e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00016,
+      "ppl": 1.00009,
       "step": 446,
       "tokens/total": 13499184,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 203919
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.09615519642829895,
+      "grad_norm": 0.0041664596647024155,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 0.0010675137164071202,
+      "loss": 4.786982390214689e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00107,
+      "ppl": 1.00005,
       "step": 447,
       "tokens/total": 13529632,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 204376
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.09744399785995483,
+      "grad_norm": 0.042264848947525024,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.001196134602651,
+      "loss": 0.00027092613163404167,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0012,
+      "ppl": 1.00027,
       "step": 448,
       "tokens/total": 13560080,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 204821
     },
     {
       "epoch": 1.75390625,
-      "grad_norm": 0.0007810555398464203,
+      "grad_norm": 0.00043735766666941345,
       "learning_rate": 1.3780999708818058e-05,
-      "loss": 8.019956112548243e-06,
+      "loss": 1.2837479516747408e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
       "ppl": 1.00001,
       "step": 449,
       "tokens/total": 13590496,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 205271
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 0.034087613224983215,
+      "grad_norm": 0.0012977999867871404,
       "learning_rate": 1.3665385037857758e-05,
-      "loss": 0.00010244422446703538,
+      "loss": 2.1102820028318092e-05,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0001,
+      "ppl": 1.00002,
       "step": 450,
       "tokens/total": 13620960,
-      "tokens/train_per_sec_per_gpu": 30.95,
+      "tokens/train_per_sec_per_gpu": 30.92,
       "tokens/trainable": 205723
     },
     {
       "epoch": 1.76171875,
-      "grad_norm": 0.02930639684200287,
+      "grad_norm": 0.005675352178514004,
       "learning_rate": 1.3551490468947126e-05,
-      "loss": 0.00021926099725533277,
+      "loss": 7.465962698915973e-05,
       "memory/device_reserved (GiB)": 35.04,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00022,
+      "ppl": 1.00007,
       "step": 451,
       "tokens/total": 13651344,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.48,
       "tokens/trainable": 206169
     },
     {
       "epoch": 1.765625,
-      "grad_norm": 0.0109552051872015,
+      "grad_norm": 0.0013997588539496064,
       "learning_rate": 1.3439320741704075e-05,
-      "loss": 7.999646186362952e-05,
+      "loss": 2.723011130001396e-05,
       "memory/device_reserved (GiB)": 35.04,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 452,
       "tokens/total": 13681792,
-      "tokens/train_per_sec_per_gpu": 32.74,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 206592
     },
     {
       "epoch": 1.76953125,
-      "grad_norm": 0.004453443922102451,
+      "grad_norm": 0.012491011992096901,
       "learning_rate": 1.3328880523968808e-05,
-      "loss": 3.79011980840005e-05,
+      "loss": 9.182744543068111e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00009,
       "step": 453,
       "tokens/total": 13712256,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.07,
       "tokens/trainable": 207055
     },
     {
       "epoch": 1.7734375,
-      "grad_norm": 0.0005650459788739681,
+      "grad_norm": 0.000924609019421041,
       "learning_rate": 1.3220174411609587e-05,
-      "loss": 1.2244867320987396e-05,
+      "loss": 2.1621295672957785e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 454,
       "tokens/total": 13742128,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.61,
       "tokens/trainable": 207494
     },
     {
       "epoch": 1.77734375,
-      "grad_norm": 0.0007141040405258536,
+      "grad_norm": 0.000411411514505744,
       "learning_rate": 1.3113206928331471e-05,
-      "loss": 1.3571594536188059e-05,
+      "loss": 1.2427874025888741e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 455,
       "tokens/total": 13772160,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.03,
       "tokens/trainable": 207956
     },
     {
       "epoch": 1.78125,
-      "grad_norm": 0.0008097058744169772,
+      "grad_norm": 0.0015909038484096527,
       "learning_rate": 1.300798252548806e-05,
-      "loss": 1.1659047231660224e-05,
+      "loss": 3.0332066671689972e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 456,
       "tokens/total": 13802448,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.42,
       "tokens/trainable": 208408
     },
     {
       "epoch": 1.78515625,
-      "grad_norm": 0.0012649539858102798,
+      "grad_norm": 0.0030019236728549004,
       "learning_rate": 1.2904505581896265e-05,
-      "loss": 9.200517524732277e-06,
+      "loss": 3.378765541128814e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 457,
       "tokens/total": 13832880,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 208889
     },
     {
       "epoch": 1.7890625,
-      "grad_norm": 0.014978500083088875,
+      "grad_norm": 0.0016427476657554507,
       "learning_rate": 1.2802780403654082e-05,
-      "loss": 0.0001208957692142576,
+      "loss": 3.3130341762444004e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00012,
+      "ppl": 1.00003,
       "step": 458,
       "tokens/total": 13862880,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 209336
     },
     {
       "epoch": 1.79296875,
-      "grad_norm": 0.004875013139098883,
+      "grad_norm": 0.0006426559993997216,
       "learning_rate": 1.2702811223961408e-05,
-      "loss": 3.241455851821229e-05,
+      "loss": 1.784306368790567e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 459,
       "tokens/total": 13893440,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.21,
       "tokens/trainable": 209797
     },
     {
       "epoch": 1.796875,
-      "grad_norm": 0.0009627968538552523,
+      "grad_norm": 0.0016685863956809044,
       "learning_rate": 1.2604602202943861e-05,
-      "loss": 1.980438355531078e-05,
+      "loss": 3.2843898225110024e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 460,
       "tokens/total": 13923824,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.03,
       "tokens/trainable": 210240
     },
     {
       "epoch": 1.80078125,
-      "grad_norm": 0.019912317395210266,
+      "grad_norm": 0.03454679995775223,
       "learning_rate": 1.2508157427479686e-05,
-      "loss": 0.00010681153071345761,
+      "loss": 7.200831169029698e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00011,
+      "ppl": 1.00007,
       "step": 461,
       "tokens/total": 13954208,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 34.9,
       "tokens/trainable": 210702
     },
     {
       "epoch": 1.8046875,
-      "grad_norm": 0.0005370390135794878,
+      "grad_norm": 0.0021488473284989595,
       "learning_rate": 1.2413480911029655e-05,
-      "loss": 1.1081473530794028e-05,
+      "loss": 3.525309875840321e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 462,
       "tokens/total": 13984784,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.44,
       "tokens/trainable": 211140
     },
     {
       "epoch": 1.80859375,
-      "grad_norm": 0.000815370527561754,
+      "grad_norm": 0.0007275328389368951,
       "learning_rate": 1.2320576593470082e-05,
-      "loss": 1.5510300727328286e-05,
+      "loss": 1.328821963397786e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 463,
       "tokens/total": 14014896,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.54,
       "tokens/trainable": 211575
     },
     {
       "epoch": 1.8125,
-      "grad_norm": 0.02875823900103569,
+      "grad_norm": 0.03173735365271568,
       "learning_rate": 1.2229448340928828e-05,
-      "loss": 0.00013141569797880948,
+      "loss": 0.00014542456483468413,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00013,
+      "ppl": 1.00015,
       "step": 464,
       "tokens/total": 14045312,
-      "tokens/train_per_sec_per_gpu": 39.03,
+      "tokens/train_per_sec_per_gpu": 38.86,
       "tokens/trainable": 212083
     },
     {
       "epoch": 1.81640625,
-      "grad_norm": 0.04558868706226349,
+      "grad_norm": 0.0005439479718916118,
       "learning_rate": 1.2140099945624458e-05,
-      "loss": 0.0002969688503071666,
+      "loss": 1.4493984053842723e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0003,
+      "ppl": 1.00001,
       "step": 465,
       "tokens/total": 14075840,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 212558
     },
     {
       "epoch": 1.8203125,
-      "grad_norm": 0.0004122587270103395,
+      "grad_norm": 0.0015477532288059592,
       "learning_rate": 1.205253512570841e-05,
-      "loss": 9.937594768416602e-06,
+      "loss": 2.787737685139291e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 466,
       "tokens/total": 14106016,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.73,
       "tokens/trainable": 213014
     },
     {
       "epoch": 1.82421875,
-      "grad_norm": 0.0010108116548508406,
+      "grad_norm": 0.0007428264361806214,
       "learning_rate": 1.1966757525110255e-05,
-      "loss": 1.5579567843815312e-05,
+      "loss": 2.1110219677211717e-05,
       "memory/device_reserved (GiB)": 37.99,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
       "ppl": 1.00002,
       "step": 467,
       "tokens/total": 14136512,
-      "tokens/train_per_sec_per_gpu": 34.03,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 213465
     },
     {
       "epoch": 1.828125,
-      "grad_norm": 0.0043095871806144714,
+      "grad_norm": 0.002182955853641033,
       "learning_rate": 1.1882770713386095e-05,
-      "loss": 3.2609641493763775e-05,
+      "loss": 4.059498314745724e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00004,
       "step": 468,
       "tokens/total": 14166832,
-      "tokens/train_per_sec_per_gpu": 36.62,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 213940
     },
     {
       "epoch": 1.83203125,
-      "grad_norm": 0.005317045841366053,
+      "grad_norm": 0.004772448446601629,
       "learning_rate": 1.180057818556998e-05,
-      "loss": 2.144884638255462e-05,
+      "loss": 3.664854375529103e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 469,
       "tokens/total": 14197360,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 214415
     },
     {
       "epoch": 1.8359375,
-      "grad_norm": 0.0003264908737037331,
+      "grad_norm": 0.00038694078102707863,
       "learning_rate": 1.1720183362028494e-05,
-      "loss": 5.6935500651889015e-06,
+      "loss": 1.1124819138785824e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 470,
       "tokens/total": 14227600,
-      "tokens/train_per_sec_per_gpu": 34.26,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 214880
     },
     {
       "epoch": 1.83984375,
-      "grad_norm": 0.013723314739763737,
+      "grad_norm": 0.0008439846569672227,
       "learning_rate": 1.1641589588318387e-05,
-      "loss": 0.00010619591921567917,
+      "loss": 2.0906983991153538e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00011,
+      "ppl": 1.00002,
       "step": 471,
       "tokens/total": 14257888,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 215334
     },
     {
       "epoch": 1.84375,
-      "grad_norm": 0.00020912640320602804,
+      "grad_norm": 0.0005926606827415526,
       "learning_rate": 1.1564800135047418e-05,
-      "loss": 4.586940121953376e-06,
+      "loss": 1.178990351036191e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 472,
       "tokens/total": 14288208,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 215822
     },
     {
       "epoch": 1.84765625,
-      "grad_norm": 0.10667876899242401,
+      "grad_norm": 0.003897819435223937,
       "learning_rate": 1.148981819773816e-05,
-      "loss": 0.0006915333215147257,
+      "loss": 5.020621756557375e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.00005,
       "step": 473,
       "tokens/total": 14318448,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 30.82,
       "tokens/trainable": 216239
     },
     {
       "epoch": 1.8515625,
-      "grad_norm": 0.013282700441777706,
+      "grad_norm": 0.0013807902578264475,
       "learning_rate": 1.1416646896695086e-05,
-      "loss": 6.861680594738573e-05,
+      "loss": 2.2795318727730773e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 474,
       "tokens/total": 14348912,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.35,
       "tokens/trainable": 216691
     },
     {
       "epoch": 1.85546875,
-      "grad_norm": 0.0074070473201572895,
+      "grad_norm": 0.00048289448022842407,
       "learning_rate": 1.1345289276874717e-05,
-      "loss": 8.334079757332802e-05,
+      "loss": 1.4630711120844353e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 475,
       "tokens/total": 14379440,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 217162
     },
     {
       "epoch": 1.859375,
-      "grad_norm": 0.0017771078273653984,
+      "grad_norm": 0.0004491541185416281,
       "learning_rate": 1.1275748307758873e-05,
-      "loss": 1.803937993827276e-05,
+      "loss": 1.4460356396739371e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 476,
       "tokens/total": 14409840,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.88,
       "tokens/trainable": 217638
     },
     {
       "epoch": 1.86328125,
-      "grad_norm": 0.00044826362864114344,
+      "grad_norm": 0.0024057701230049133,
       "learning_rate": 1.1208026883231147e-05,
-      "loss": 9.577534910931718e-06,
+      "loss": 4.9342866986989975e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 477,
       "tokens/total": 14440256,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 218053
     },
     {
       "epoch": 1.8671875,
-      "grad_norm": 0.0009560272446833551,
+      "grad_norm": 0.0008485029684379697,
       "learning_rate": 1.1142127821456433e-05,
-      "loss": 1.0945323083433323e-05,
+      "loss": 2.05296601052396e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 478,
       "tokens/total": 14470704,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 218503
     },
     {
       "epoch": 1.87109375,
-      "grad_norm": 0.2086740881204605,
+      "grad_norm": 0.011945155449211597,
       "learning_rate": 1.1078053864763674e-05,
-      "loss": 0.0012299084337428212,
+      "loss": 8.75981932040304e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00123,
+      "ppl": 1.00009,
       "step": 479,
       "tokens/total": 14501072,
       "tokens/train_per_sec_per_gpu": 36.19,
@@ -6717,13 +6717,13 @@
     },
     {
       "epoch": 1.875,
-      "grad_norm": 0.019221138209104538,
+      "grad_norm": 0.004776590969413519,
       "learning_rate": 1.1015807679531756e-05,
-      "loss": 0.00011615331459324807,
+      "loss": 7.07923318259418e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00012,
+      "ppl": 1.00007,
       "step": 480,
       "tokens/total": 14531200,
       "tokens/train_per_sec_per_gpu": 35.51,
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_model.safetensors
index ea3cea0a354d5b9ee2785a0c48f7ebf1c6877a9d..39b53dcd92648efb307c35556c5bb95fa36cebce 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16cc65acfba87debf950e49fc05f5dce80ea16745d793c826ed96faac31a47a1
+oid sha256:daeafd5d686b4b39c5b3dd656eca36514ea93ef225c5afab298c74bbfab4b632
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/optimizer.pt
index 5528e895609544bbcb34ba418d72923cff27a19b..9244c05de713bc8bc639710777cf556fe75699b1 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe1efd539e57db6375dbfc409cf1e2a434b05b88c1f3bd7e58c16a8d2075a365
+oid sha256:05bb512f0593d6339b6dd5bd7fd23b00fbb5ccbe33fcfaa642269827ce75bdc6
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/trainer_state.json
index 2499b3b7e698733d7ad57d04aa419bb5db763c37..dd12f510a42892888ddf9cc0585226435f8e071d 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-512/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,1105 +4715,1105 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.370597779750824,
+      "grad_norm": 0.03195308893918991,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 0.007550997193902731,
+      "loss": 0.0002666166110429913,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00758,
+      "ppl": 1.00027,
       "step": 353,
       "tokens/total": 10684176,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 161420
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0031376827973872423,
+      "grad_norm": 0.07012991607189178,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 2.3453332687495276e-05,
+      "loss": 0.000741704716347158,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00074,
       "step": 354,
       "tokens/total": 10714352,
-      "tokens/train_per_sec_per_gpu": 34.02,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 161868
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.0021399864926934242,
+      "grad_norm": 0.0005822654929943383,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 2.5832894607447088e-05,
+      "loss": 1.0378402294008993e-05,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 355,
       "tokens/total": 10744464,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 162318
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0028148347046226263,
+      "grad_norm": 0.012435230426490307,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 3.7286932638380677e-05,
+      "loss": 7.89838086348027e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00008,
       "step": 356,
       "tokens/total": 10774976,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.54,
       "tokens/trainable": 162789
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.1642119586467743,
+      "grad_norm": 0.0045598647557199,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 0.0027845175936818123,
+      "loss": 2.48450869548833e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00279,
+      "ppl": 1.00002,
       "step": 357,
       "tokens/total": 10805600,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 163254
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.066535085439682,
+      "grad_norm": 0.3044262230396271,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.0002488417667336762,
+      "loss": 0.002298796083778143,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00025,
+      "ppl": 1.0023,
       "step": 358,
       "tokens/total": 10835920,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.62,
       "tokens/trainable": 163715
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.006617655046284199,
+      "grad_norm": 0.002458421979099512,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 7.841112528694794e-05,
+      "loss": 1.9634167983895168e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 359,
       "tokens/total": 10866048,
-      "tokens/train_per_sec_per_gpu": 34.4,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 164197
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.010409035719931126,
+      "grad_norm": 0.0024773837067186832,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010489403939573094,
+      "loss": 2.744927041931078e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 360,
       "tokens/total": 10896288,
-      "tokens/train_per_sec_per_gpu": 31.6,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 164661
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.002426267135888338,
+      "grad_norm": 0.00026717092259787023,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 4.0343060391023755e-05,
+      "loss": 6.422977094189264e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 361,
       "tokens/total": 10926816,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 165111
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.015666797757148743,
+      "grad_norm": 0.0008353493758477271,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 0.00014991794887464494,
+      "loss": 1.3549893083109055e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 362,
       "tokens/total": 10954928,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 165552
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.2695651650428772,
+      "grad_norm": 0.006071125157177448,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0014983330620452762,
+      "loss": 6.428411870729178e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.0015,
+      "ppl": 1.00006,
       "step": 363,
       "tokens/total": 10985328,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 166026
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.03586212173104286,
+      "grad_norm": 0.25424960255622864,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.00030509717180393636,
+      "loss": 0.004243595991283655,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00031,
+      "ppl": 1.00425,
       "step": 364,
       "tokens/total": 11015664,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 166486
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.0019308789633214474,
+      "grad_norm": 0.0005575277027674019,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 2.4129443772835657e-05,
+      "loss": 8.403902938880492e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 365,
       "tokens/total": 11046256,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 166959
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005940837785601616,
+      "grad_norm": 0.0014335239538922906,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 2.8238933737156913e-05,
+      "loss": 1.4738036043127067e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 366,
       "tokens/total": 11076288,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 167394
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.002125627128407359,
+      "grad_norm": 0.0010152696631848812,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 1.602789416210726e-05,
+      "loss": 1.3278609912958927e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 367,
       "tokens/total": 11106416,
-      "tokens/train_per_sec_per_gpu": 28.5,
+      "tokens/train_per_sec_per_gpu": 28.58,
       "tokens/trainable": 167832
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.2510211765766144,
+      "grad_norm": 0.11941836029291153,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 0.00010395953722763807,
+      "loss": 0.0006476733833551407,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00065,
       "step": 368,
       "tokens/total": 11136848,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 31.02,
       "tokens/trainable": 168242
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.001904280623421073,
+      "grad_norm": 0.006877629552036524,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 2.6418363631819375e-05,
+      "loss": 7.449048280250281e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 369,
       "tokens/total": 11167376,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 168715
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.0028152521699666977,
+      "grad_norm": 0.0007151139434427023,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 2.9974533390486613e-05,
+      "loss": 1.1437590728746727e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 370,
       "tokens/total": 11197776,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.86,
       "tokens/trainable": 169147
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.0010822078911587596,
+      "grad_norm": 0.009094453416764736,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 1.34217716549756e-05,
+      "loss": 5.9242345741949975e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11227840,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 169628
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.005367911420762539,
+      "grad_norm": 0.000247267191298306,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 6.65646803099662e-05,
+      "loss": 7.548428584414069e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 372,
       "tokens/total": 11258176,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 170085
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.047570567578077316,
+      "grad_norm": 0.0003029497747775167,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.0003471036907285452,
+      "loss": 5.4013939916330855e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00035,
+      "ppl": 1.00001,
       "step": 373,
       "tokens/total": 11288336,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 170584
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.040886107832193375,
+      "grad_norm": 0.001942179980687797,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.0002880148240365088,
+      "loss": 2.380511250521522e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00029,
+      "ppl": 1.00002,
       "step": 374,
       "tokens/total": 11318848,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 171058
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.0031216361094266176,
+      "grad_norm": 0.0019308892078697681,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 2.690855944820214e-05,
+      "loss": 1.5233906196954194e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 375,
       "tokens/total": 11349168,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 171518
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.012176645919680595,
+      "grad_norm": 0.6044301986694336,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 8.402287494391203e-05,
+      "loss": 0.006350134499371052,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00008,
+      "ppl": 1.00637,
       "step": 376,
       "tokens/total": 11379328,
-      "tokens/train_per_sec_per_gpu": 40.48,
+      "tokens/train_per_sec_per_gpu": 40.62,
       "tokens/trainable": 172007
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.24136756360530853,
+      "grad_norm": 0.25446587800979614,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 0.0013303000014275312,
+      "loss": 0.0018724403344094753,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00133,
+      "ppl": 1.00187,
       "step": 377,
       "tokens/total": 11409568,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 172468
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.021714258939027786,
+      "grad_norm": 0.0019207323202863336,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 0.00020160498388577253,
+      "loss": 3.049923907383345e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 378,
       "tokens/total": 11439968,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 172923
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.35412073135375977,
+      "grad_norm": 0.33491548895835876,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 0.015761105343699455,
+      "loss": 0.02673855796456337,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01589,
+      "ppl": 1.0271,
       "step": 379,
       "tokens/total": 11470496,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 173370
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005766173824667931,
+      "grad_norm": 0.006177723873406649,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 2.5790239305933937e-05,
+      "loss": 5.195035191718489e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 380,
       "tokens/total": 11500720,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 173837
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.005898744333535433,
+      "grad_norm": 0.06973031163215637,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 5.821501690661535e-05,
+      "loss": 0.0006310560274869204,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00006,
+      "ppl": 1.00063,
       "step": 381,
       "tokens/total": 11531280,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 174294
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005451703444123268,
+      "grad_norm": 0.0039021887350827456,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 4.0181505028158426e-05,
+      "loss": 4.849781544180587e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 382,
       "tokens/total": 11561264,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.84,
       "tokens/trainable": 174701
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0038302047178149223,
+      "grad_norm": 0.0007975143962539732,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 3.403786467970349e-05,
+      "loss": 2.3489263185183518e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 383,
       "tokens/total": 11591616,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 175131
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.004227485507726669,
+      "grad_norm": 0.012541128322482109,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 3.562243364285678e-05,
+      "loss": 0.00011495217040646821,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00011,
       "step": 384,
       "tokens/total": 11622000,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 175586
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.013361346907913685,
+      "grad_norm": 0.004752015229314566,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0001464220113120973,
+      "loss": 6.502695032395422e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00007,
       "step": 385,
       "tokens/total": 11652352,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176026
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.008831475861370564,
+      "grad_norm": 0.014192809350788593,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 8.2662510976661e-05,
+      "loss": 8.075163350440562e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00008,
       "step": 386,
       "tokens/total": 11682736,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.91,
       "tokens/trainable": 176512
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.009311008267104626,
+      "grad_norm": 0.007288333959877491,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 0.0001226613821927458,
+      "loss": 8.845872798701748e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00009,
       "step": 387,
       "tokens/total": 11713136,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.45,
       "tokens/trainable": 176943
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.0017816838808357716,
+      "grad_norm": 0.10162956267595291,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 2.3221660740091465e-05,
+      "loss": 0.0007253064541146159,
       "memory/device_reserved (GiB)": 35.23,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00073,
       "step": 388,
       "tokens/total": 11743808,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 177370
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.023602057248353958,
+      "grad_norm": 0.005729300435632467,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 0.00012638044427148998,
+      "loss": 8.652975520817563e-05,
       "memory/device_reserved (GiB)": 35.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00013,
+      "ppl": 1.00009,
       "step": 389,
       "tokens/total": 11774144,
-      "tokens/train_per_sec_per_gpu": 35.14,
+      "tokens/train_per_sec_per_gpu": 35.2,
       "tokens/trainable": 177851
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.04383797571063042,
+      "grad_norm": 0.0715414360165596,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 0.0004640861588995904,
+      "loss": 0.0007950748549774289,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00046,
+      "ppl": 1.0008,
       "step": 390,
       "tokens/total": 11804272,
-      "tokens/train_per_sec_per_gpu": 34.68,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 178310
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.061932601034641266,
+      "grad_norm": 0.009775097481906414,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.00033243370126001537,
+      "loss": 0.00010615254723234102,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00033,
+      "ppl": 1.00011,
       "step": 391,
       "tokens/total": 11832432,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 178736
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.04983547702431679,
+      "grad_norm": 0.0015552763361483812,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 0.000599355495069176,
+      "loss": 2.954876617877744e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0006,
+      "ppl": 1.00003,
       "step": 392,
       "tokens/total": 11862960,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 179233
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 0.012036106549203396,
+      "grad_norm": 0.026106838136911392,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0001237639953615144,
+      "loss": 0.00021109850786160678,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00021,
       "step": 393,
       "tokens/total": 11893232,
-      "tokens/train_per_sec_per_gpu": 38.49,
+      "tokens/train_per_sec_per_gpu": 38.57,
       "tokens/trainable": 179730
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.008992817252874374,
+      "grad_norm": 0.020347680896520615,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 0.00010385089262854308,
+      "loss": 0.0002601295418571681,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0001,
+      "ppl": 1.00026,
       "step": 394,
       "tokens/total": 11923664,
-      "tokens/train_per_sec_per_gpu": 37.81,
+      "tokens/train_per_sec_per_gpu": 37.87,
       "tokens/trainable": 180227
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.24631057679653168,
+      "grad_norm": 0.005419979803264141,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.0031685903668403625,
+      "loss": 4.929217175231315e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00317,
+      "ppl": 1.00005,
       "step": 395,
       "tokens/total": 11954032,
-      "tokens/train_per_sec_per_gpu": 32.85,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 180687
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.06871633976697922,
+      "grad_norm": 0.0031002764590084553,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 0.0005168755305930972,
+      "loss": 6.144218787085265e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00052,
+      "ppl": 1.00006,
       "step": 396,
       "tokens/total": 11984512,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 181141
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.011674679815769196,
+      "grad_norm": 0.006748533807694912,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 0.00017496946384198964,
+      "loss": 9.171784040518105e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 397,
       "tokens/total": 12014880,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 36.02,
       "tokens/trainable": 181648
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.14610664546489716,
+      "grad_norm": 0.08927815407514572,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.0014660547021776438,
+      "loss": 0.001000746968202293,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00147,
+      "ppl": 1.001,
       "step": 398,
       "tokens/total": 12044928,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 182109
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.016359565779566765,
+      "grad_norm": 0.004651801194995642,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 7.928608101792634e-05,
+      "loss": 5.3439554903889075e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00008,
+      "ppl": 1.00005,
       "step": 399,
       "tokens/total": 12075344,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 182559
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0010484450031071901,
+      "grad_norm": 0.0017514342907816172,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 1.526270352769643e-05,
+      "loss": 4.570486271404661e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00005,
       "step": 400,
       "tokens/total": 12105616,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 182985
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.0005900752730667591,
+      "grad_norm": 0.01574699766933918,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 1.2166316082584672e-05,
+      "loss": 0.0001242965809069574,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 401,
       "tokens/total": 12136080,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 183440
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.013194791041314602,
+      "grad_norm": 0.007152364123612642,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 0.0001970212033484131,
+      "loss": 0.0001222842838615179,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0002,
+      "ppl": 1.00012,
       "step": 402,
       "tokens/total": 12166240,
-      "tokens/train_per_sec_per_gpu": 34.52,
+      "tokens/train_per_sec_per_gpu": 34.51,
       "tokens/trainable": 183917
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.02566305734217167,
+      "grad_norm": 0.0015342400874942541,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.00022389904188457876,
+      "loss": 3.621872019721195e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00022,
+      "ppl": 1.00004,
       "step": 403,
       "tokens/total": 12196544,
-      "tokens/train_per_sec_per_gpu": 30.39,
+      "tokens/train_per_sec_per_gpu": 30.36,
       "tokens/trainable": 184348
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.008275284431874752,
+      "grad_norm": 0.0035214691888540983,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 8.35009923321195e-05,
+      "loss": 6.367819150909781e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00006,
       "step": 404,
       "tokens/total": 12226784,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 184786
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.038177311420440674,
+      "grad_norm": 0.0007211520569398999,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 0.00018527230713516474,
+      "loss": 2.0325338482507505e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00019,
+      "ppl": 1.00002,
       "step": 405,
       "tokens/total": 12257264,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 185249
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.007672510575503111,
+      "grad_norm": 0.0014804014936089516,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 0.00011439670925028622,
+      "loss": 3.379736153874546e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00003,
       "step": 406,
       "tokens/total": 12287584,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 185680
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.003159885760396719,
+      "grad_norm": 0.004397980403155088,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 4.649449692806229e-05,
+      "loss": 5.1796458137687296e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00005,
       "step": 407,
       "tokens/total": 12317984,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 186100
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.32744285464286804,
+      "grad_norm": 0.0026473007164895535,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 0.002739987801760435,
+      "loss": 3.441090666456148e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00274,
+      "ppl": 1.00003,
       "step": 408,
       "tokens/total": 12348144,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.84,
       "tokens/trainable": 186565
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.0006675662589259446,
+      "grad_norm": 0.0289909727871418,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 1.2970660463906825e-05,
+      "loss": 0.00021602815832011402,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00022,
       "step": 409,
       "tokens/total": 12378544,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 187017
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.003976705949753523,
+      "grad_norm": 0.007092812564224005,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 3.857718547806144e-05,
+      "loss": 9.646185935707763e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.0001,
       "step": 410,
       "tokens/total": 12408784,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 187468
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.0024577646981924772,
+      "grad_norm": 0.02710823155939579,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 2.431379834888503e-05,
+      "loss": 0.0003458422143012285,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00035,
       "step": 411,
       "tokens/total": 12438832,
-      "tokens/train_per_sec_per_gpu": 37.61,
+      "tokens/train_per_sec_per_gpu": 37.57,
       "tokens/trainable": 187961
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003754909848794341,
+      "grad_norm": 0.29485711455345154,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 4.119630830246024e-05,
+      "loss": 0.0032923028338700533,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.0033,
       "step": 412,
       "tokens/total": 12469296,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 188447
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.3583323657512665,
+      "grad_norm": 0.22881586849689484,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.004735157359391451,
+      "loss": 0.0035809341352432966,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00475,
+      "ppl": 1.00359,
       "step": 413,
       "tokens/total": 12499392,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 188903
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.049485232681035995,
+      "grad_norm": 0.0026347371749579906,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 0.00019404500199016184,
+      "loss": 6.0475373174995184e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.00006,
       "step": 414,
       "tokens/total": 12529952,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 189398
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0030447980388998985,
+      "grad_norm": 0.0017223359318450093,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 2.0038012735312805e-05,
+      "loss": 4.137849100516178e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 415,
       "tokens/total": 12560240,
       "tokens/train_per_sec_per_gpu": 34.16,
@@ -5821,293 +5821,293 @@
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.25415608286857605,
+      "grad_norm": 0.011837545782327652,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 0.0016201161779463291,
+      "loss": 0.00011106643069069833,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00162,
+      "ppl": 1.00011,
       "step": 416,
       "tokens/total": 12590736,
-      "tokens/train_per_sec_per_gpu": 32.64,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 190306
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.006711115129292011,
+      "grad_norm": 0.01777251996099949,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 4.321872620494105e-05,
+      "loss": 0.0001219596597366035,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 417,
       "tokens/total": 12621168,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 190746
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.004931971430778503,
+      "grad_norm": 0.003218573285266757,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 4.740363510791212e-05,
+      "loss": 5.22282425663434e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00005,
       "step": 418,
       "tokens/total": 12651456,
-      "tokens/train_per_sec_per_gpu": 31.07,
+      "tokens/train_per_sec_per_gpu": 31.15,
       "tokens/trainable": 191156
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.007677357643842697,
+      "grad_norm": 0.013744015246629715,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 7.545409607701004e-05,
+      "loss": 0.0001364837517030537,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00014,
       "step": 419,
       "tokens/total": 12681760,
-      "tokens/train_per_sec_per_gpu": 29.47,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 191573
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.015445503406226635,
+      "grad_norm": 0.25013473629951477,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 0.0001045453900587745,
+      "loss": 0.002094803610816598,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0001,
+      "ppl": 1.0021,
       "step": 420,
       "tokens/total": 12712144,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.16882061958312988,
+      "grad_norm": 0.14290185272693634,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 0.0024146074429154396,
+      "loss": 0.001426510512828827,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00242,
+      "ppl": 1.00143,
       "step": 421,
       "tokens/total": 12742400,
-      "tokens/train_per_sec_per_gpu": 32.82,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 192453
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.002461223630234599,
+      "grad_norm": 0.0009390547638759017,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 3.336594454594888e-05,
+      "loss": 1.822916055971291e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12772288,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 192886
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.0005388484569266438,
+      "grad_norm": 0.004544281866401434,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 8.696397344465367e-06,
+      "loss": 6.211431173142046e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 423,
       "tokens/total": 12802880,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.87,
       "tokens/trainable": 193323
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.005009706597775221,
+      "grad_norm": 0.039859045296907425,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 4.8607362259645015e-05,
+      "loss": 0.0005113891093060374,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00051,
       "step": 424,
       "tokens/total": 12833456,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 193753
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.001982118235900998,
+      "grad_norm": 0.00408409908413887,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 3.262238169554621e-05,
+      "loss": 7.249199552461505e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 425,
       "tokens/total": 12863952,
-      "tokens/train_per_sec_per_gpu": 36.33,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 194240
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.05305991321802139,
+      "grad_norm": 0.13116440176963806,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 0.0005229170201346278,
+      "loss": 0.0015061571029946208,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00052,
+      "ppl": 1.00151,
       "step": 426,
       "tokens/total": 12894352,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 194680
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.001265498693101108,
+      "grad_norm": 0.010440024547278881,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 1.5427456673933193e-05,
+      "loss": 8.445358253084123e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00008,
       "step": 427,
       "tokens/total": 12924688,
-      "tokens/train_per_sec_per_gpu": 38.73,
+      "tokens/train_per_sec_per_gpu": 38.77,
       "tokens/trainable": 195186
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.0016263640718534589,
+      "grad_norm": 0.007594608701765537,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 2.7778178264270537e-05,
+      "loss": 9.946282807504758e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 428,
       "tokens/total": 12955072,
-      "tokens/train_per_sec_per_gpu": 36.51,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 195693
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.013330154120922089,
+      "grad_norm": 0.006465950049459934,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 6.972272240091115e-05,
+      "loss": 0.00013557568308897316,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 429,
       "tokens/total": 12985328,
-      "tokens/train_per_sec_per_gpu": 36.93,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 196130
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.0034615101758390665,
+      "grad_norm": 0.00230405549518764,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 4.500148497754708e-05,
+      "loss": 3.7114587030373514e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 430,
       "tokens/total": 13015600,
-      "tokens/train_per_sec_per_gpu": 29.58,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 196530
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.03424045443534851,
+      "grad_norm": 0.0041900877840816975,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0002949235204141587,
+      "loss": 6.116987788118422e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00029,
+      "ppl": 1.00006,
       "step": 431,
       "tokens/total": 13045728,
-      "tokens/train_per_sec_per_gpu": 29.88,
+      "tokens/train_per_sec_per_gpu": 29.94,
       "tokens/trainable": 196974
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0064499350264668465,
+      "grad_norm": 0.02951621450483799,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 7.787663344061002e-05,
+      "loss": 0.0004236411186866462,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00008,
+      "ppl": 1.00042,
       "step": 432,
       "tokens/total": 13075920,
-      "tokens/train_per_sec_per_gpu": 35.25,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 197454
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.00021908426424488425,
+      "grad_norm": 0.0003202867810614407,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 6.603059773624409e-06,
+      "loss": 1.1190046279807575e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 433,
       "tokens/total": 13106128,
-      "tokens/train_per_sec_per_gpu": 31.4,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 197902
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.0002612156968098134,
+      "grad_norm": 0.0006868013297207654,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 6.425582796509843e-06,
+      "loss": 2.1185776859056205e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 434,
       "tokens/total": 13136368,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.06,
       "tokens/trainable": 198374
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.001639618189074099,
+      "grad_norm": 0.0025407401844859123,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 2.5485322112217546e-05,
+      "loss": 4.928320413455367e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 435,
       "tokens/total": 13166704,
-      "tokens/train_per_sec_per_gpu": 37.54,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 198846
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.01887959986925125,
+      "grad_norm": 0.0013901089550927281,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0002141120348824188,
+      "loss": 3.577578900149092e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00021,
+      "ppl": 1.00004,
       "step": 436,
       "tokens/total": 13197344,
       "tokens/train_per_sec_per_gpu": 37.67,
@@ -6115,41 +6115,41 @@
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.0012294527841731906,
+      "grad_norm": 0.003646081080660224,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.316079735464882e-05,
+      "loss": 6.294051127042621e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 437,
       "tokens/total": 13227664,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 199777
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.005585651844739914,
+      "grad_norm": 0.030430182814598083,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 4.5935248635942116e-05,
+      "loss": 0.0001770013477653265,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00018,
       "step": 438,
       "tokens/total": 13257632,
-      "tokens/train_per_sec_per_gpu": 30.82,
+      "tokens/train_per_sec_per_gpu": 30.93,
       "tokens/trainable": 200194
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.001553745474666357,
+      "grad_norm": 0.0018230377463623881,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 2.2851421817904338e-05,
+      "loss": 3.7613608583342284e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13288240,
       "tokens/train_per_sec_per_gpu": 33.88,
@@ -6157,559 +6157,559 @@
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0029791900888085365,
+      "grad_norm": 0.0029376039747148752,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 4.093274037586525e-05,
+      "loss": 4.779352093464695e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 440,
       "tokens/total": 13318992,
-      "tokens/train_per_sec_per_gpu": 33.79,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 201121
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0017392379231750965,
+      "grad_norm": 0.0018373411148786545,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 7.29740804672474e-06,
+      "loss": 3.149824624415487e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 441,
       "tokens/total": 13349392,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.81,
       "tokens/trainable": 201598
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0009943305049091578,
+      "grad_norm": 0.0015428521437570453,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 1.7162077710963786e-05,
+      "loss": 2.0810390196857043e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 442,
       "tokens/total": 13379648,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 202068
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0011473585618659854,
+      "grad_norm": 0.0010770867811515927,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 2.3819740817998536e-05,
+      "loss": 2.303836117789615e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 443,
       "tokens/total": 13409936,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 202524
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.006799118127673864,
+      "grad_norm": 0.006372133269906044,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 6.556943117175251e-05,
+      "loss": 7.449327677022666e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
       "ppl": 1.00007,
       "step": 444,
       "tokens/total": 13440288,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 202996
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.0011645558988675475,
+      "grad_norm": 0.007094603031873703,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 1.8389995602774434e-05,
+      "loss": 4.211071063764393e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 445,
       "tokens/total": 13468880,
-      "tokens/train_per_sec_per_gpu": 39.14,
+      "tokens/train_per_sec_per_gpu": 39.19,
       "tokens/trainable": 203470
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.020891210064291954,
+      "grad_norm": 0.006931444630026817,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 0.00016399261949118227,
+      "loss": 8.66219779709354e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00016,
+      "ppl": 1.00009,
       "step": 446,
       "tokens/total": 13499184,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 203919
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.09615519642829895,
+      "grad_norm": 0.0041664596647024155,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 0.0010675137164071202,
+      "loss": 4.786982390214689e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00107,
+      "ppl": 1.00005,
       "step": 447,
       "tokens/total": 13529632,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 204376
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.09744399785995483,
+      "grad_norm": 0.042264848947525024,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.001196134602651,
+      "loss": 0.00027092613163404167,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0012,
+      "ppl": 1.00027,
       "step": 448,
       "tokens/total": 13560080,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 204821
     },
     {
       "epoch": 1.75390625,
-      "grad_norm": 0.0007810555398464203,
+      "grad_norm": 0.00043735766666941345,
       "learning_rate": 1.3780999708818058e-05,
-      "loss": 8.019956112548243e-06,
+      "loss": 1.2837479516747408e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
       "ppl": 1.00001,
       "step": 449,
       "tokens/total": 13590496,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 205271
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 0.034087613224983215,
+      "grad_norm": 0.0012977999867871404,
       "learning_rate": 1.3665385037857758e-05,
-      "loss": 0.00010244422446703538,
+      "loss": 2.1102820028318092e-05,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0001,
+      "ppl": 1.00002,
       "step": 450,
       "tokens/total": 13620960,
-      "tokens/train_per_sec_per_gpu": 30.95,
+      "tokens/train_per_sec_per_gpu": 30.92,
       "tokens/trainable": 205723
     },
     {
       "epoch": 1.76171875,
-      "grad_norm": 0.02930639684200287,
+      "grad_norm": 0.005675352178514004,
       "learning_rate": 1.3551490468947126e-05,
-      "loss": 0.00021926099725533277,
+      "loss": 7.465962698915973e-05,
       "memory/device_reserved (GiB)": 35.04,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00022,
+      "ppl": 1.00007,
       "step": 451,
       "tokens/total": 13651344,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.48,
       "tokens/trainable": 206169
     },
     {
       "epoch": 1.765625,
-      "grad_norm": 0.0109552051872015,
+      "grad_norm": 0.0013997588539496064,
       "learning_rate": 1.3439320741704075e-05,
-      "loss": 7.999646186362952e-05,
+      "loss": 2.723011130001396e-05,
       "memory/device_reserved (GiB)": 35.04,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 452,
       "tokens/total": 13681792,
-      "tokens/train_per_sec_per_gpu": 32.74,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 206592
     },
     {
       "epoch": 1.76953125,
-      "grad_norm": 0.004453443922102451,
+      "grad_norm": 0.012491011992096901,
       "learning_rate": 1.3328880523968808e-05,
-      "loss": 3.79011980840005e-05,
+      "loss": 9.182744543068111e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00009,
       "step": 453,
       "tokens/total": 13712256,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.07,
       "tokens/trainable": 207055
     },
     {
       "epoch": 1.7734375,
-      "grad_norm": 0.0005650459788739681,
+      "grad_norm": 0.000924609019421041,
       "learning_rate": 1.3220174411609587e-05,
-      "loss": 1.2244867320987396e-05,
+      "loss": 2.1621295672957785e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 454,
       "tokens/total": 13742128,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.61,
       "tokens/trainable": 207494
     },
     {
       "epoch": 1.77734375,
-      "grad_norm": 0.0007141040405258536,
+      "grad_norm": 0.000411411514505744,
       "learning_rate": 1.3113206928331471e-05,
-      "loss": 1.3571594536188059e-05,
+      "loss": 1.2427874025888741e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 455,
       "tokens/total": 13772160,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.03,
       "tokens/trainable": 207956
     },
     {
       "epoch": 1.78125,
-      "grad_norm": 0.0008097058744169772,
+      "grad_norm": 0.0015909038484096527,
       "learning_rate": 1.300798252548806e-05,
-      "loss": 1.1659047231660224e-05,
+      "loss": 3.0332066671689972e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 456,
       "tokens/total": 13802448,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.42,
       "tokens/trainable": 208408
     },
     {
       "epoch": 1.78515625,
-      "grad_norm": 0.0012649539858102798,
+      "grad_norm": 0.0030019236728549004,
       "learning_rate": 1.2904505581896265e-05,
-      "loss": 9.200517524732277e-06,
+      "loss": 3.378765541128814e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 457,
       "tokens/total": 13832880,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 208889
     },
     {
       "epoch": 1.7890625,
-      "grad_norm": 0.014978500083088875,
+      "grad_norm": 0.0016427476657554507,
       "learning_rate": 1.2802780403654082e-05,
-      "loss": 0.0001208957692142576,
+      "loss": 3.3130341762444004e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00012,
+      "ppl": 1.00003,
       "step": 458,
       "tokens/total": 13862880,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 209336
     },
     {
       "epoch": 1.79296875,
-      "grad_norm": 0.004875013139098883,
+      "grad_norm": 0.0006426559993997216,
       "learning_rate": 1.2702811223961408e-05,
-      "loss": 3.241455851821229e-05,
+      "loss": 1.784306368790567e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 459,
       "tokens/total": 13893440,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.21,
       "tokens/trainable": 209797
     },
     {
       "epoch": 1.796875,
-      "grad_norm": 0.0009627968538552523,
+      "grad_norm": 0.0016685863956809044,
       "learning_rate": 1.2604602202943861e-05,
-      "loss": 1.980438355531078e-05,
+      "loss": 3.2843898225110024e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 460,
       "tokens/total": 13923824,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.03,
       "tokens/trainable": 210240
     },
     {
       "epoch": 1.80078125,
-      "grad_norm": 0.019912317395210266,
+      "grad_norm": 0.03454679995775223,
       "learning_rate": 1.2508157427479686e-05,
-      "loss": 0.00010681153071345761,
+      "loss": 7.200831169029698e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00011,
+      "ppl": 1.00007,
       "step": 461,
       "tokens/total": 13954208,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 34.9,
       "tokens/trainable": 210702
     },
     {
       "epoch": 1.8046875,
-      "grad_norm": 0.0005370390135794878,
+      "grad_norm": 0.0021488473284989595,
       "learning_rate": 1.2413480911029655e-05,
-      "loss": 1.1081473530794028e-05,
+      "loss": 3.525309875840321e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 462,
       "tokens/total": 13984784,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.44,
       "tokens/trainable": 211140
     },
     {
       "epoch": 1.80859375,
-      "grad_norm": 0.000815370527561754,
+      "grad_norm": 0.0007275328389368951,
       "learning_rate": 1.2320576593470082e-05,
-      "loss": 1.5510300727328286e-05,
+      "loss": 1.328821963397786e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 463,
       "tokens/total": 14014896,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.54,
       "tokens/trainable": 211575
     },
     {
       "epoch": 1.8125,
-      "grad_norm": 0.02875823900103569,
+      "grad_norm": 0.03173735365271568,
       "learning_rate": 1.2229448340928828e-05,
-      "loss": 0.00013141569797880948,
+      "loss": 0.00014542456483468413,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00013,
+      "ppl": 1.00015,
       "step": 464,
       "tokens/total": 14045312,
-      "tokens/train_per_sec_per_gpu": 39.03,
+      "tokens/train_per_sec_per_gpu": 38.86,
       "tokens/trainable": 212083
     },
     {
       "epoch": 1.81640625,
-      "grad_norm": 0.04558868706226349,
+      "grad_norm": 0.0005439479718916118,
       "learning_rate": 1.2140099945624458e-05,
-      "loss": 0.0002969688503071666,
+      "loss": 1.4493984053842723e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0003,
+      "ppl": 1.00001,
       "step": 465,
       "tokens/total": 14075840,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 212558
     },
     {
       "epoch": 1.8203125,
-      "grad_norm": 0.0004122587270103395,
+      "grad_norm": 0.0015477532288059592,
       "learning_rate": 1.205253512570841e-05,
-      "loss": 9.937594768416602e-06,
+      "loss": 2.787737685139291e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 466,
       "tokens/total": 14106016,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.73,
       "tokens/trainable": 213014
     },
     {
       "epoch": 1.82421875,
-      "grad_norm": 0.0010108116548508406,
+      "grad_norm": 0.0007428264361806214,
       "learning_rate": 1.1966757525110255e-05,
-      "loss": 1.5579567843815312e-05,
+      "loss": 2.1110219677211717e-05,
       "memory/device_reserved (GiB)": 37.99,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
       "ppl": 1.00002,
       "step": 467,
       "tokens/total": 14136512,
-      "tokens/train_per_sec_per_gpu": 34.03,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 213465
     },
     {
       "epoch": 1.828125,
-      "grad_norm": 0.0043095871806144714,
+      "grad_norm": 0.002182955853641033,
       "learning_rate": 1.1882770713386095e-05,
-      "loss": 3.2609641493763775e-05,
+      "loss": 4.059498314745724e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00004,
       "step": 468,
       "tokens/total": 14166832,
-      "tokens/train_per_sec_per_gpu": 36.62,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 213940
     },
     {
       "epoch": 1.83203125,
-      "grad_norm": 0.005317045841366053,
+      "grad_norm": 0.004772448446601629,
       "learning_rate": 1.180057818556998e-05,
-      "loss": 2.144884638255462e-05,
+      "loss": 3.664854375529103e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 469,
       "tokens/total": 14197360,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 214415
     },
     {
       "epoch": 1.8359375,
-      "grad_norm": 0.0003264908737037331,
+      "grad_norm": 0.00038694078102707863,
       "learning_rate": 1.1720183362028494e-05,
-      "loss": 5.6935500651889015e-06,
+      "loss": 1.1124819138785824e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 470,
       "tokens/total": 14227600,
-      "tokens/train_per_sec_per_gpu": 34.26,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 214880
     },
     {
       "epoch": 1.83984375,
-      "grad_norm": 0.013723314739763737,
+      "grad_norm": 0.0008439846569672227,
       "learning_rate": 1.1641589588318387e-05,
-      "loss": 0.00010619591921567917,
+      "loss": 2.0906983991153538e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00011,
+      "ppl": 1.00002,
       "step": 471,
       "tokens/total": 14257888,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 215334
     },
     {
       "epoch": 1.84375,
-      "grad_norm": 0.00020912640320602804,
+      "grad_norm": 0.0005926606827415526,
       "learning_rate": 1.1564800135047418e-05,
-      "loss": 4.586940121953376e-06,
+      "loss": 1.178990351036191e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 472,
       "tokens/total": 14288208,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 215822
     },
     {
       "epoch": 1.84765625,
-      "grad_norm": 0.10667876899242401,
+      "grad_norm": 0.003897819435223937,
       "learning_rate": 1.148981819773816e-05,
-      "loss": 0.0006915333215147257,
+      "loss": 5.020621756557375e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.00005,
       "step": 473,
       "tokens/total": 14318448,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 30.82,
       "tokens/trainable": 216239
     },
     {
       "epoch": 1.8515625,
-      "grad_norm": 0.013282700441777706,
+      "grad_norm": 0.0013807902578264475,
       "learning_rate": 1.1416646896695086e-05,
-      "loss": 6.861680594738573e-05,
+      "loss": 2.2795318727730773e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 474,
       "tokens/total": 14348912,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.35,
       "tokens/trainable": 216691
     },
     {
       "epoch": 1.85546875,
-      "grad_norm": 0.0074070473201572895,
+      "grad_norm": 0.00048289448022842407,
       "learning_rate": 1.1345289276874717e-05,
-      "loss": 8.334079757332802e-05,
+      "loss": 1.4630711120844353e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 475,
       "tokens/total": 14379440,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 217162
     },
     {
       "epoch": 1.859375,
-      "grad_norm": 0.0017771078273653984,
+      "grad_norm": 0.0004491541185416281,
       "learning_rate": 1.1275748307758873e-05,
-      "loss": 1.803937993827276e-05,
+      "loss": 1.4460356396739371e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 476,
       "tokens/total": 14409840,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.88,
       "tokens/trainable": 217638
     },
     {
       "epoch": 1.86328125,
-      "grad_norm": 0.00044826362864114344,
+      "grad_norm": 0.0024057701230049133,
       "learning_rate": 1.1208026883231147e-05,
-      "loss": 9.577534910931718e-06,
+      "loss": 4.9342866986989975e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 477,
       "tokens/total": 14440256,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 218053
     },
     {
       "epoch": 1.8671875,
-      "grad_norm": 0.0009560272446833551,
+      "grad_norm": 0.0008485029684379697,
       "learning_rate": 1.1142127821456433e-05,
-      "loss": 1.0945323083433323e-05,
+      "loss": 2.05296601052396e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 478,
       "tokens/total": 14470704,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 218503
     },
     {
       "epoch": 1.87109375,
-      "grad_norm": 0.2086740881204605,
+      "grad_norm": 0.011945155449211597,
       "learning_rate": 1.1078053864763674e-05,
-      "loss": 0.0012299084337428212,
+      "loss": 8.75981932040304e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00123,
+      "ppl": 1.00009,
       "step": 479,
       "tokens/total": 14501072,
       "tokens/train_per_sec_per_gpu": 36.19,
@@ -6717,13 +6717,13 @@
     },
     {
       "epoch": 1.875,
-      "grad_norm": 0.019221138209104538,
+      "grad_norm": 0.004776590969413519,
       "learning_rate": 1.1015807679531756e-05,
-      "loss": 0.00011615331459324807,
+      "loss": 7.07923318259418e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00012,
+      "ppl": 1.00007,
       "step": 480,
       "tokens/total": 14531200,
       "tokens/train_per_sec_per_gpu": 35.51,
@@ -6731,251 +6731,251 @@
     },
     {
       "epoch": 1.87890625,
-      "grad_norm": 0.002074754796922207,
+      "grad_norm": 0.0006404300802387297,
       "learning_rate": 1.0955391856078528e-05,
-      "loss": 2.3620896172360517e-05,
+      "loss": 1.6759000573074445e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
       "ppl": 1.00002,
       "step": 481,
       "tokens/total": 14561152,
-      "tokens/train_per_sec_per_gpu": 38.29,
+      "tokens/train_per_sec_per_gpu": 38.25,
       "tokens/trainable": 219886
     },
     {
       "epoch": 1.8828125,
-      "grad_norm": 0.011151358485221863,
+      "grad_norm": 0.005522869061678648,
       "learning_rate": 1.0896808908553007e-05,
-      "loss": 5.527659959625453e-05,
+      "loss": 2.3341217456618324e-05,
       "memory/device_reserved (GiB)": 37.77,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 482,
       "tokens/total": 14591264,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.52,
       "tokens/trainable": 220343
     },
     {
       "epoch": 1.88671875,
-      "grad_norm": 0.010913084261119366,
+      "grad_norm": 0.001015707035548985,
       "learning_rate": 1.0840061274830763e-05,
-      "loss": 6.711594323860481e-05,
+      "loss": 2.3611488359165378e-05,
       "memory/device_reserved (GiB)": 37.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 483,
       "tokens/total": 14621568,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 220825
     },
     {
       "epoch": 1.890625,
-      "grad_norm": 0.3236236274242401,
+      "grad_norm": 0.004034205339848995,
       "learning_rate": 1.0785151316412473e-05,
-      "loss": 0.002836203668266535,
+      "loss": 5.646366116707213e-05,
       "memory/device_reserved (GiB)": 37.77,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00284,
+      "ppl": 1.00006,
       "step": 484,
       "tokens/total": 14652064,
-      "tokens/train_per_sec_per_gpu": 32.52,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 221282
     },
     {
       "epoch": 1.89453125,
-      "grad_norm": 0.0002806074626278132,
+      "grad_norm": 0.008657694794237614,
       "learning_rate": 1.0732081318325639e-05,
-      "loss": 5.300881639414001e-06,
+      "loss": 5.676104410667904e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 485,
       "tokens/total": 14682608,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 221780
     },
     {
       "epoch": 1.8984375,
-      "grad_norm": 0.009111613035202026,
+      "grad_norm": 0.0036515207029879093,
       "learning_rate": 1.0680853489029501e-05,
-      "loss": 3.664881660370156e-05,
+      "loss": 3.9208807720569894e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00004,
       "step": 486,
       "tokens/total": 14713168,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 222210
     },
     {
       "epoch": 1.90234375,
-      "grad_norm": 0.04484618082642555,
+      "grad_norm": 0.005618095863610506,
       "learning_rate": 1.0631469960323152e-05,
-      "loss": 0.0002056290686596185,
+      "loss": 5.3539326472673565e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00005,
       "step": 487,
       "tokens/total": 14743536,
-      "tokens/train_per_sec_per_gpu": 29.33,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 222646
     },
     {
       "epoch": 1.90625,
-      "grad_norm": 0.2784590721130371,
+      "grad_norm": 0.07197980582714081,
       "learning_rate": 1.0583932787256783e-05,
-      "loss": 0.008498833514750004,
+      "loss": 0.0006614382145926356,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00854,
+      "ppl": 1.00066,
       "step": 488,
       "tokens/total": 14773728,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 223066
     },
     {
       "epoch": 1.91015625,
-      "grad_norm": 0.0006373917567543685,
+      "grad_norm": 0.001169998780824244,
       "learning_rate": 1.0538243948046206e-05,
-      "loss": 9.457155101699755e-06,
+      "loss": 1.62669166456908e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 489,
       "tokens/total": 14803744,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 223512
     },
     {
       "epoch": 1.9140625,
-      "grad_norm": 0.00038701840094290674,
+      "grad_norm": 0.0014546489110216498,
       "learning_rate": 1.0494405343990523e-05,
-      "loss": 6.92771118337987e-06,
+      "loss": 3.051830208278261e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 490,
       "tokens/total": 14832112,
-      "tokens/train_per_sec_per_gpu": 38.96,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 223956
     },
     {
       "epoch": 1.91796875,
-      "grad_norm": 0.0015320515958592296,
+      "grad_norm": 0.0007275546086020768,
       "learning_rate": 1.0452418799392985e-05,
-      "loss": 2.272017445648089e-05,
+      "loss": 1.4892379113007337e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 491,
       "tokens/total": 14860560,
-      "tokens/train_per_sec_per_gpu": 36.33,
+      "tokens/train_per_sec_per_gpu": 36.3,
       "tokens/trainable": 224416
     },
     {
       "epoch": 1.921875,
-      "grad_norm": 0.016094110906124115,
+      "grad_norm": 0.0007378064910881221,
       "learning_rate": 1.0412286061485102e-05,
-      "loss": 0.00011934755457332358,
+      "loss": 1.902497206174303e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 492,
       "tokens/total": 14890848,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.12,
       "tokens/trainable": 224883
     },
     {
       "epoch": 1.92578125,
-      "grad_norm": 0.0016222589183598757,
+      "grad_norm": 0.0006582152564078569,
       "learning_rate": 1.03740088003539e-05,
-      "loss": 1.4765413652639836e-05,
+      "loss": 1.498015808465425e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
       "ppl": 1.00001,
       "step": 493,
       "tokens/total": 14921008,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.42,
       "tokens/trainable": 225324
     },
     {
       "epoch": 1.9296875,
-      "grad_norm": 0.004442836623638868,
+      "grad_norm": 0.000849059084430337,
       "learning_rate": 1.0337588608872463e-05,
-      "loss": 3.44704239978455e-05,
+      "loss": 1.0077818842546549e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 494,
       "tokens/total": 14951360,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 225783
     },
     {
       "epoch": 1.93359375,
-      "grad_norm": 0.002437110524624586,
+      "grad_norm": 0.0029649219941347837,
       "learning_rate": 1.0303027002633622e-05,
-      "loss": 1.8051592633128166e-05,
+      "loss": 2.8222328182891943e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 495,
       "tokens/total": 14981552,
-      "tokens/train_per_sec_per_gpu": 36.46,
+      "tokens/train_per_sec_per_gpu": 36.48,
       "tokens/trainable": 226281
     },
     {
       "epoch": 1.9375,
-      "grad_norm": 0.0015804325230419636,
+      "grad_norm": 0.001169885159470141,
       "learning_rate": 1.0270325419886884e-05,
-      "loss": 1.691515171842184e-05,
+      "loss": 2.592003693280276e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 496,
       "tokens/total": 15011584,
-      "tokens/train_per_sec_per_gpu": 36.72,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 226735
     },
     {
       "epoch": 1.94140625,
-      "grad_norm": 0.0036208343226462603,
+      "grad_norm": 0.006588887423276901,
       "learning_rate": 1.0239485221478599e-05,
-      "loss": 4.592180994222872e-05,
+      "loss": 9.911794040817767e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00005,
+      "ppl": 1.0001,
       "step": 497,
       "tokens/total": 15042016,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.3,
       "tokens/trainable": 227198
     },
     {
       "epoch": 1.9453125,
-      "grad_norm": 0.0005918977549299598,
+      "grad_norm": 0.0013167925644665956,
       "learning_rate": 1.0210507690795292e-05,
-      "loss": 1.5305626220651902e-05,
+      "loss": 2.8513590223155916e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 498,
       "tokens/total": 15071984,
       "tokens/train_per_sec_per_gpu": 31.07,
@@ -6983,69 +6983,69 @@
     },
     {
       "epoch": 1.94921875,
-      "grad_norm": 0.0013667694292962551,
+      "grad_norm": 0.0010883790673688054,
       "learning_rate": 1.0183394033710305e-05,
-      "loss": 1.3490713172359392e-05,
+      "loss": 2.4797469450277276e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 499,
       "tokens/total": 15102192,
-      "tokens/train_per_sec_per_gpu": 32.15,
+      "tokens/train_per_sec_per_gpu": 32.17,
       "tokens/trainable": 228074
     },
     {
       "epoch": 1.953125,
-      "grad_norm": 0.0027930096257478,
+      "grad_norm": 0.010283510200679302,
       "learning_rate": 1.0158145378533583e-05,
-      "loss": 2.8409334845491685e-05,
+      "loss": 4.6255343477241695e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 500,
       "tokens/total": 15132560,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.59,
       "tokens/trainable": 228541
     },
     {
       "epoch": 1.95703125,
-      "grad_norm": 0.0028320043347775936,
+      "grad_norm": 0.0011426788987591863,
       "learning_rate": 1.0134762775964726e-05,
-      "loss": 3.1829928047955036e-05,
+      "loss": 1.940131733135786e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 501,
       "tokens/total": 15163152,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.78,
       "tokens/trainable": 229031
     },
     {
       "epoch": 1.9609375,
-      "grad_norm": 0.0030925904866307974,
+      "grad_norm": 0.0012880039867013693,
       "learning_rate": 1.0113247199049278e-05,
-      "loss": 2.238066190329846e-05,
+      "loss": 2.073413270409219e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 502,
       "tokens/total": 15193456,
-      "tokens/train_per_sec_per_gpu": 38.68,
+      "tokens/train_per_sec_per_gpu": 38.76,
       "tokens/trainable": 229512
     },
     {
       "epoch": 1.96484375,
-      "grad_norm": 0.06389517337083817,
+      "grad_norm": 0.022319281473755836,
       "learning_rate": 1.0093599543138205e-05,
-      "loss": 0.0004140711098443717,
+      "loss": 0.00016162173415068537,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00041,
+      "ppl": 1.00016,
       "step": 503,
       "tokens/total": 15223680,
       "tokens/train_per_sec_per_gpu": 34.26,
@@ -7053,128 +7053,128 @@
     },
     {
       "epoch": 1.96875,
-      "grad_norm": 0.003928070887923241,
+      "grad_norm": 0.00031853109248913825,
       "learning_rate": 1.0075820625850675e-05,
-      "loss": 4.3977754103252664e-05,
+      "loss": 9.707431672723033e-06,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 504,
       "tokens/total": 15254336,
-      "tokens/train_per_sec_per_gpu": 35.81,
+      "tokens/train_per_sec_per_gpu": 35.84,
       "tokens/trainable": 230447
     },
     {
       "epoch": 1.97265625,
-      "grad_norm": 0.0022790853399783373,
+      "grad_norm": 0.0005880650132894516,
       "learning_rate": 1.0059911187040013e-05,
-      "loss": 2.180483534175437e-05,
+      "loss": 1.7119673429988325e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
       "ppl": 1.00002,
       "step": 505,
       "tokens/total": 15284496,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.3,
       "tokens/trainable": 230879
     },
     {
       "epoch": 1.9765625,
-      "grad_norm": 0.0022161444649100304,
+      "grad_norm": 0.0011501131812110543,
       "learning_rate": 1.0045871888762893e-05,
-      "loss": 2.00994672923116e-05,
+      "loss": 1.6414911442552693e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
       "ppl": 1.00002,
       "step": 506,
       "tokens/total": 15314976,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 231348
     },
     {
       "epoch": 1.98046875,
-      "grad_norm": 0.0109304990619421,
+      "grad_norm": 0.0007460714550688863,
       "learning_rate": 1.003370331525184e-05,
-      "loss": 9.907589264912531e-05,
+      "loss": 2.1825573639944196e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00002,
       "step": 507,
       "tokens/total": 15345280,
-      "tokens/train_per_sec_per_gpu": 35.22,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 231819
     },
     {
       "epoch": 1.984375,
-      "grad_norm": 0.0008610524819232523,
+      "grad_norm": 0.003101626643911004,
       "learning_rate": 1.002340597289085e-05,
-      "loss": 1.4185266991262324e-05,
+      "loss": 4.144266858929768e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 508,
       "tokens/total": 15375424,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 232269
     },
     {
       "epoch": 1.98828125,
-      "grad_norm": 0.18468032777309418,
+      "grad_norm": 0.0010362571338191628,
       "learning_rate": 1.0014980290194387e-05,
-      "loss": 0.0011317847529426217,
+      "loss": 1.4500470570055768e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00113,
+      "ppl": 1.00001,
       "step": 509,
       "tokens/total": 15405872,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 232716
     },
     {
       "epoch": 1.9921875,
-      "grad_norm": 0.0012291419552639127,
+      "grad_norm": 0.0019138582283630967,
       "learning_rate": 1.0008426617789489e-05,
-      "loss": 1.4661756722489372e-05,
+      "loss": 2.179224611609243e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 510,
       "tokens/total": 15436128,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 233178
     },
     {
       "epoch": 1.99609375,
-      "grad_norm": 0.003051232313737273,
+      "grad_norm": 0.0017450153827667236,
       "learning_rate": 1.0003745228401215e-05,
-      "loss": 2.0963430870324373e-05,
+      "loss": 2.1991130779497325e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 511,
       "tokens/total": 15466464,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 233647
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.001929801655933261,
+      "grad_norm": 0.06926226615905762,
       "learning_rate": 1.0000936316841296e-05,
-      "loss": 1.9281407730886713e-05,
+      "loss": 0.0005933041102252901,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00002,
+      "ppl": 1.00059,
       "step": 512,
       "tokens/total": 15496368,
-      "tokens/train_per_sec_per_gpu": 29.42,
+      "tokens/train_per_sec_per_gpu": 29.4,
       "tokens/trainable": 234060
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_model.safetensors
index ec21044dd16514741ed50ac4fdbba207fd095d1f..b7aa46dc518d1cfa0e16d663660b9a0af7a46934 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:051830de7f1d163f4de9749f244e6981723b34f0246ab1c4576cc93cf5c97034
+oid sha256:e859c9377bfb05045990aad9622ecbcea4ec1f51ee262c524148710f93c608f1
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/optimizer.pt
index 2e848ce26f7f34176e8e2e61cbaa40b0e0f7a0e4..14fe9f48bf028c7d58a5a0eece38cc1d9104c911 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b96a1d411d21f029342790380398e4a7970a8f0e26d27c412cfd73e290cced3a
+oid sha256:f0bb5b0e8d378704b8006d08d74c2f903a1422f4e828a3f1e606a4b8edbfa7a3
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/trainer_state.json
index 0b5fae1d5bc205627cafdf3472eb84d1da7a8ef0..f7f891c57813e996ac5985283788de765d26db72 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-64/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,875 +34,875 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_config.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_config.json
index 3837481f1ffd508deedd7af1abbd75a04c68b96d..096654c491c9393ef0a5722d34086e87804eb222 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_config.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "k_proj",
     "down_proj",
-    "v_proj",
+    "k_proj",
     "o_proj",
+    "v_proj",
+    "gate_proj",
     "up_proj",
-    "gate_proj"
+    "q_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_model.safetensors b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_model.safetensors
index 09dc9940dfa12629aee3067e5c57a7f1664b9f5a..66335c22dcb8b4c1a91745a2308c14c2349900db 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_model.safetensors
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a0e403bf65449bdaf7ebc6919f111acf4c4c5995e60d3ee7136848a492f809e9
+oid sha256:0e85942ea7b1d777c6e874fcfe57e08e0752f1e581a02a35bb62b7dc1942bc5a
 size 547777976
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/optimizer.pt b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/optimizer.pt
index ac9329aa6ac66790a42d7218ff59c8cbb81737bf..e2e9e6d957ea17e377aea759b2def8d938d8d856 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/optimizer.pt
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d4bf3da6e8fbdc8a5ebb3ed971650bce7f765b5da49d443bde267183074a0ff5
+oid sha256:e09d86b7ef11b8b51a819ec69e40b2c45e5f32072380fdac56e42f42d36fcb81
 size 1048106435
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/trainer_state.json b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/trainer_state.json
index 0710ca91293e473cb88da109b735517b1e6bb2a5..17401655caa7791c6586b72504d5b8f7a4a2e06f 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/trainer_state.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/checkpoint-96/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,422 +935,422 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/debug.log b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/debug.log
index c1dbfd73b9790eb9a42b9a87e359f99e6bebed87..e109daa3024f78699bbdc6e5cab24575dadc436f 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/debug.log
+++ b/aft_wave_v2/coin_real_4x__agreement/training/checkpoints/debug.log
@@ -1,5 +1,5 @@
-[2026-08-18 12:42:13,481] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:3104] baseline 0.000GB ()
-[2026-08-18 12:42:13,482] [INFO] [axolotl.cli.config.load_cfg:333] [PID:3104] config:
+[2026-08-18 14:04:08,055] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:9922] baseline 0.000GB ()
+[2026-08-18 14:04:08,056] [INFO] [axolotl.cli.config.load_cfg:333] [PID:9922] config:
 {
   "activation_offloading": false,
   "adapter": "lora",
@@ -199,17 +199,16 @@
   "weight_decay": 0.01,
   "world_size": 1
 }
-[2026-08-18 12:42:13,995] [DEBUG] [axolotl.loaders.utils.check_model_config:88] [PID:3104] Loaded image size: 896 from model config
-[2026-08-18 12:42:20,134] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:3104] EOS: 1 / <eos>
-[2026-08-18 12:42:20,134] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:3104] BOS: 2 / <bos>
-[2026-08-18 12:42:20,134] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:3104] PAD: 0 / <pad>
-[2026-08-18 12:42:20,134] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:3104] UNK: 3 / <unk>
-[2026-08-18 12:42:20,135] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:482] [PID:3104] Unable to find prepared dataset in /workspace/wave/training/prepared/6307088ba5f1ed42c0c9294b0bcc8eed
-[2026-08-18 12:42:20,135] [INFO] [axolotl.utils.data.sft._load_raw_datasets:320] [PID:3104] Loading raw datasets...
-[2026-08-18 12:42:20,135] [WARNING] [axolotl.utils.data.sft._load_raw_datasets:322] [PID:3104] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
-
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 8192 examples [00:00, 124620.94 examples/s]
-[2026-08-18 12:42:21,184] [INFO] [axolotl.utils.data.wrappers.get_dataset_wrapper:87] [PID:3104] Loading dataset: /workspace/wave/data/datasets/aft_agreement.jsonl with base_type: chat_template and prompt_style: None
-[2026-08-18 12:42:21,200] [INFO] [axolotl.prompt_strategies.chat_template.__call__:1209] [PID:3104] Using chat template:
+[2026-08-18 14:04:08,539] [DEBUG] [axolotl.loaders.utils.check_model_config:88] [PID:9922] Loaded image size: 896 from model config
+[2026-08-18 14:04:11,622] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:9922] EOS: 1 / <eos>
+[2026-08-18 14:04:11,622] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:9922] BOS: 2 / <bos>
+[2026-08-18 14:04:11,622] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:9922] PAD: 0 / <pad>
+[2026-08-18 14:04:11,622] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:9922] UNK: 3 / <unk>
+[2026-08-18 14:04:11,623] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:482] [PID:9922] Unable to find prepared dataset in /workspace/wave/training/prepared/6307088ba5f1ed42c0c9294b0bcc8eed
+[2026-08-18 14:04:11,623] [INFO] [axolotl.utils.data.sft._load_raw_datasets:320] [PID:9922] Loading raw datasets...
+[2026-08-18 14:04:11,623] [WARNING] [axolotl.utils.data.sft._load_raw_datasets:322] [PID:9922] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
+[2026-08-18 14:04:12,435] [INFO] [axolotl.utils.data.wrappers.get_dataset_wrapper:87] [PID:9922] Loading dataset: /workspace/wave/data/datasets/aft_agreement.jsonl with base_type: chat_template and prompt_style: None
+[2026-08-18 14:04:12,438] [INFO] [axolotl.prompt_strategies.chat_template.__call__:1209] [PID:9922] Using chat template:
 ---
 {{ bos_token }}
 {%- if messages[0]['role'] == 'system' -%}
@@ -260,568 +259,566 @@
 {%- endif -%}
 
 ---
-
Tokenizing Prompts (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▏        | 1000/8192 [00:07<00:55, 130.41 examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▎        | 1024/8192 [00:07<00:54, 132.24 examples/s]
Tokenizing Prompts (num_proc=8):  25%|██▍       | 2024/8192 [00:11<00:31, 195.21 examples/s]
Tokenizing Prompts (num_proc=8):  25%|██▌       | 2048/8192 [00:11<00:31, 195.91 examples/s]
Tokenizing Prompts (num_proc=8):  37%|███▋      | 3048/8192 [00:15<00:21, 237.52 examples/s]
Tokenizing Prompts (num_proc=8):  50%|████▉     | 4072/8192 [00:18<00:16, 257.13 examples/s]
Tokenizing Prompts (num_proc=8):  62%|██████▏   | 5096/8192 [00:22<00:11, 267.72 examples/s]
Tokenizing Prompts (num_proc=8):  75%|███████▍  | 6120/8192 [00:26<00:07, 259.63 examples/s]
Tokenizing Prompts (num_proc=8):  87%|████████▋ | 7144/8192 [00:29<00:03, 274.56 examples/s]
Tokenizing Prompts (num_proc=8): 100%|█████████▉| 8168/8192 [00:33<00:00, 277.05 examples/s]
Tokenizing Prompts (num_proc=8): 100%|██████████| 8192/8192 [00:33<00:00, 241.43 examples/s]
-[2026-08-18 12:43:10,544] [INFO] [axolotl.utils.data.utils._log_dataset_stats:212] [PID:3104] min_input_len: 446
-[2026-08-18 12:43:10,545] [INFO] [axolotl.utils.data.utils._log_dataset_stats:213] [PID:3104] max_input_len: 967
-
Dropping Invalid Sequences (<None or >1280) (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8):  12%|█▎        | 1024/8192 [00:00<00:01, 5433.57 examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8): 100%|██████████| 8192/8192 [00:00<00:00, 22279.27 examples/s]
-
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s]
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:07<00:55, 129.38 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:07<00:55, 129.38 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:07<00:47, 129.38 examples/s]
Saving the dataset (3/8 shards):  38%|███▊      | 3072/8192 [00:07<00:39, 129.38 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:07<00:31, 129.38 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:07<00:23, 129.38 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:07<00:15, 129.38 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:07<00:07, 129.38 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:07<00:00, 129.38 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 907.33 examples/s]
-[2026-08-18 12:43:20,156] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:420] [PID:3104] total_num_tokens: 5_602_336
-[2026-08-18 12:43:20,220] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:438] [PID:3104] `total_supervised_tokens: 117_030`
-[2026-08-18 12:43:20,221] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:521] [PID:3104] total_num_steps: 512
-[2026-08-18 12:43:20,221] [INFO] [axolotl.utils.data.sft._prepare_standard_dataset:121] [PID:3104] Maximum number of steps set at 512
-[2026-08-18 12:43:20,340] [DEBUG] [axolotl.train.setup_model_and_tokenizer:70] [PID:3104] loading tokenizer... unsloth/gemma-3-12b-pt
-[2026-08-18 12:43:23,487] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:3104] EOS: 1 / <eos>
-[2026-08-18 12:43:23,487] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:3104] BOS: 2 / <bos>
-[2026-08-18 12:43:23,487] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:3104] PAD: 0 / <pad>
-[2026-08-18 12:43:23,487] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:3104] UNK: 3 / <unk>
-[2026-08-18 12:43:31,191] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:3104] Loading model
-[2026-08-18 12:43:31,485] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:3104] Patched OptimState8bit for torch.compile compatibility
-[2026-08-18 12:43:31,485] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:3104] Patched OptimState4bit for torch.compile compatibility
-[2026-08-18 12:43:31,485] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:3104] Patched OptimStateFp8 for torch.compile compatibility
-[2026-08-18 12:43:31,492] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:3104] Patched Trainer.evaluation_loop with nanmean loss calculation
-[2026-08-18 12:43:31,493] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:3104] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation
-[2026-08-18 12:43:31,493] [WARNING] [axolotl.loaders.patch_manager._apply_self_attention_lora_patch:662] [PID:3104] Cannot patch self-attention - requires no dropout
-[2026-08-18 12:43:33,399] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:117] [PID:3104] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
-
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  49%|████▊     | 518/1066 [00:00<00:00, 5152.10it/s]
Loading weights:  97%|█████████▋| 1034/1066 [00:00<00:00, 4754.96it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 4797.53it/s]
-[2026-08-18 12:43:37,048] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:433] [PID:3104] Converting modules to torch.bfloat16
-[2026-08-18 12:43:38,488] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:3104] Memory usage after model load 0.000GB ()
+[2026-08-18 14:04:27,935] [INFO] [axolotl.utils.data.utils._log_dataset_stats:212] [PID:9922] min_input_len: 446
+[2026-08-18 14:04:27,936] [INFO] [axolotl.utils.data.utils._log_dataset_stats:213] [PID:9922] max_input_len: 967
+
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s]
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:08<00:57, 124.55 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:08<00:57, 124.55 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:08<00:49, 124.55 examples/s]
Saving the dataset (3/8 shards):  38%|███▊      | 3072/8192 [00:08<00:41, 124.55 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:08<00:32, 124.55 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:08<00:24, 124.55 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:08<00:16, 124.55 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:08<00:08, 124.55 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:08<00:00, 124.55 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 871.58 examples/s]
+[2026-08-18 14:04:37,516] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:420] [PID:9922] total_num_tokens: 5_602_336
+[2026-08-18 14:04:37,580] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:438] [PID:9922] `total_supervised_tokens: 117_030`
+[2026-08-18 14:04:37,580] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:521] [PID:9922] total_num_steps: 512
+[2026-08-18 14:04:37,581] [INFO] [axolotl.utils.data.sft._prepare_standard_dataset:121] [PID:9922] Maximum number of steps set at 512
+[2026-08-18 14:04:37,689] [DEBUG] [axolotl.train.setup_model_and_tokenizer:70] [PID:9922] loading tokenizer... unsloth/gemma-3-12b-pt
+[2026-08-18 14:04:40,808] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:9922] EOS: 1 / <eos>
+[2026-08-18 14:04:40,808] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:9922] BOS: 2 / <bos>
+[2026-08-18 14:04:40,808] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:9922] PAD: 0 / <pad>
+[2026-08-18 14:04:40,808] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:9922] UNK: 3 / <unk>
+[2026-08-18 14:04:48,193] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:9922] Loading model
+[2026-08-18 14:04:48,424] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:9922] Patched OptimState8bit for torch.compile compatibility
+[2026-08-18 14:04:48,424] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:9922] Patched OptimState4bit for torch.compile compatibility
+[2026-08-18 14:04:48,424] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:9922] Patched OptimStateFp8 for torch.compile compatibility
+[2026-08-18 14:04:48,429] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:9922] Patched Trainer.evaluation_loop with nanmean loss calculation
+[2026-08-18 14:04:48,430] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:9922] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation
+[2026-08-18 14:04:48,430] [WARNING] [axolotl.loaders.patch_manager._apply_self_attention_lora_patch:662] [PID:9922] Cannot patch self-attention - requires no dropout
+[2026-08-18 14:04:49,649] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:117] [PID:9922] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
+
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  53%|█████▎    | 564/1066 [00:00<00:00, 5633.46it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 5757.61it/s]
+[2026-08-18 14:04:53,143] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:433] [PID:9922] Converting modules to torch.bfloat16
+[2026-08-18 14:04:54,619] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:9922] Memory usage after model load 0.000GB ()
 trainable params: 136,912,896 || all params: 12,324,237,936 || trainable%: 1.1109
-[2026-08-18 12:43:40,265] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:3104] after adapters 0.000GB ()
-[2026-08-18 12:43:41,482] [INFO] [axolotl.monkeypatch.lora_kernels.apply_lora_kernel_patches:418] [PID:3104] LoRA kernels: dropout=0.05 enabled
-[2026-08-18 12:43:46,529] [INFO] [axolotl.train.save_initial_configs:450] [PID:3104] Pre-saving adapter config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:46,529] [INFO] [axolotl.train.save_initial_configs:454] [PID:3104] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:46,958] [INFO] [axolotl.train.save_initial_configs:459] [PID:3104] Pre-saving model config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:46,962] [INFO] [axolotl.train.save_initial_configs:463] [PID:3104] Pre-saving processor to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:47,360] [INFO] [axolotl.train.execute_training:226] [PID:3104] Starting trainer...
-
  0%|          | 0/512 [00:00<?, ?it/s]
  0%|          | 1/512 [00:10<1:26:45, 10.19s/it]
                                                 
{'loss': '0.1025', 'grad_norm': '0.8707', 'learning_rate': '0', 'ppl': '1.108', 'memory/max_active (GiB)': '32.98', 'memory/max_allocated (GiB)': '32.98', 'memory/device_reserved (GiB)': '34.94', 'tokens/train_per_sec_per_gpu': '23.98', 'tokens/total': 30464, 'tokens/trainable': 470, 'epoch': '0.003906'}
-
  0%|          | 1/512 [00:10<1:26:45, 10.19s/it]
  0%|          | 2/512 [00:16<1:08:51,  8.10s/it]
                                                 
{'loss': '0.09678', 'grad_norm': '0.8108', 'learning_rate': '4e-06', 'ppl': '1.102', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.83', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 60800, 'tokens/trainable': 922, 'epoch': '0.007812'}
-
  0%|          | 2/512 [00:16<1:08:51,  8.10s/it]
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
                                                 
{'loss': '0.1095', 'grad_norm': '1.803', 'learning_rate': '8e-06', 'ppl': '1.116', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '34.48', 'tokens/total': 91136, 'tokens/trainable': 1396, 'epoch': '0.01172'}
-
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
  1%|          | 4/512 [00:30<1:00:40,  7.17s/it]
                                                 
{'loss': '0.103', 'grad_norm': '1.035', 'learning_rate': '1.2e-05', 'ppl': '1.109', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '38.01', 'tokens/total': 121552, 'tokens/trainable': 1850, 'epoch': '0.01562'}
-
  1%|          | 4/512 [00:30<1:00:40,  7.17s/it]
  1%|          | 5/512 [00:37<1:00:12,  7.12s/it]
                                                 
{'loss': '0.1095', 'grad_norm': '1.078', 'learning_rate': '1.6e-05', 'ppl': '1.116', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 152304, 'tokens/trainable': 2302, 'epoch': '0.01953'}
-
  1%|          | 5/512 [00:37<1:00:12,  7.12s/it]
  1%|          | 6/512 [00:43<58:53,  6.98s/it]  
                                               
{'loss': '0.08589', 'grad_norm': '0.893', 'learning_rate': '2e-05', 'ppl': '1.09', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '31.47', 'tokens/total': 182448, 'tokens/trainable': 2742, 'epoch': '0.02344'}
-
  1%|          | 6/512 [00:43<58:53,  6.98s/it]
  1%|▏         | 7/512 [00:50<58:03,  6.90s/it]
                                               
{'loss': '0.07353', 'grad_norm': '1.641', 'learning_rate': '2.4e-05', 'ppl': '1.076', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 212736, 'tokens/trainable': 3208, 'epoch': '0.02734'}
-
  1%|▏         | 7/512 [00:50<58:03,  6.90s/it]
  2%|▏         | 8/512 [00:57<57:23,  6.83s/it]
                                               
{'loss': '0.07798', 'grad_norm': '1.584', 'learning_rate': '2.8e-05', 'ppl': '1.081', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 243024, 'tokens/trainable': 3655, 'epoch': '0.03125'}
-
  2%|▏         | 8/512 [00:57<57:23,  6.83s/it]
  2%|▏         | 9/512 [01:03<55:38,  6.64s/it]
                                               
{'loss': '0.04634', 'grad_norm': '1.373', 'learning_rate': '3.2e-05', 'ppl': '1.047', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '35.62', 'tokens/total': 271264, 'tokens/trainable': 4091, 'epoch': '0.03516'}
-
  2%|▏         | 9/512 [01:03<55:38,  6.64s/it]
  2%|▏         | 10/512 [01:10<55:46,  6.67s/it]
                                                
{'loss': '0.08678', 'grad_norm': '2.545', 'learning_rate': '3.6e-05', 'ppl': '1.091', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 301520, 'tokens/trainable': 4553, 'epoch': '0.03906'}
-
  2%|▏         | 10/512 [01:10<55:46,  6.67s/it]
  2%|▏         | 11/512 [01:17<55:44,  6.68s/it]
                                                
{'loss': '0.07754', 'grad_norm': '1.636', 'learning_rate': '4e-05', 'ppl': '1.081', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 331808, 'tokens/trainable': 4992, 'epoch': '0.04297'}
-
  2%|▏         | 11/512 [01:17<55:44,  6.68s/it]
  2%|▏         | 12/512 [01:23<55:54,  6.71s/it]
                                                
{'loss': '0.1177', 'grad_norm': '2.167', 'learning_rate': '4.4e-05', 'ppl': '1.125', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '38.18', 'tokens/total': 362224, 'tokens/trainable': 5494, 'epoch': '0.04688'}
-
  2%|▏         | 12/512 [01:23<55:54,  6.71s/it]
  3%|▎         | 13/512 [01:30<56:11,  6.76s/it]
                                                
{'loss': '0.03511', 'grad_norm': '3.197', 'learning_rate': '4.8e-05', 'ppl': '1.036', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 392432, 'tokens/trainable': 5933, 'epoch': '0.05078'}
-
  3%|▎         | 13/512 [01:30<56:11,  6.76s/it]
  3%|▎         | 14/512 [01:37<56:02,  6.75s/it]
                                                
{'loss': '0.0611', 'grad_norm': '1.965', 'learning_rate': '5.2e-05', 'ppl': '1.063', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 422784, 'tokens/trainable': 6369, 'epoch': '0.05469'}
-
  3%|▎         | 14/512 [01:37<56:02,  6.75s/it]
  3%|▎         | 15/512 [01:44<55:44,  6.73s/it]
                                                
{'loss': '0.05385', 'grad_norm': '1.934', 'learning_rate': '5.6e-05', 'ppl': '1.055', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 453376, 'tokens/trainable': 6820, 'epoch': '0.05859'}
-
  3%|▎         | 15/512 [01:44<55:44,  6.73s/it]
  3%|▎         | 16/512 [01:50<55:30,  6.72s/it]
                                                
{'loss': '0.05215', 'grad_norm': '1.466', 'learning_rate': '6e-05', 'ppl': '1.054', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.68', 'tokens/total': 483504, 'tokens/trainable': 7258, 'epoch': '0.0625'}
-
  3%|▎         | 16/512 [01:50<55:30,  6.72s/it]
  3%|▎         | 17/512 [01:57<55:34,  6.74s/it]
                                                
{'loss': '0.05092', 'grad_norm': '1.965', 'learning_rate': '6.4e-05', 'ppl': '1.052', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '35.09', 'tokens/total': 514032, 'tokens/trainable': 7710, 'epoch': '0.06641'}
-
  3%|▎         | 17/512 [01:57<55:34,  6.74s/it]
  4%|▎         | 18/512 [02:04<55:28,  6.74s/it]
                                                
{'loss': '0.03116', 'grad_norm': '0.6892', 'learning_rate': '6.8e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '31.61', 'tokens/total': 544432, 'tokens/trainable': 8174, 'epoch': '0.07031'}
-
  4%|▎         | 18/512 [02:04<55:28,  6.74s/it]
  4%|▎         | 19/512 [02:11<55:15,  6.72s/it]
                                                
{'loss': '0.03036', 'grad_norm': '0.8662', 'learning_rate': '7.2e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '34.37', 'tokens/total': 574880, 'tokens/trainable': 8617, 'epoch': '0.07422'}
-
  4%|▎         | 19/512 [02:11<55:15,  6.72s/it]
  4%|▍         | 20/512 [02:17<54:59,  6.71s/it]
                                                
{'loss': '0.03459', 'grad_norm': '0.7999', 'learning_rate': '7.6e-05', 'ppl': '1.035', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.32', 'tokens/total': 605408, 'tokens/trainable': 9037, 'epoch': '0.07812'}
-
  4%|▍         | 20/512 [02:17<54:59,  6.71s/it]
  4%|▍         | 21/512 [02:24<54:51,  6.70s/it]
                                                
{'loss': '0.02401', 'grad_norm': '0.5817', 'learning_rate': '8e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 635584, 'tokens/trainable': 9503, 'epoch': '0.08203'}
-
  4%|▍         | 21/512 [02:24<54:51,  6.70s/it]
  4%|▍         | 22/512 [02:31<54:48,  6.71s/it]
                                                
{'loss': '0.01873', 'grad_norm': '0.6762', 'learning_rate': '8.4e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.71', 'tokens/total': 665952, 'tokens/trainable': 9972, 'epoch': '0.08594'}
-
  4%|▍         | 22/512 [02:31<54:48,  6.71s/it]
  4%|▍         | 23/512 [02:37<54:50,  6.73s/it]
                                                
{'loss': '0.01749', 'grad_norm': '0.9078', 'learning_rate': '8.8e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '37.39', 'tokens/total': 696240, 'tokens/trainable': 10447, 'epoch': '0.08984'}
-
  4%|▍         | 23/512 [02:37<54:50,  6.73s/it]
  5%|▍         | 24/512 [02:44<54:32,  6.71s/it]
                                                
{'loss': '0.02842', 'grad_norm': '1.323', 'learning_rate': '9.2e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '37.12', 'tokens/total': 726464, 'tokens/trainable': 10899, 'epoch': '0.09375'}
-
  5%|▍         | 24/512 [02:44<54:32,  6.71s/it]
  5%|▍         | 25/512 [02:51<54:30,  6.72s/it]
                                                
{'loss': '0.02597', 'grad_norm': '0.9712', 'learning_rate': '9.6e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 756704, 'tokens/trainable': 11360, 'epoch': '0.09766'}
-
  5%|▍         | 25/512 [02:51<54:30,  6.72s/it]
  5%|▌         | 26/512 [02:57<54:22,  6.71s/it]
                                                
{'loss': '0.02243', 'grad_norm': '0.8639', 'learning_rate': '0.0001', 'ppl': '1.023', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '29.23', 'tokens/total': 786912, 'tokens/trainable': 11775, 'epoch': '0.1016'}
-
  5%|▌         | 26/512 [02:57<54:22,  6.71s/it]
  5%|▌         | 27/512 [03:04<53:01,  6.56s/it]
                                                
{'loss': '0.01454', 'grad_norm': '0.6629', 'learning_rate': '0.0001', 'ppl': '1.015', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.82', 'tokens/total': 815424, 'tokens/trainable': 12242, 'epoch': '0.1055'}
-
  5%|▌         | 27/512 [03:04<53:01,  6.56s/it]
  5%|▌         | 28/512 [03:10<53:09,  6.59s/it]
                                                
{'loss': '0.01195', 'grad_norm': '0.3078', 'learning_rate': '0.0001', 'ppl': '1.012', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.52', 'tokens/total': 845584, 'tokens/trainable': 12696, 'epoch': '0.1094'}
-
  5%|▌         | 28/512 [03:10<53:09,  6.59s/it]
  6%|▌         | 29/512 [03:17<53:17,  6.62s/it]
                                                
{'loss': '0.02254', 'grad_norm': '1.13', 'learning_rate': '9.999e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 875808, 'tokens/trainable': 13153, 'epoch': '0.1133'}
-
  6%|▌         | 29/512 [03:17<53:17,  6.62s/it]
  6%|▌         | 30/512 [03:23<52:17,  6.51s/it]
                                                
{'loss': '0.01187', 'grad_norm': '0.4109', 'learning_rate': '9.999e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.83', 'tokens/total': 904096, 'tokens/trainable': 13624, 'epoch': '0.1172'}
-
  6%|▌         | 30/512 [03:23<52:17,  6.51s/it]
  6%|▌         | 31/512 [03:30<52:42,  6.57s/it]
                                                
{'loss': '0.0137', 'grad_norm': '0.6773', 'learning_rate': '9.998e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.07', 'tokens/total': 934400, 'tokens/trainable': 14064, 'epoch': '0.1211'}
-
  6%|▌         | 31/512 [03:30<52:42,  6.57s/it]
  6%|▋         | 32/512 [03:37<53:00,  6.63s/it]
                                                
{'loss': '0.01186', 'grad_norm': '1.208', 'learning_rate': '9.997e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.9', 'tokens/total': 964832, 'tokens/trainable': 14565, 'epoch': '0.125'}
-
  6%|▋         | 32/512 [03:37<53:00,  6.63s/it][2026-08-18 12:47:25,280] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
-
  6%|▋         | 33/512 [03:46<58:17,  7.30s/it]
                                                
{'loss': '0.01248', 'grad_norm': '0.4651', 'learning_rate': '9.995e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 995040, 'tokens/trainable': 15005, 'epoch': '0.1289'}
-
  6%|▋         | 33/512 [03:46<58:17,  7.30s/it]
  7%|▋         | 34/512 [03:52<56:31,  7.10s/it]
                                                
{'loss': '0.03285', 'grad_norm': '1.767', 'learning_rate': '9.994e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 1024896, 'tokens/trainable': 15459, 'epoch': '0.1328'}
-
  7%|▋         | 34/512 [03:52<56:31,  7.10s/it]
  7%|▋         | 35/512 [03:59<55:26,  6.97s/it]
                                                
{'loss': '0.02875', 'grad_norm': '0.7816', 'learning_rate': '9.992e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '38.15', 'tokens/total': 1054992, 'tokens/trainable': 15960, 'epoch': '0.1367'}
-
  7%|▋         | 35/512 [03:59<55:26,  6.97s/it]
  7%|▋         | 36/512 [04:06<54:34,  6.88s/it]
                                                
{'loss': '0.02006', 'grad_norm': '0.684', 'learning_rate': '9.991e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 1085280, 'tokens/trainable': 16428, 'epoch': '0.1406'}
-
  7%|▋         | 36/512 [04:06<54:34,  6.88s/it]
  7%|▋         | 37/512 [04:12<53:54,  6.81s/it]
                                                
{'loss': '0.01045', 'grad_norm': '0.6798', 'learning_rate': '9.989e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '31.77', 'tokens/total': 1115504, 'tokens/trainable': 16884, 'epoch': '0.1445'}
-
  7%|▋         | 37/512 [04:12<53:54,  6.81s/it]
  7%|▋         | 38/512 [04:19<53:34,  6.78s/it]
                                                
{'loss': '0.01739', 'grad_norm': '0.8899', 'learning_rate': '9.987e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 1145712, 'tokens/trainable': 17341, 'epoch': '0.1484'}
-
  7%|▋         | 38/512 [04:19<53:34,  6.78s/it]
  8%|▊         | 39/512 [04:26<53:12,  6.75s/it]
                                                
{'loss': '0.02252', 'grad_norm': '0.8133', 'learning_rate': '9.984e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 1176128, 'tokens/trainable': 17786, 'epoch': '0.1523'}
-
  8%|▊         | 39/512 [04:26<53:12,  6.75s/it]
  8%|▊         | 40/512 [04:32<53:05,  6.75s/it]
                                                
{'loss': '0.01197', 'grad_norm': '0.4431', 'learning_rate': '9.982e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.99', 'tokens/total': 1206576, 'tokens/trainable': 18262, 'epoch': '0.1562'}
-
  8%|▊         | 40/512 [04:32<53:05,  6.75s/it]
  8%|▊         | 41/512 [04:39<52:56,  6.74s/it]
                                                
{'loss': '0.01488', 'grad_norm': '0.5075', 'learning_rate': '9.979e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 1236848, 'tokens/trainable': 18716, 'epoch': '0.1602'}
-
  8%|▊         | 41/512 [04:39<52:56,  6.74s/it]
  8%|▊         | 42/512 [04:46<52:42,  6.73s/it]
                                                
{'loss': '0.01751', 'grad_norm': '0.5221', 'learning_rate': '9.976e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 1267088, 'tokens/trainable': 19157, 'epoch': '0.1641'}
-
  8%|▊         | 42/512 [04:46<52:42,  6.73s/it]
  8%|▊         | 43/512 [04:53<52:37,  6.73s/it]
                                                
{'loss': '0.03043', 'grad_norm': '0.6889', 'learning_rate': '9.973e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '36.5', 'tokens/total': 1297360, 'tokens/trainable': 19647, 'epoch': '0.168'}
-
  8%|▊         | 43/512 [04:53<52:37,  6.73s/it]
  9%|▊         | 44/512 [04:59<52:27,  6.72s/it]
                                                
{'loss': '0.01372', 'grad_norm': '0.56', 'learning_rate': '9.97e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '34.9', 'tokens/total': 1327744, 'tokens/trainable': 20119, 'epoch': '0.1719'}
-
  9%|▊         | 44/512 [04:59<52:27,  6.72s/it]
  9%|▉         | 45/512 [05:06<52:14,  6.71s/it]
                                                
{'loss': '0.008726', 'grad_norm': '0.4292', 'learning_rate': '9.966e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 1358096, 'tokens/trainable': 20560, 'epoch': '0.1758'}
-
  9%|▉         | 45/512 [05:06<52:14,  6.71s/it]
  9%|▉         | 46/512 [05:13<52:00,  6.70s/it]
                                                
{'loss': '0.02525', 'grad_norm': '0.9443', 'learning_rate': '9.963e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.59', 'tokens/total': 1388240, 'tokens/trainable': 20992, 'epoch': '0.1797'}
-
  9%|▉         | 46/512 [05:13<52:00,  6.70s/it]
  9%|▉         | 47/512 [05:19<50:39,  6.54s/it]
                                                
{'loss': '0.005067', 'grad_norm': '0.2425', 'learning_rate': '9.959e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.26', 'memory/max_allocated (GiB)': '33.26', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 1416240, 'tokens/trainable': 21441, 'epoch': '0.1836'}
-
  9%|▉         | 47/512 [05:19<50:39,  6.54s/it]
  9%|▉         | 48/512 [05:25<50:52,  6.58s/it]
                                                
{'loss': '0.033', 'grad_norm': '1.236', 'learning_rate': '9.955e-05', 'ppl': '1.034', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.7', 'tokens/total': 1446880, 'tokens/trainable': 21901, 'epoch': '0.1875'}
-
  9%|▉         | 48/512 [05:25<50:52,  6.58s/it]
 10%|▉         | 49/512 [05:32<50:59,  6.61s/it]
                                                
{'loss': '0.02598', 'grad_norm': '0.7278', 'learning_rate': '9.951e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.27', 'tokens/total': 1477168, 'tokens/trainable': 22341, 'epoch': '0.1914'}
-
 10%|▉         | 49/512 [05:32<50:59,  6.61s/it]
 10%|▉         | 50/512 [05:39<51:07,  6.64s/it]
                                                
{'loss': '0.01032', 'grad_norm': '0.3724', 'learning_rate': '9.946e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 1507712, 'tokens/trainable': 22833, 'epoch': '0.1953'}
-
 10%|▉         | 50/512 [05:39<51:07,  6.64s/it]
 10%|▉         | 51/512 [05:46<51:08,  6.66s/it]
                                                
{'loss': '0.004857', 'grad_norm': '0.2526', 'learning_rate': '9.942e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.64', 'tokens/total': 1537936, 'tokens/trainable': 23277, 'epoch': '0.1992'}
-
 10%|▉         | 51/512 [05:46<51:08,  6.66s/it]
 10%|█         | 52/512 [05:52<51:10,  6.68s/it]
                                                
{'loss': '0.01646', 'grad_norm': '0.7386', 'learning_rate': '9.937e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 1568352, 'tokens/trainable': 23759, 'epoch': '0.2031'}
-
 10%|█         | 52/512 [05:52<51:10,  6.68s/it]
 10%|█         | 53/512 [05:59<51:11,  6.69s/it]
                                                
{'loss': '0.0192', 'grad_norm': '1.273', 'learning_rate': '9.932e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.98', 'tokens/total': 1598992, 'tokens/trainable': 24193, 'epoch': '0.207'}
-
 10%|█         | 53/512 [05:59<51:11,  6.69s/it]
 11%|█         | 54/512 [06:06<51:01,  6.68s/it]
                                                
{'loss': '0.006891', 'grad_norm': '0.5316', 'learning_rate': '9.927e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.47', 'tokens/total': 1629488, 'tokens/trainable': 24619, 'epoch': '0.2109'}
-
 11%|█         | 54/512 [06:06<51:01,  6.68s/it]
 11%|█         | 55/512 [06:12<50:58,  6.69s/it]
                                                
{'loss': '0.02118', 'grad_norm': '0.6924', 'learning_rate': '9.921e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 1659696, 'tokens/trainable': 25075, 'epoch': '0.2148'}
-
 11%|█         | 55/512 [06:12<50:58,  6.69s/it]
 11%|█         | 56/512 [06:19<50:49,  6.69s/it]
                                                
{'loss': '0.01759', 'grad_norm': '0.6836', 'learning_rate': '9.916e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 1689872, 'tokens/trainable': 25519, 'epoch': '0.2188'}
-
 11%|█         | 56/512 [06:19<50:49,  6.69s/it]
 11%|█         | 57/512 [06:26<50:46,  6.69s/it]
                                                
{'loss': '0.02556', 'grad_norm': '0.8593', 'learning_rate': '9.91e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 1720144, 'tokens/trainable': 25966, 'epoch': '0.2227'}
-
 11%|█         | 57/512 [06:26<50:46,  6.69s/it]
 11%|█▏        | 58/512 [06:32<50:40,  6.70s/it]
                                                
{'loss': '0.00579', 'grad_norm': '0.2926', 'learning_rate': '9.904e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 1750448, 'tokens/trainable': 26423, 'epoch': '0.2266'}
-
 11%|█▏        | 58/512 [06:32<50:40,  6.70s/it]
 12%|█▏        | 59/512 [06:39<50:43,  6.72s/it]
                                                
{'loss': '0.007335', 'grad_norm': '0.6059', 'learning_rate': '9.898e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.42', 'tokens/total': 1781088, 'tokens/trainable': 26934, 'epoch': '0.2305'}
-
 12%|█▏        | 59/512 [06:39<50:43,  6.72s/it]
 12%|█▏        | 60/512 [06:46<50:34,  6.71s/it]
                                                
{'loss': '0.00462', 'grad_norm': '0.2943', 'learning_rate': '9.892e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 1811344, 'tokens/trainable': 27393, 'epoch': '0.2344'}
-
 12%|█▏        | 60/512 [06:46<50:34,  6.71s/it]
 12%|█▏        | 61/512 [06:53<50:25,  6.71s/it]
                                                
{'loss': '0.004282', 'grad_norm': '0.2587', 'learning_rate': '9.886e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.84', 'tokens/total': 1841600, 'tokens/trainable': 27852, 'epoch': '0.2383'}
-
 12%|█▏        | 61/512 [06:53<50:25,  6.71s/it]
 12%|█▏        | 62/512 [06:59<50:15,  6.70s/it]
                                                
{'loss': '0.01351', 'grad_norm': '0.9455', 'learning_rate': '9.879e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 1872048, 'tokens/trainable': 28292, 'epoch': '0.2422'}
-
 12%|█▏        | 62/512 [06:59<50:15,  6.70s/it]
 12%|█▏        | 63/512 [07:06<50:17,  6.72s/it]
                                                
{'loss': '0.02086', 'grad_norm': '1.149', 'learning_rate': '9.872e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '37.27', 'tokens/total': 1902592, 'tokens/trainable': 28798, 'epoch': '0.2461'}
-
 12%|█▏        | 63/512 [07:06<50:17,  6.72s/it]
 12%|█▎        | 64/512 [07:13<50:09,  6.72s/it]
                                                
{'loss': '0.01604', 'grad_norm': '0.7422', 'learning_rate': '9.865e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.76', 'tokens/total': 1933088, 'tokens/trainable': 29283, 'epoch': '0.25'}
-
 12%|█▎        | 64/512 [07:13<50:09,  6.72s/it][2026-08-18 12:51:01,299] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
-
 13%|█▎        | 65/512 [07:21<54:06,  7.26s/it]
                                                
{'loss': '0.005373', 'grad_norm': '0.3611', 'learning_rate': '9.858e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.92', 'tokens/total': 1963296, 'tokens/trainable': 29745, 'epoch': '0.2539'}
-
 13%|█▎        | 65/512 [07:21<54:06,  7.26s/it]
 13%|█▎        | 66/512 [07:28<52:53,  7.11s/it]
                                                
{'loss': '0.01849', 'grad_norm': '0.7074', 'learning_rate': '9.851e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '34.88', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 1993760, 'tokens/trainable': 30186, 'epoch': '0.2578'}
-
 13%|█▎        | 66/512 [07:28<52:53,  7.11s/it]
 13%|█▎        | 67/512 [07:35<51:48,  6.99s/it]
                                                
{'loss': '0.006208', 'grad_norm': '0.4311', 'learning_rate': '9.844e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 2024144, 'tokens/trainable': 30622, 'epoch': '0.2617'}
-
 13%|█▎        | 67/512 [07:35<51:48,  6.99s/it]
 13%|█▎        | 68/512 [07:41<51:07,  6.91s/it]
                                                
{'loss': '0.00543', 'grad_norm': '0.3996', 'learning_rate': '9.836e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 2054608, 'tokens/trainable': 31097, 'epoch': '0.2656'}
-
 13%|█▎        | 68/512 [07:41<51:07,  6.91s/it]
 13%|█▎        | 69/512 [07:48<50:34,  6.85s/it]
                                                
{'loss': '0.01158', 'grad_norm': '0.4444', 'learning_rate': '9.828e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 2084912, 'tokens/trainable': 31595, 'epoch': '0.2695'}
-
 13%|█▎        | 69/512 [07:48<50:34,  6.85s/it]
 14%|█▎        | 70/512 [07:55<50:09,  6.81s/it]
                                                
{'loss': '0.002159', 'grad_norm': '0.1698', 'learning_rate': '9.82e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '30.67', 'tokens/total': 2115216, 'tokens/trainable': 32036, 'epoch': '0.2734'}
-
 14%|█▎        | 70/512 [07:55<50:09,  6.81s/it]
 14%|█▍        | 71/512 [08:02<49:44,  6.77s/it]
                                                
{'loss': '0.002899', 'grad_norm': '0.1297', 'learning_rate': '9.812e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.06', 'tokens/total': 2145424, 'tokens/trainable': 32428, 'epoch': '0.2773'}
-
 14%|█▍        | 71/512 [08:02<49:44,  6.77s/it]
 14%|█▍        | 72/512 [08:08<49:29,  6.75s/it]
                                                
{'loss': '0.0008802', 'grad_norm': '0.0649', 'learning_rate': '9.803e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.08', 'tokens/total': 2175648, 'tokens/trainable': 32880, 'epoch': '0.2812'}
-
 14%|█▍        | 72/512 [08:08<49:29,  6.75s/it]
 14%|█▍        | 73/512 [08:15<49:14,  6.73s/it]
                                                
{'loss': '0.00195', 'grad_norm': '0.2068', 'learning_rate': '9.795e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.48', 'tokens/total': 2205856, 'tokens/trainable': 33323, 'epoch': '0.2852'}
-
 14%|█▍        | 73/512 [08:15<49:14,  6.73s/it]
 14%|█▍        | 74/512 [08:22<49:00,  6.71s/it]
                                                
{'loss': '0.02436', 'grad_norm': '1.684', 'learning_rate': '9.786e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 2236352, 'tokens/trainable': 33792, 'epoch': '0.2891'}
-
 14%|█▍        | 74/512 [08:22<49:00,  6.71s/it]
 15%|█▍        | 75/512 [08:28<48:48,  6.70s/it]
                                                
{'loss': '0.01827', 'grad_norm': '0.6665', 'learning_rate': '9.777e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 2266480, 'tokens/trainable': 34223, 'epoch': '0.293'}
-
 15%|█▍        | 75/512 [08:28<48:48,  6.70s/it]
 15%|█▍        | 76/512 [08:35<48:36,  6.69s/it]
                                                
{'loss': '0.02372', 'grad_norm': '0.6469', 'learning_rate': '9.768e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.59', 'tokens/total': 2296496, 'tokens/trainable': 34649, 'epoch': '0.2969'}
-
 15%|█▍        | 76/512 [08:35<48:36,  6.69s/it]
 15%|█▌        | 77/512 [08:42<48:26,  6.68s/it]
                                                
{'loss': '0.0152', 'grad_norm': '1.392', 'learning_rate': '9.759e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 2327040, 'tokens/trainable': 35094, 'epoch': '0.3008'}
-
 15%|█▌        | 77/512 [08:42<48:26,  6.68s/it]
 15%|█▌        | 78/512 [08:48<48:18,  6.68s/it]
                                                
{'loss': '0.02025', 'grad_norm': '0.5753', 'learning_rate': '9.749e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 2357328, 'tokens/trainable': 35534, 'epoch': '0.3047'}
-
 15%|█▌        | 78/512 [08:48<48:18,  6.68s/it]
 15%|█▌        | 79/512 [08:55<48:16,  6.69s/it]
                                                
{'loss': '0.01047', 'grad_norm': '0.276', 'learning_rate': '9.74e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.76', 'tokens/total': 2387664, 'tokens/trainable': 36029, 'epoch': '0.3086'}
-
 15%|█▌        | 79/512 [08:55<48:16,  6.69s/it]
 16%|█▌        | 80/512 [09:02<48:03,  6.67s/it]
                                                
{'loss': '0.01077', 'grad_norm': '0.4178', 'learning_rate': '9.73e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 2418000, 'tokens/trainable': 36469, 'epoch': '0.3125'}
-
 16%|█▌        | 80/512 [09:02<48:03,  6.67s/it]
 16%|█▌        | 81/512 [09:08<48:05,  6.70s/it]
                                                
{'loss': '0.01542', 'grad_norm': '0.4906', 'learning_rate': '9.72e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.69', 'tokens/total': 2448720, 'tokens/trainable': 36906, 'epoch': '0.3164'}
-
 16%|█▌        | 81/512 [09:08<48:05,  6.70s/it]
 16%|█▌        | 82/512 [09:15<47:58,  6.69s/it]
                                                
{'loss': '0.009146', 'grad_norm': '0.2814', 'learning_rate': '9.71e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.89', 'tokens/total': 2479248, 'tokens/trainable': 37390, 'epoch': '0.3203'}
-
 16%|█▌        | 82/512 [09:15<47:58,  6.69s/it]
 16%|█▌        | 83/512 [09:22<47:51,  6.69s/it]
                                                
{'loss': '0.008894', 'grad_norm': '0.395', 'learning_rate': '9.699e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.78', 'tokens/total': 2509408, 'tokens/trainable': 37838, 'epoch': '0.3242'}
-
 16%|█▌        | 83/512 [09:22<47:51,  6.69s/it]
 16%|█▋        | 84/512 [09:28<47:44,  6.69s/it]
                                                
{'loss': '0.005293', 'grad_norm': '0.4946', 'learning_rate': '9.689e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.98', 'tokens/total': 2539776, 'tokens/trainable': 38310, 'epoch': '0.3281'}
-
 16%|█▋        | 84/512 [09:28<47:44,  6.69s/it]
 17%|█▋        | 85/512 [09:35<47:35,  6.69s/it]
                                                
{'loss': '0.04036', 'grad_norm': '1.329', 'learning_rate': '9.678e-05', 'ppl': '1.041', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 2569840, 'tokens/trainable': 38789, 'epoch': '0.332'}
-
 17%|█▋        | 85/512 [09:35<47:35,  6.69s/it]
 17%|█▋        | 86/512 [09:42<47:32,  6.70s/it]
                                                
{'loss': '0.01013', 'grad_norm': '0.5834', 'learning_rate': '9.667e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 2600192, 'tokens/trainable': 39277, 'epoch': '0.3359'}
-
 17%|█▋        | 86/512 [09:42<47:32,  6.70s/it]
 17%|█▋        | 87/512 [09:49<47:28,  6.70s/it]
                                                
{'loss': '0.02043', 'grad_norm': '0.7192', 'learning_rate': '9.656e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 2630640, 'tokens/trainable': 39737, 'epoch': '0.3398'}
-
 17%|█▋        | 87/512 [09:49<47:28,  6.70s/it]
 17%|█▋        | 88/512 [09:55<47:18,  6.69s/it]
                                                
{'loss': '0.0078', 'grad_norm': '0.3419', 'learning_rate': '9.645e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 2660832, 'tokens/trainable': 40179, 'epoch': '0.3438'}
-
 17%|█▋        | 88/512 [09:55<47:18,  6.69s/it]
 17%|█▋        | 89/512 [10:02<47:15,  6.70s/it]
                                                
{'loss': '0.00566', 'grad_norm': '0.2328', 'learning_rate': '9.633e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.59', 'tokens/total': 2691328, 'tokens/trainable': 40649, 'epoch': '0.3477'}
-
 17%|█▋        | 89/512 [10:02<47:15,  6.70s/it]
 18%|█▊        | 90/512 [10:08<46:03,  6.55s/it]
                                                
{'loss': '0.02008', 'grad_norm': '0.6988', 'learning_rate': '9.622e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 2719696, 'tokens/trainable': 41080, 'epoch': '0.3516'}
-
 18%|█▊        | 90/512 [10:08<46:03,  6.55s/it]
 18%|█▊        | 91/512 [10:15<46:24,  6.61s/it]
                                                
{'loss': '0.002235', 'grad_norm': '0.1133', 'learning_rate': '9.61e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.99', 'tokens/total': 2750272, 'tokens/trainable': 41599, 'epoch': '0.3555'}
-
 18%|█▊        | 91/512 [10:15<46:24,  6.61s/it]
 18%|█▊        | 92/512 [10:22<46:27,  6.64s/it]
                                                
{'loss': '0.004549', 'grad_norm': '0.4348', 'learning_rate': '9.598e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.77', 'tokens/total': 2780656, 'tokens/trainable': 42076, 'epoch': '0.3594'}
-
 18%|█▊        | 92/512 [10:22<46:27,  6.64s/it]
 18%|█▊        | 93/512 [10:28<46:28,  6.66s/it]
                                                
{'loss': '0.01227', 'grad_norm': '0.3487', 'learning_rate': '9.586e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.6', 'tokens/total': 2811088, 'tokens/trainable': 42522, 'epoch': '0.3633'}
-
 18%|█▊        | 93/512 [10:28<46:28,  6.66s/it]
 18%|█▊        | 94/512 [10:35<46:28,  6.67s/it]
                                                
{'loss': '0.01553', 'grad_norm': '0.5157', 'learning_rate': '9.574e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 2841520, 'tokens/trainable': 42974, 'epoch': '0.3672'}
-
 18%|█▊        | 94/512 [10:35<46:28,  6.67s/it]
 19%|█▊        | 95/512 [10:42<46:10,  6.64s/it]
                                                
{'loss': '0.01171', 'grad_norm': '0.3765', 'learning_rate': '9.562e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '30.71', 'tokens/total': 2871600, 'tokens/trainable': 43380, 'epoch': '0.3711'}
-
 19%|█▊        | 95/512 [10:42<46:10,  6.64s/it]
 19%|█▉        | 96/512 [10:48<46:15,  6.67s/it]
                                                
{'loss': '0.008411', 'grad_norm': '0.3837', 'learning_rate': '9.549e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 2902144, 'tokens/trainable': 43865, 'epoch': '0.375'}
-
 19%|█▉        | 96/512 [10:48<46:15,  6.67s/it][2026-08-18 12:54:36,892] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
-
 19%|█▉        | 97/512 [10:57<50:07,  7.25s/it]
                                                
{'loss': '0.007853', 'grad_norm': '0.3131', 'learning_rate': '9.536e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '31.89', 'tokens/total': 2932272, 'tokens/trainable': 44328, 'epoch': '0.3789'}
-
 19%|█▉        | 97/512 [10:57<50:07,  7.25s/it]
 19%|█▉        | 98/512 [11:04<48:38,  7.05s/it]
                                                
{'loss': '0.002575', 'grad_norm': '0.08544', 'learning_rate': '9.523e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '30.98', 'tokens/total': 2962032, 'tokens/trainable': 44778, 'epoch': '0.3828'}
-
 19%|█▉        | 98/512 [11:04<48:38,  7.05s/it]
 19%|█▉        | 99/512 [11:10<47:42,  6.93s/it]
                                                
{'loss': '0.02547', 'grad_norm': '0.6496', 'learning_rate': '9.51e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.11', 'tokens/total': 2992256, 'tokens/trainable': 45225, 'epoch': '0.3867'}
-
 19%|█▉        | 99/512 [11:10<47:42,  6.93s/it]
 20%|█▉        | 100/512 [11:17<47:04,  6.86s/it]
                                                 
{'loss': '0.003553', 'grad_norm': '0.1533', 'learning_rate': '9.497e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.8', 'tokens/total': 3022608, 'tokens/trainable': 45678, 'epoch': '0.3906'}
-
 20%|█▉        | 100/512 [11:17<47:04,  6.86s/it]
 20%|█▉        | 101/512 [11:24<46:33,  6.80s/it]
                                                 
{'loss': '0.004096', 'grad_norm': '0.2531', 'learning_rate': '9.484e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 3052992, 'tokens/trainable': 46125, 'epoch': '0.3945'}
-
 20%|█▉        | 101/512 [11:24<46:33,  6.80s/it]
 20%|█▉        | 102/512 [11:30<46:21,  6.78s/it]
                                                 
{'loss': '0.001389', 'grad_norm': '0.1501', 'learning_rate': '9.47e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '39.71', 'tokens/total': 3083392, 'tokens/trainable': 46612, 'epoch': '0.3984'}
-
 20%|█▉        | 102/512 [11:30<46:21,  6.78s/it]
 20%|██        | 103/512 [11:37<46:07,  6.77s/it]
                                                 
{'loss': '0.02253', 'grad_norm': '0.577', 'learning_rate': '9.456e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '32.57', 'tokens/total': 3113744, 'tokens/trainable': 47059, 'epoch': '0.4023'}
-
 20%|██        | 103/512 [11:37<46:07,  6.77s/it]
 20%|██        | 104/512 [11:44<46:10,  6.79s/it]
                                                 
{'loss': '0.01689', 'grad_norm': '0.5657', 'learning_rate': '9.442e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 3144272, 'tokens/trainable': 47536, 'epoch': '0.4062'}
-
 20%|██        | 104/512 [11:44<46:10,  6.79s/it]
 21%|██        | 105/512 [11:51<45:54,  6.77s/it]
                                                 
{'loss': '0.01797', 'grad_norm': '0.2961', 'learning_rate': '9.428e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 3174512, 'tokens/trainable': 48037, 'epoch': '0.4102'}
-
 21%|██        | 105/512 [11:51<45:54,  6.77s/it]
 21%|██        | 106/512 [11:57<45:36,  6.74s/it]
                                                 
{'loss': '0.01619', 'grad_norm': '0.5242', 'learning_rate': '9.414e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 3204896, 'tokens/trainable': 48491, 'epoch': '0.4141'}
-
 21%|██        | 106/512 [11:57<45:36,  6.74s/it]
 21%|██        | 107/512 [12:04<45:32,  6.75s/it]
                                                 
{'loss': '0.01091', 'grad_norm': '0.3886', 'learning_rate': '9.4e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 3235360, 'tokens/trainable': 48940, 'epoch': '0.418'}
-
 21%|██        | 107/512 [12:04<45:32,  6.75s/it]
 21%|██        | 108/512 [12:11<45:16,  6.72s/it]
                                                 
{'loss': '0.008071', 'grad_norm': '0.2443', 'learning_rate': '9.385e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 3265552, 'tokens/trainable': 49389, 'epoch': '0.4219'}
-
 21%|██        | 108/512 [12:11<45:16,  6.72s/it]
 21%|██▏       | 109/512 [12:17<43:58,  6.55s/it]
                                                 
{'loss': '0.003786', 'grad_norm': '0.1457', 'learning_rate': '9.37e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '36.01', 'tokens/total': 3293856, 'tokens/trainable': 49838, 'epoch': '0.4258'}
-
 21%|██▏       | 109/512 [12:17<43:58,  6.55s/it]
 21%|██▏       | 110/512 [12:24<44:10,  6.59s/it]
                                                 
{'loss': '0.006992', 'grad_norm': '0.3956', 'learning_rate': '9.355e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 3323984, 'tokens/trainable': 50310, 'epoch': '0.4297'}
-
 21%|██▏       | 110/512 [12:24<44:10,  6.59s/it]
 22%|██▏       | 111/512 [12:30<44:11,  6.61s/it]
                                                 
{'loss': '0.01485', 'grad_norm': '0.6225', 'learning_rate': '9.34e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 3354320, 'tokens/trainable': 50755, 'epoch': '0.4336'}
-
 22%|██▏       | 111/512 [12:30<44:11,  6.61s/it]
 22%|██▏       | 112/512 [12:37<44:15,  6.64s/it]
                                                 
{'loss': '0.004368', 'grad_norm': '0.3701', 'learning_rate': '9.325e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.65', 'tokens/total': 3384880, 'tokens/trainable': 51213, 'epoch': '0.4375'}
-
 22%|██▏       | 112/512 [12:37<44:15,  6.64s/it]
 22%|██▏       | 113/512 [12:44<44:06,  6.63s/it]
                                                 
{'loss': '0.002598', 'grad_norm': '0.1353', 'learning_rate': '9.31e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.83', 'tokens/total': 3415104, 'tokens/trainable': 51660, 'epoch': '0.4414'}
-
 22%|██▏       | 113/512 [12:44<44:06,  6.63s/it]
 22%|██▏       | 114/512 [12:50<44:07,  6.65s/it]
                                                 
{'loss': '0.002497', 'grad_norm': '0.1898', 'learning_rate': '9.294e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.43', 'tokens/total': 3445504, 'tokens/trainable': 52133, 'epoch': '0.4453'}
-
 22%|██▏       | 114/512 [12:50<44:07,  6.65s/it]
 22%|██▏       | 115/512 [12:57<44:06,  6.67s/it]
                                                 
{'loss': '0.02949', 'grad_norm': '1.182', 'learning_rate': '9.278e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.47', 'tokens/total': 3475744, 'tokens/trainable': 52580, 'epoch': '0.4492'}
-
 22%|██▏       | 115/512 [12:57<44:06,  6.67s/it]
 23%|██▎       | 116/512 [13:04<44:06,  6.68s/it]
                                                 
{'loss': '0.009586', 'grad_norm': '0.4436', 'learning_rate': '9.263e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.83', 'tokens/total': 3506016, 'tokens/trainable': 53033, 'epoch': '0.4531'}
-
 23%|██▎       | 116/512 [13:04<44:06,  6.68s/it]
 23%|██▎       | 117/512 [13:10<44:04,  6.69s/it]
                                                 
{'loss': '0.0202', 'grad_norm': '0.5075', 'learning_rate': '9.247e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 3536256, 'tokens/trainable': 53517, 'epoch': '0.457'}
-
 23%|██▎       | 117/512 [13:10<44:04,  6.69s/it]
 23%|██▎       | 118/512 [13:17<43:49,  6.67s/it]
                                                 
{'loss': '0.005689', 'grad_norm': '0.3565', 'learning_rate': '9.23e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 3566480, 'tokens/trainable': 53967, 'epoch': '0.4609'}
-
 23%|██▎       | 118/512 [13:17<43:49,  6.67s/it]
 23%|██▎       | 119/512 [13:24<43:45,  6.68s/it]
                                                 
{'loss': '0.003189', 'grad_norm': '0.1655', 'learning_rate': '9.214e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 3596624, 'tokens/trainable': 54430, 'epoch': '0.4648'}
-
 23%|██▎       | 119/512 [13:24<43:45,  6.68s/it]
 23%|██▎       | 120/512 [13:30<43:41,  6.69s/it]
                                                 
{'loss': '0.003235', 'grad_norm': '0.2072', 'learning_rate': '9.198e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 3627168, 'tokens/trainable': 54896, 'epoch': '0.4688'}
-
 23%|██▎       | 120/512 [13:30<43:41,  6.69s/it]
 24%|██▎       | 121/512 [13:37<43:39,  6.70s/it]
                                                 
{'loss': '0.005671', 'grad_norm': '0.4755', 'learning_rate': '9.181e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.77', 'tokens/total': 3657632, 'tokens/trainable': 55379, 'epoch': '0.4727'}
-
 24%|██▎       | 121/512 [13:37<43:39,  6.70s/it]
 24%|██▍       | 122/512 [13:44<43:24,  6.68s/it]
                                                 
{'loss': '0.0009903', 'grad_norm': '0.08305', 'learning_rate': '9.164e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.7', 'tokens/total': 3687824, 'tokens/trainable': 55803, 'epoch': '0.4766'}
-
 24%|██▍       | 122/512 [13:44<43:24,  6.68s/it]
 24%|██▍       | 123/512 [13:50<43:16,  6.68s/it]
                                                 
{'loss': '0.007413', 'grad_norm': '0.7253', 'learning_rate': '9.147e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '36.45', 'tokens/total': 3718320, 'tokens/trainable': 56249, 'epoch': '0.4805'}
-
 24%|██▍       | 123/512 [13:50<43:16,  6.68s/it]
 24%|██▍       | 124/512 [13:57<43:16,  6.69s/it]
                                                 
{'loss': '0.00297', 'grad_norm': '0.3141', 'learning_rate': '9.13e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.41', 'tokens/total': 3748672, 'tokens/trainable': 56713, 'epoch': '0.4844'}
-
 24%|██▍       | 124/512 [13:57<43:16,  6.69s/it]
 24%|██▍       | 125/512 [14:04<43:14,  6.70s/it]
                                                 
{'loss': '0.003796', 'grad_norm': '0.6083', 'learning_rate': '9.113e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '39.08', 'tokens/total': 3778976, 'tokens/trainable': 57196, 'epoch': '0.4883'}
-
 24%|██▍       | 125/512 [14:04<43:14,  6.70s/it]
 25%|██▍       | 126/512 [14:11<43:08,  6.71s/it]
                                                 
{'loss': '0.001475', 'grad_norm': '0.0604', 'learning_rate': '9.096e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '36.49', 'tokens/total': 3809296, 'tokens/trainable': 57680, 'epoch': '0.4922'}
-
 25%|██▍       | 126/512 [14:11<43:08,  6.71s/it]
 25%|██▍       | 127/512 [14:17<42:54,  6.69s/it]
                                                 
{'loss': '0.004137', 'grad_norm': '0.2111', 'learning_rate': '9.078e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.8', 'tokens/total': 3839328, 'tokens/trainable': 58110, 'epoch': '0.4961'}
-
 25%|██▍       | 127/512 [14:17<42:54,  6.69s/it]
 25%|██▌       | 128/512 [14:24<42:49,  6.69s/it]
                                                 
{'loss': '0.007372', 'grad_norm': '0.3234', 'learning_rate': '9.06e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 3869824, 'tokens/trainable': 58534, 'epoch': '0.5'}
-
 25%|██▌       | 128/512 [14:24<42:49,  6.69s/it][2026-08-18 12:58:12,429] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
-
 25%|██▌       | 129/512 [14:33<46:35,  7.30s/it]
                                                 
{'loss': '0.00798', 'grad_norm': '0.5826', 'learning_rate': '9.043e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 3900080, 'tokens/trainable': 58998, 'epoch': '0.5039'}
-
 25%|██▌       | 129/512 [14:33<46:35,  7.30s/it]
 25%|██▌       | 130/512 [14:39<45:19,  7.12s/it]
                                                 
{'loss': '0.005638', 'grad_norm': '0.3385', 'learning_rate': '9.025e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.81', 'tokens/total': 3930560, 'tokens/trainable': 59448, 'epoch': '0.5078'}
-
 25%|██▌       | 130/512 [14:39<45:19,  7.12s/it]
 26%|██▌       | 131/512 [14:46<44:16,  6.97s/it]
                                                 
{'loss': '0.003793', 'grad_norm': '0.2839', 'learning_rate': '9.007e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.97', 'tokens/total': 3960496, 'tokens/trainable': 59906, 'epoch': '0.5117'}
-
 26%|██▌       | 131/512 [14:46<44:16,  6.97s/it]
 26%|██▌       | 132/512 [14:53<43:37,  6.89s/it]
                                                 
{'loss': '0.01023', 'grad_norm': '0.4598', 'learning_rate': '8.988e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 3990928, 'tokens/trainable': 60350, 'epoch': '0.5156'}
-
 26%|██▌       | 132/512 [14:53<43:37,  6.89s/it]
 26%|██▌       | 133/512 [14:59<43:08,  6.83s/it]
                                                 
{'loss': '0.009468', 'grad_norm': '0.7073', 'learning_rate': '8.97e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 4021264, 'tokens/trainable': 60831, 'epoch': '0.5195'}
-
 26%|██▌       | 133/512 [14:59<43:08,  6.83s/it]
 26%|██▌       | 134/512 [15:06<42:46,  6.79s/it]
                                                 
{'loss': '0.006774', 'grad_norm': '0.3642', 'learning_rate': '8.951e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 4051728, 'tokens/trainable': 61305, 'epoch': '0.5234'}
-
 26%|██▌       | 134/512 [15:06<42:46,  6.79s/it]
 26%|██▋       | 135/512 [15:13<42:33,  6.77s/it]
                                                 
{'loss': '0.002285', 'grad_norm': '0.1071', 'learning_rate': '8.933e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.57', 'tokens/total': 4081920, 'tokens/trainable': 61792, 'epoch': '0.5273'}
-
 26%|██▋       | 135/512 [15:13<42:33,  6.77s/it]
 27%|██▋       | 136/512 [15:20<42:22,  6.76s/it]
                                                 
{'loss': '0.001668', 'grad_norm': '0.2247', 'learning_rate': '8.914e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.33', 'tokens/total': 4112304, 'tokens/trainable': 62253, 'epoch': '0.5312'}
-
 27%|██▋       | 136/512 [15:20<42:22,  6.76s/it]
 27%|██▋       | 137/512 [15:26<42:10,  6.75s/it]
                                                 
{'loss': '0.01744', 'grad_norm': '0.5423', 'learning_rate': '8.895e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 4142512, 'tokens/trainable': 62707, 'epoch': '0.5352'}
-
 27%|██▋       | 137/512 [15:26<42:10,  6.75s/it]
 27%|██▋       | 138/512 [15:33<41:53,  6.72s/it]
                                                 
{'loss': '0.01526', 'grad_norm': '0.4749', 'learning_rate': '8.876e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '29.32', 'tokens/total': 4172992, 'tokens/trainable': 63145, 'epoch': '0.5391'}
-
 27%|██▋       | 138/512 [15:33<41:53,  6.72s/it]
 27%|██▋       | 139/512 [15:40<41:44,  6.72s/it]
                                                 
{'loss': '0.00106', 'grad_norm': '0.05535', 'learning_rate': '8.856e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 4203136, 'tokens/trainable': 63587, 'epoch': '0.543'}
-
 27%|██▋       | 139/512 [15:40<41:44,  6.72s/it]
 27%|██▋       | 140/512 [15:46<41:35,  6.71s/it]
                                                 
{'loss': '0.002773', 'grad_norm': '0.1527', 'learning_rate': '8.837e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '36.75', 'tokens/total': 4233456, 'tokens/trainable': 64052, 'epoch': '0.5469'}
-
 27%|██▋       | 140/512 [15:46<41:35,  6.71s/it]
 28%|██▊       | 141/512 [15:53<41:27,  6.71s/it]
                                                 
{'loss': '0.00517', 'grad_norm': '0.2169', 'learning_rate': '8.817e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '34.36', 'tokens/total': 4263728, 'tokens/trainable': 64526, 'epoch': '0.5508'}
-
 28%|██▊       | 141/512 [15:53<41:27,  6.71s/it]
 28%|██▊       | 142/512 [16:00<41:32,  6.74s/it]
                                                 
{'loss': '0.003176', 'grad_norm': '0.1491', 'learning_rate': '8.798e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '34.66', 'tokens/total': 4294432, 'tokens/trainable': 64983, 'epoch': '0.5547'}
-
 28%|██▊       | 142/512 [16:00<41:32,  6.74s/it]
 28%|██▊       | 143/512 [16:07<41:31,  6.75s/it]
                                                 
{'loss': '0.003301', 'grad_norm': '0.1937', 'learning_rate': '8.778e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '37.6', 'tokens/total': 4324960, 'tokens/trainable': 65492, 'epoch': '0.5586'}
-
 28%|██▊       | 143/512 [16:07<41:31,  6.75s/it]
 28%|██▊       | 144/512 [16:13<41:11,  6.72s/it]
                                                 
{'loss': '0.004676', 'grad_norm': '0.4305', 'learning_rate': '8.758e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '35.71', 'tokens/total': 4355424, 'tokens/trainable': 65925, 'epoch': '0.5625'}
-
 28%|██▊       | 144/512 [16:13<41:11,  6.72s/it]
 28%|██▊       | 145/512 [16:20<41:06,  6.72s/it]
                                                 
{'loss': '0.005307', 'grad_norm': '1.153', 'learning_rate': '8.738e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.24', 'tokens/total': 4385712, 'tokens/trainable': 66383, 'epoch': '0.5664'}
-
 28%|██▊       | 145/512 [16:20<41:06,  6.72s/it]
 29%|██▊       | 146/512 [16:27<40:56,  6.71s/it]
                                                 
{'loss': '0.01282', 'grad_norm': '0.774', 'learning_rate': '8.718e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '31.37', 'tokens/total': 4416016, 'tokens/trainable': 66840, 'epoch': '0.5703'}
-
 29%|██▊       | 146/512 [16:27<40:56,  6.71s/it]
 29%|██▊       | 147/512 [16:33<40:47,  6.71s/it]
                                                 
{'loss': '0.02721', 'grad_norm': '0.6979', 'learning_rate': '8.697e-05', 'ppl': '1.028', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 4446368, 'tokens/trainable': 67297, 'epoch': '0.5742'}
-
 29%|██▊       | 147/512 [16:33<40:47,  6.71s/it]
 29%|██▉       | 148/512 [16:40<40:42,  6.71s/it]
                                                 
{'loss': '0.008016', 'grad_norm': '0.1734', 'learning_rate': '8.677e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '37.36', 'tokens/total': 4476880, 'tokens/trainable': 67769, 'epoch': '0.5781'}
-
 29%|██▉       | 148/512 [16:40<40:42,  6.71s/it]
 29%|██▉       | 149/512 [16:47<40:35,  6.71s/it]
                                                 
{'loss': '0.01083', 'grad_norm': '0.6356', 'learning_rate': '8.656e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 4507232, 'tokens/trainable': 68257, 'epoch': '0.582'}
-
 29%|██▉       | 149/512 [16:47<40:35,  6.71s/it]
 29%|██▉       | 150/512 [16:53<40:22,  6.69s/it]
                                                 
{'loss': '0.005296', 'grad_norm': '0.2509', 'learning_rate': '8.635e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 4537376, 'tokens/trainable': 68706, 'epoch': '0.5859'}
-
 29%|██▉       | 150/512 [16:53<40:22,  6.69s/it]
 29%|██▉       | 151/512 [17:00<40:14,  6.69s/it]
                                                 
{'loss': '0.02201', 'grad_norm': '0.5293', 'learning_rate': '8.615e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.25', 'tokens/total': 4567584, 'tokens/trainable': 69131, 'epoch': '0.5898'}
-
 29%|██▉       | 151/512 [17:00<40:14,  6.69s/it]
 30%|██▉       | 152/512 [17:07<40:02,  6.67s/it]
                                                 
{'loss': '0.02459', 'grad_norm': '0.3471', 'learning_rate': '8.594e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.92', 'tokens/total': 4597792, 'tokens/trainable': 69586, 'epoch': '0.5938'}
-
 30%|██▉       | 152/512 [17:07<40:02,  6.67s/it]
 30%|██▉       | 153/512 [17:13<39:59,  6.68s/it]
                                                 
{'loss': '0.002612', 'grad_norm': '0.08057', 'learning_rate': '8.572e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 4627936, 'tokens/trainable': 70061, 'epoch': '0.5977'}
-
 30%|██▉       | 153/512 [17:13<39:59,  6.68s/it]
 30%|███       | 154/512 [17:20<39:50,  6.68s/it]
                                                 
{'loss': '0.003361', 'grad_norm': '0.1623', 'learning_rate': '8.551e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '27.93', 'tokens/total': 4658352, 'tokens/trainable': 70467, 'epoch': '0.6016'}
-
 30%|███       | 154/512 [17:20<39:50,  6.68s/it]
 30%|███       | 155/512 [17:27<39:47,  6.69s/it]
                                                 
{'loss': '0.00435', 'grad_norm': '0.1761', 'learning_rate': '8.53e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 4688896, 'tokens/trainable': 70933, 'epoch': '0.6055'}
-
 30%|███       | 155/512 [17:27<39:47,  6.69s/it]
 30%|███       | 156/512 [17:34<39:44,  6.70s/it]
                                                 
{'loss': '0.01062', 'grad_norm': '0.2591', 'learning_rate': '8.508e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 4719168, 'tokens/trainable': 71382, 'epoch': '0.6094'}
-
 30%|███       | 156/512 [17:34<39:44,  6.70s/it]
 31%|███       | 157/512 [17:40<39:39,  6.70s/it]
                                                 
{'loss': '0.002628', 'grad_norm': '0.1564', 'learning_rate': '8.487e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '30.22', 'tokens/total': 4749136, 'tokens/trainable': 71808, 'epoch': '0.6133'}
-
 31%|███       | 157/512 [17:40<39:39,  6.70s/it]
 31%|███       | 158/512 [17:47<39:32,  6.70s/it]
                                                 
{'loss': '0.01087', 'grad_norm': '0.3443', 'learning_rate': '8.465e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 4779472, 'tokens/trainable': 72249, 'epoch': '0.6172'}
-
 31%|███       | 158/512 [17:47<39:32,  6.70s/it]
 31%|███       | 159/512 [17:54<39:23,  6.69s/it]
                                                 
{'loss': '0.02005', 'grad_norm': '0.3859', 'learning_rate': '8.443e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 4809632, 'tokens/trainable': 72726, 'epoch': '0.6211'}
-
 31%|███       | 159/512 [17:54<39:23,  6.69s/it]
 31%|███▏      | 160/512 [18:00<39:13,  6.68s/it]
                                                 
{'loss': '0.009756', 'grad_norm': '0.3002', 'learning_rate': '8.421e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 4840112, 'tokens/trainable': 73148, 'epoch': '0.625'}
-
 31%|███▏      | 160/512 [18:00<39:13,  6.68s/it][2026-08-18 13:01:48,787] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
-
 31%|███▏      | 161/512 [18:09<42:32,  7.27s/it]
                                                 
{'loss': '0.009085', 'grad_norm': '0.2029', 'learning_rate': '8.399e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 4870656, 'tokens/trainable': 73600, 'epoch': '0.6289'}
-
 31%|███▏      | 161/512 [18:09<42:32,  7.27s/it]
 32%|███▏      | 162/512 [18:16<41:15,  7.07s/it]
                                                 
{'loss': '0.01259', 'grad_norm': '0.2961', 'learning_rate': '8.376e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.71', 'tokens/total': 4900608, 'tokens/trainable': 74068, 'epoch': '0.6328'}
-
 32%|███▏      | 162/512 [18:16<41:15,  7.07s/it]
 32%|███▏      | 163/512 [18:22<40:25,  6.95s/it]
                                                 
{'loss': '0.003114', 'grad_norm': '0.1145', 'learning_rate': '8.354e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.71', 'tokens/total': 4930752, 'tokens/trainable': 74527, 'epoch': '0.6367'}
-
 32%|███▏      | 163/512 [18:22<40:25,  6.95s/it]
 32%|███▏      | 164/512 [18:29<39:49,  6.87s/it]
                                                 
{'loss': '0.005665', 'grad_norm': '0.1261', 'learning_rate': '8.332e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '31.81', 'tokens/total': 4961008, 'tokens/trainable': 74992, 'epoch': '0.6406'}
-
 32%|███▏      | 164/512 [18:29<39:49,  6.87s/it]
 32%|███▏      | 165/512 [18:36<39:24,  6.81s/it]
                                                 
{'loss': '0.01254', 'grad_norm': '0.4959', 'learning_rate': '8.309e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 4991200, 'tokens/trainable': 75442, 'epoch': '0.6445'}
-
 32%|███▏      | 165/512 [18:36<39:24,  6.81s/it]
 32%|███▏      | 166/512 [18:42<39:09,  6.79s/it]
                                                 
{'loss': '0.007678', 'grad_norm': '0.1712', 'learning_rate': '8.286e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 5021600, 'tokens/trainable': 75887, 'epoch': '0.6484'}
-
 32%|███▏      | 166/512 [18:42<39:09,  6.79s/it]
 33%|███▎      | 167/512 [18:49<38:51,  6.76s/it]
                                                 
{'loss': '0.007361', 'grad_norm': '0.1903', 'learning_rate': '8.263e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 5052032, 'tokens/trainable': 76331, 'epoch': '0.6523'}
-
 33%|███▎      | 167/512 [18:49<38:51,  6.76s/it]
 33%|███▎      | 168/512 [18:55<37:45,  6.59s/it]
                                                 
{'loss': '0.001689', 'grad_norm': '0.08689', 'learning_rate': '8.24e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.9', 'tokens/total': 5080256, 'tokens/trainable': 76745, 'epoch': '0.6562'}
-
 33%|███▎      | 168/512 [18:55<37:45,  6.59s/it]
 33%|███▎      | 169/512 [19:02<37:45,  6.61s/it]
                                                 
{'loss': '0.02567', 'grad_norm': '1.222', 'learning_rate': '8.217e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.3', 'tokens/total': 5110784, 'tokens/trainable': 77201, 'epoch': '0.6602'}
-
 33%|███▎      | 169/512 [19:02<37:45,  6.61s/it]
 33%|███▎      | 170/512 [19:08<37:47,  6.63s/it]
                                                 
{'loss': '0.005606', 'grad_norm': '0.2243', 'learning_rate': '8.194e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.38', 'tokens/total': 5141200, 'tokens/trainable': 77655, 'epoch': '0.6641'}
-
 33%|███▎      | 170/512 [19:08<37:47,  6.63s/it]
 33%|███▎      | 171/512 [19:15<37:45,  6.64s/it]
                                                 
{'loss': '0.001366', 'grad_norm': '0.06314', 'learning_rate': '8.171e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '39.25', 'tokens/total': 5171760, 'tokens/trainable': 78122, 'epoch': '0.668'}
-
 33%|███▎      | 171/512 [19:15<37:45,  6.64s/it]
 34%|███▎      | 172/512 [19:22<37:44,  6.66s/it]
                                                 
{'loss': '0.008099', 'grad_norm': '0.3466', 'learning_rate': '8.147e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.59', 'tokens/total': 5202272, 'tokens/trainable': 78576, 'epoch': '0.6719'}
-
 34%|███▎      | 172/512 [19:22<37:44,  6.66s/it]
 34%|███▍      | 173/512 [19:29<37:40,  6.67s/it]
                                                 
{'loss': '0.02078', 'grad_norm': '0.4534', 'learning_rate': '8.124e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '36.49', 'tokens/total': 5232800, 'tokens/trainable': 79059, 'epoch': '0.6758'}
-
 34%|███▍      | 173/512 [19:29<37:40,  6.67s/it]
 34%|███▍      | 174/512 [19:35<37:26,  6.65s/it]
                                                 
{'loss': '0.01096', 'grad_norm': '0.3241', 'learning_rate': '8.1e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '29.73', 'tokens/total': 5262672, 'tokens/trainable': 79498, 'epoch': '0.6797'}
-
 34%|███▍      | 174/512 [19:35<37:26,  6.65s/it]
 34%|███▍      | 175/512 [19:42<37:18,  6.64s/it]
                                                 
{'loss': '0.009046', 'grad_norm': '0.5539', 'learning_rate': '8.076e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '30.45', 'tokens/total': 5293072, 'tokens/trainable': 79922, 'epoch': '0.6836'}
-
 34%|███▍      | 175/512 [19:42<37:18,  6.64s/it]
 34%|███▍      | 176/512 [19:48<36:24,  6.50s/it]
                                                 
{'loss': '0.01483', 'grad_norm': '0.2632', 'learning_rate': '8.053e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.67', 'tokens/total': 5321520, 'tokens/trainable': 80383, 'epoch': '0.6875'}
-
 34%|███▍      | 176/512 [19:48<36:24,  6.50s/it]
 35%|███▍      | 177/512 [19:55<36:35,  6.55s/it]
                                                 
{'loss': '0.006698', 'grad_norm': '0.2084', 'learning_rate': '8.029e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.09', 'tokens/total': 5351904, 'tokens/trainable': 80824, 'epoch': '0.6914'}
-
 35%|███▍      | 177/512 [19:55<36:35,  6.55s/it]
 35%|███▍      | 178/512 [20:01<36:37,  6.58s/it]
                                                 
{'loss': '0.00976', 'grad_norm': '0.3557', 'learning_rate': '8.005e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '32.27', 'tokens/total': 5382432, 'tokens/trainable': 81243, 'epoch': '0.6953'}
-
 35%|███▍      | 178/512 [20:01<36:37,  6.58s/it]
 35%|███▍      | 179/512 [20:08<36:39,  6.60s/it]
                                                 
{'loss': '0.009939', 'grad_norm': '0.2144', 'learning_rate': '7.98e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.36', 'tokens/total': 5412960, 'tokens/trainable': 81716, 'epoch': '0.6992'}
-
 35%|███▍      | 179/512 [20:08<36:39,  6.60s/it]
 35%|███▌      | 180/512 [20:15<36:43,  6.64s/it]
                                                 
{'loss': '0.005616', 'grad_norm': '0.1301', 'learning_rate': '7.956e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.61', 'tokens/total': 5443232, 'tokens/trainable': 82181, 'epoch': '0.7031'}
-
 35%|███▌      | 180/512 [20:15<36:43,  6.64s/it]
 35%|███▌      | 181/512 [20:21<35:51,  6.50s/it]
                                                 
{'loss': '0.007541', 'grad_norm': '0.2382', 'learning_rate': '7.932e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.29', 'memory/max_allocated (GiB)': '33.29', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.63', 'tokens/total': 5471440, 'tokens/trainable': 82626, 'epoch': '0.707'}
-
 35%|███▌      | 181/512 [20:21<35:51,  6.50s/it]
 36%|███▌      | 182/512 [20:28<36:03,  6.56s/it]
                                                 
{'loss': '0.003529', 'grad_norm': '0.1383', 'learning_rate': '7.907e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.17', 'tokens/total': 5501744, 'tokens/trainable': 83089, 'epoch': '0.7109'}
-
 36%|███▌      | 182/512 [20:28<36:03,  6.56s/it]
 36%|███▌      | 183/512 [20:34<36:11,  6.60s/it]
                                                 
{'loss': '0.002233', 'grad_norm': '0.08244', 'learning_rate': '7.883e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 5532272, 'tokens/trainable': 83590, 'epoch': '0.7148'}
-
 36%|███▌      | 183/512 [20:34<36:11,  6.60s/it]
 36%|███▌      | 184/512 [20:41<36:16,  6.64s/it]
                                                 
{'loss': '0.0112', 'grad_norm': '0.2647', 'learning_rate': '7.858e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 5562784, 'tokens/trainable': 84046, 'epoch': '0.7188'}
-
 36%|███▌      | 184/512 [20:41<36:16,  6.64s/it]
 36%|███▌      | 185/512 [20:48<36:10,  6.64s/it]
                                                 
{'loss': '0.0033', 'grad_norm': '0.1043', 'learning_rate': '7.833e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 5592800, 'tokens/trainable': 84517, 'epoch': '0.7227'}
-
 36%|███▌      | 185/512 [20:48<36:10,  6.64s/it]
 36%|███▋      | 186/512 [20:54<36:13,  6.67s/it]
                                                 
{'loss': '0.03358', 'grad_norm': '0.6546', 'learning_rate': '7.808e-05', 'ppl': '1.034', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 5623088, 'tokens/trainable': 84974, 'epoch': '0.7266'}
-
 36%|███▋      | 186/512 [20:54<36:13,  6.67s/it]
 37%|███▋      | 187/512 [21:01<36:13,  6.69s/it]
                                                 
{'loss': '0.0004142', 'grad_norm': '0.0167', 'learning_rate': '7.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 5653344, 'tokens/trainable': 85460, 'epoch': '0.7305'}
-
 37%|███▋      | 187/512 [21:01<36:13,  6.69s/it]
 37%|███▋      | 188/512 [21:08<36:09,  6.70s/it]
                                                 
{'loss': '0.002929', 'grad_norm': '0.1661', 'learning_rate': '7.758e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.9', 'tokens/total': 5683600, 'tokens/trainable': 85939, 'epoch': '0.7344'}
-
 37%|███▋      | 188/512 [21:08<36:09,  6.70s/it]
 37%|███▋      | 189/512 [21:14<35:11,  6.54s/it]
                                                 
{'loss': '0.00125', 'grad_norm': '0.3206', 'learning_rate': '7.733e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '38.26', 'tokens/total': 5711952, 'tokens/trainable': 86377, 'epoch': '0.7383'}
-
 37%|███▋      | 189/512 [21:14<35:11,  6.54s/it]
 37%|███▋      | 190/512 [21:21<35:21,  6.59s/it]
                                                 
{'loss': '0.009377', 'grad_norm': '0.5194', 'learning_rate': '7.708e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 5742336, 'tokens/trainable': 86834, 'epoch': '0.7422'}
-
 37%|███▋      | 190/512 [21:21<35:21,  6.59s/it]
 37%|███▋      | 191/512 [21:27<35:20,  6.60s/it]
                                                 
{'loss': '0.009413', 'grad_norm': '0.6359', 'learning_rate': '7.683e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '32.08', 'tokens/total': 5772560, 'tokens/trainable': 87265, 'epoch': '0.7461'}
-
 37%|███▋      | 191/512 [21:27<35:20,  6.60s/it]
 38%|███▊      | 192/512 [21:34<35:28,  6.65s/it]
                                                 
{'loss': '0.001757', 'grad_norm': '0.1094', 'learning_rate': '7.657e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 5803136, 'tokens/trainable': 87747, 'epoch': '0.75'}
-
 38%|███▊      | 192/512 [21:34<35:28,  6.65s/it][2026-08-18 13:05:22,559] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
-
 38%|███▊      | 193/512 [21:43<38:42,  7.28s/it]
                                                 
{'loss': '0.001762', 'grad_norm': '0.08651', 'learning_rate': '7.632e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 5833440, 'tokens/trainable': 88239, 'epoch': '0.7539'}
-
 38%|███▊      | 193/512 [21:43<38:42,  7.28s/it]
 38%|███▊      | 194/512 [21:49<37:40,  7.11s/it]
                                                 
{'loss': '0.0002287', 'grad_norm': '0.01772', 'learning_rate': '7.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 5863552, 'tokens/trainable': 88718, 'epoch': '0.7578'}
-
 38%|███▊      | 194/512 [21:49<37:40,  7.11s/it]
 38%|███▊      | 195/512 [21:56<36:49,  6.97s/it]
                                                 
{'loss': '0.003605', 'grad_norm': '0.2394', 'learning_rate': '7.58e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '31.27', 'tokens/total': 5893680, 'tokens/trainable': 89129, 'epoch': '0.7617'}
-
 38%|███▊      | 195/512 [21:56<36:49,  6.97s/it]
 38%|███▊      | 196/512 [22:03<36:14,  6.88s/it]
                                                 
{'loss': '0.0001226', 'grad_norm': '0.006238', 'learning_rate': '7.555e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '32.25', 'tokens/total': 5923744, 'tokens/trainable': 89567, 'epoch': '0.7656'}
-
 38%|███▊      | 196/512 [22:03<36:14,  6.88s/it]
 38%|███▊      | 197/512 [22:09<35:48,  6.82s/it]
                                                 
{'loss': '0.03544', 'grad_norm': '0.9131', 'learning_rate': '7.529e-05', 'ppl': '1.036', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '30.25', 'tokens/total': 5954128, 'tokens/trainable': 89988, 'epoch': '0.7695'}
-
 38%|███▊      | 197/512 [22:09<35:48,  6.82s/it]
 39%|███▊      | 198/512 [22:16<35:30,  6.79s/it]
                                                 
{'loss': '0.001015', 'grad_norm': '0.1169', 'learning_rate': '7.503e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 5984352, 'tokens/trainable': 90434, 'epoch': '0.7734'}
-
 39%|███▊      | 198/512 [22:16<35:30,  6.79s/it]
 39%|███▉      | 199/512 [22:23<35:16,  6.76s/it]
                                                 
{'loss': '0.009798', 'grad_norm': '0.5136', 'learning_rate': '7.477e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.61', 'tokens/train_per_sec_per_gpu': '34.17', 'tokens/total': 6014704, 'tokens/trainable': 90913, 'epoch': '0.7773'}
-
 39%|███▉      | 199/512 [22:23<35:16,  6.76s/it]
 39%|███▉      | 200/512 [22:30<34:58,  6.73s/it]
                                                 
{'loss': '0.009519', 'grad_norm': '0.267', 'learning_rate': '7.451e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.17', 'tokens/total': 6045056, 'tokens/trainable': 91355, 'epoch': '0.7812'}
-
 39%|███▉      | 200/512 [22:30<34:58,  6.73s/it]
 39%|███▉      | 201/512 [22:36<34:58,  6.75s/it]
                                                 
{'loss': '0.006804', 'grad_norm': '0.315', 'learning_rate': '7.424e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '37.38', 'tokens/total': 6075504, 'tokens/trainable': 91856, 'epoch': '0.7852'}
-
 39%|███▉      | 201/512 [22:36<34:58,  6.75s/it]
 39%|███▉      | 202/512 [22:43<34:43,  6.72s/it]
                                                 
{'loss': '0.01742', 'grad_norm': '0.5194', 'learning_rate': '7.398e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 6105904, 'tokens/trainable': 92311, 'epoch': '0.7891'}
-
 39%|███▉      | 202/512 [22:43<34:43,  6.72s/it]
 40%|███▉      | 203/512 [22:50<34:38,  6.73s/it]
                                                 
{'loss': '0.002557', 'grad_norm': '0.1296', 'learning_rate': '7.372e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.55', 'tokens/total': 6136272, 'tokens/trainable': 92764, 'epoch': '0.793'}
-
 40%|███▉      | 203/512 [22:50<34:38,  6.73s/it]
 40%|███▉      | 204/512 [22:56<34:24,  6.70s/it]
                                                 
{'loss': '0.00419', 'grad_norm': '0.1787', 'learning_rate': '7.345e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 6166336, 'tokens/trainable': 93227, 'epoch': '0.7969'}
-
 40%|███▉      | 204/512 [22:56<34:24,  6.70s/it]
 40%|████      | 205/512 [23:03<34:12,  6.69s/it]
                                                 
{'loss': '0.0102', 'grad_norm': '0.3807', 'learning_rate': '7.319e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 6196720, 'tokens/trainable': 93675, 'epoch': '0.8008'}
-
 40%|████      | 205/512 [23:03<34:12,  6.69s/it]
 40%|████      | 206/512 [23:10<34:11,  6.70s/it]
                                                 
{'loss': '0.008401', 'grad_norm': '0.3957', 'learning_rate': '7.292e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.33', 'tokens/total': 6227424, 'tokens/trainable': 94120, 'epoch': '0.8047'}
-
 40%|████      | 206/512 [23:10<34:11,  6.70s/it]
 40%|████      | 207/512 [23:16<34:02,  6.70s/it]
                                                 
{'loss': '0.002036', 'grad_norm': '0.1237', 'learning_rate': '7.266e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 6257616, 'tokens/trainable': 94557, 'epoch': '0.8086'}
-
 40%|████      | 207/512 [23:16<34:02,  6.70s/it]
 41%|████      | 208/512 [23:23<33:59,  6.71s/it]
                                                 
{'loss': '0.01048', 'grad_norm': '0.2043', 'learning_rate': '7.239e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '40.52', 'tokens/total': 6288000, 'tokens/trainable': 95035, 'epoch': '0.8125'}
-
 41%|████      | 208/512 [23:23<33:59,  6.71s/it]
 41%|████      | 209/512 [23:30<33:52,  6.71s/it]
                                                 
{'loss': '0.001463', 'grad_norm': '0.06323', 'learning_rate': '7.212e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '37.41', 'tokens/total': 6318336, 'tokens/trainable': 95517, 'epoch': '0.8164'}
-
 41%|████      | 209/512 [23:30<33:52,  6.71s/it]
 41%|████      | 210/512 [23:37<33:47,  6.71s/it]
                                                 
{'loss': '0.007305', 'grad_norm': '0.1743', 'learning_rate': '7.185e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.31', 'tokens/total': 6348656, 'tokens/trainable': 96026, 'epoch': '0.8203'}
-
 41%|████      | 210/512 [23:37<33:47,  6.71s/it]
 41%|████      | 211/512 [23:43<33:38,  6.71s/it]
                                                 
{'loss': '0.002277', 'grad_norm': '0.08656', 'learning_rate': '7.158e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.44', 'tokens/total': 6379040, 'tokens/trainable': 96477, 'epoch': '0.8242'}
-
 41%|████      | 211/512 [23:43<33:38,  6.71s/it]
 41%|████▏     | 212/512 [23:50<33:30,  6.70s/it]
                                                 
{'loss': '0.003339', 'grad_norm': '0.1084', 'learning_rate': '7.131e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 6409312, 'tokens/trainable': 96951, 'epoch': '0.8281'}
-
 41%|████▏     | 212/512 [23:50<33:30,  6.70s/it]
 42%|████▏     | 213/512 [23:57<33:22,  6.70s/it]
                                                 
{'loss': '0.00445', 'grad_norm': '0.2011', 'learning_rate': '7.104e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '30.68', 'tokens/total': 6439520, 'tokens/trainable': 97350, 'epoch': '0.832'}
-
 42%|████▏     | 213/512 [23:57<33:22,  6.70s/it]
 42%|████▏     | 214/512 [24:03<33:14,  6.69s/it]
                                                 
{'loss': '0.001114', 'grad_norm': '0.08952', 'learning_rate': '7.077e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '29.79', 'tokens/total': 6469888, 'tokens/trainable': 97792, 'epoch': '0.8359'}
-
 42%|████▏     | 214/512 [24:03<33:14,  6.69s/it]
 42%|████▏     | 215/512 [24:10<33:10,  6.70s/it]
                                                 
{'loss': '0.002481', 'grad_norm': '0.2406', 'learning_rate': '7.05e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 6500128, 'tokens/trainable': 98257, 'epoch': '0.8398'}
-
 42%|████▏     | 215/512 [24:10<33:10,  6.70s/it]
 42%|████▏     | 216/512 [24:17<33:13,  6.73s/it]
                                                 
{'loss': '0.005798', 'grad_norm': '0.2198', 'learning_rate': '7.022e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.46', 'tokens/total': 6530832, 'tokens/trainable': 98731, 'epoch': '0.8438'}
-
 42%|████▏     | 216/512 [24:17<33:13,  6.73s/it]
 42%|████▏     | 217/512 [24:24<33:04,  6.73s/it]
                                                 
{'loss': '0.0002352', 'grad_norm': '0.02031', 'learning_rate': '6.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.37', 'tokens/total': 6561248, 'tokens/trainable': 99212, 'epoch': '0.8477'}
-
 42%|████▏     | 217/512 [24:24<33:04,  6.73s/it]
 43%|████▎     | 218/512 [24:30<32:50,  6.70s/it]
                                                 
{'loss': '0.0007598', 'grad_norm': '0.04731', 'learning_rate': '6.968e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 6591648, 'tokens/trainable': 99654, 'epoch': '0.8516'}
-
 43%|████▎     | 218/512 [24:30<32:50,  6.70s/it]
 43%|████▎     | 219/512 [24:37<32:48,  6.72s/it]
                                                 
{'loss': '0.006445', 'grad_norm': '0.3725', 'learning_rate': '6.94e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 6622368, 'tokens/trainable': 100086, 'epoch': '0.8555'}
-
 43%|████▎     | 219/512 [24:37<32:48,  6.72s/it]
 43%|████▎     | 220/512 [24:44<32:39,  6.71s/it]
                                                 
{'loss': '0.00061', 'grad_norm': '0.04439', 'learning_rate': '6.913e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 6652848, 'tokens/trainable': 100524, 'epoch': '0.8594'}
-
 43%|████▎     | 220/512 [24:44<32:39,  6.71s/it]
 43%|████▎     | 221/512 [24:51<32:39,  6.73s/it]
                                                 
{'loss': '0.01045', 'grad_norm': '0.4204', 'learning_rate': '6.885e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.27', 'tokens/total': 6683392, 'tokens/trainable': 100996, 'epoch': '0.8633'}
-
 43%|████▎     | 221/512 [24:51<32:39,  6.73s/it]
 43%|████▎     | 222/512 [24:57<32:28,  6.72s/it]
                                                 
{'loss': '0.0005217', 'grad_norm': '0.03999', 'learning_rate': '6.857e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '31.9', 'tokens/total': 6713712, 'tokens/trainable': 101449, 'epoch': '0.8672'}
-
 43%|████▎     | 222/512 [24:57<32:28,  6.72s/it]
 44%|████▎     | 223/512 [25:04<32:17,  6.70s/it]
                                                 
{'loss': '0.01121', 'grad_norm': '0.4914', 'learning_rate': '6.83e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 6744176, 'tokens/trainable': 101905, 'epoch': '0.8711'}
-
 44%|████▎     | 223/512 [25:04<32:17,  6.70s/it]
 44%|████▍     | 224/512 [25:11<32:13,  6.71s/it]
                                                 
{'loss': '0.001121', 'grad_norm': '0.1168', 'learning_rate': '6.802e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.36', 'tokens/total': 6774416, 'tokens/trainable': 102411, 'epoch': '0.875'}
-
 44%|████▍     | 224/512 [25:11<32:13,  6.71s/it][2026-08-18 13:08:59,122] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
-
 44%|████▍     | 225/512 [25:19<34:54,  7.30s/it]
                                                 
{'loss': '0.001277', 'grad_norm': '0.08471', 'learning_rate': '6.774e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 6804592, 'tokens/trainable': 102882, 'epoch': '0.8789'}
-
 44%|████▍     | 225/512 [25:19<34:54,  7.30s/it]
 44%|████▍     | 226/512 [25:26<33:54,  7.11s/it]
                                                 
{'loss': '0.0003806', 'grad_norm': '0.03987', 'learning_rate': '6.746e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '34.57', 'tokens/total': 6834976, 'tokens/trainable': 103332, 'epoch': '0.8828'}
-
 44%|████▍     | 226/512 [25:26<33:54,  7.11s/it]
 44%|████▍     | 227/512 [25:33<33:14,  7.00s/it]
                                                 
{'loss': '0.008083', 'grad_norm': '0.4415', 'learning_rate': '6.718e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 6865408, 'tokens/trainable': 103790, 'epoch': '0.8867'}
-
 44%|████▍     | 227/512 [25:33<33:14,  7.00s/it]
 45%|████▍     | 228/512 [25:39<32:39,  6.90s/it]
                                                 
{'loss': '0.004411', 'grad_norm': '0.2599', 'learning_rate': '6.69e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.35', 'tokens/total': 6895664, 'tokens/trainable': 104246, 'epoch': '0.8906'}
-
 45%|████▍     | 228/512 [25:39<32:39,  6.90s/it]
 45%|████▍     | 229/512 [25:46<32:15,  6.84s/it]
                                                 
{'loss': '0.01891', 'grad_norm': '0.5188', 'learning_rate': '6.662e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.66', 'tokens/total': 6926016, 'tokens/trainable': 104707, 'epoch': '0.8945'}
-
 45%|████▍     | 229/512 [25:46<32:15,  6.84s/it]
 45%|████▍     | 230/512 [25:53<31:57,  6.80s/it]
                                                 
{'loss': '0.006382', 'grad_norm': '0.2526', 'learning_rate': '6.634e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.83', 'tokens/total': 6956384, 'tokens/trainable': 105167, 'epoch': '0.8984'}
-
 45%|████▍     | 230/512 [25:53<31:57,  6.80s/it]
 45%|████▌     | 231/512 [25:59<31:37,  6.75s/it]
                                                 
{'loss': '0.0007957', 'grad_norm': '0.04988', 'learning_rate': '6.606e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 6986768, 'tokens/trainable': 105576, 'epoch': '0.9023'}
-
 45%|████▌     | 231/512 [25:59<31:37,  6.75s/it]
 45%|████▌     | 232/512 [26:06<31:23,  6.73s/it]
                                                 
{'loss': '0.0148', 'grad_norm': '0.4156', 'learning_rate': '6.578e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.76', 'tokens/total': 7017232, 'tokens/trainable': 106043, 'epoch': '0.9062'}
-
 45%|████▌     | 232/512 [26:06<31:23,  6.73s/it]
 46%|████▌     | 233/512 [26:13<31:14,  6.72s/it]
                                                 
{'loss': '0.003381', 'grad_norm': '0.16', 'learning_rate': '6.55e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 7047568, 'tokens/trainable': 106506, 'epoch': '0.9102'}
-
 46%|████▌     | 233/512 [26:13<31:14,  6.72s/it]
 46%|████▌     | 234/512 [26:19<31:00,  6.69s/it]
                                                 
{'loss': '0.0007064', 'grad_norm': '0.04383', 'learning_rate': '6.522e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 7077760, 'tokens/trainable': 106951, 'epoch': '0.9141'}
-
 46%|████▌     | 234/512 [26:19<31:00,  6.69s/it]
 46%|████▌     | 235/512 [26:26<30:45,  6.66s/it]
                                                 
{'loss': '0.003099', 'grad_norm': '0.133', 'learning_rate': '6.493e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.78', 'tokens/total': 7107760, 'tokens/trainable': 107382, 'epoch': '0.918'}
-
 46%|████▌     | 235/512 [26:26<30:45,  6.66s/it]
 46%|████▌     | 236/512 [26:33<30:40,  6.67s/it]
                                                 
{'loss': '0.00322', 'grad_norm': '0.3202', 'learning_rate': '6.465e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 7138144, 'tokens/trainable': 107814, 'epoch': '0.9219'}
-
 46%|████▌     | 236/512 [26:33<30:40,  6.67s/it]
 46%|████▋     | 237/512 [26:39<30:34,  6.67s/it]
                                                 
{'loss': '0.002884', 'grad_norm': '0.1548', 'learning_rate': '6.437e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 7168352, 'tokens/trainable': 108274, 'epoch': '0.9258'}
-
 46%|████▋     | 237/512 [26:39<30:34,  6.67s/it]
 46%|████▋     | 238/512 [26:46<30:28,  6.67s/it]
                                                 
{'loss': '0.002136', 'grad_norm': '0.09865', 'learning_rate': '6.408e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.22', 'tokens/total': 7198624, 'tokens/trainable': 108714, 'epoch': '0.9297'}
-
 46%|████▋     | 238/512 [26:46<30:28,  6.67s/it]
 47%|████▋     | 239/512 [26:53<30:22,  6.68s/it]
                                                 
{'loss': '0.001417', 'grad_norm': '0.06801', 'learning_rate': '6.38e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.79', 'tokens/total': 7229216, 'tokens/trainable': 109220, 'epoch': '0.9336'}
-
 47%|████▋     | 239/512 [26:53<30:22,  6.68s/it]
 47%|████▋     | 240/512 [26:59<30:15,  6.67s/it]
                                                 
{'loss': '0.002807', 'grad_norm': '0.217', 'learning_rate': '6.351e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.12', 'tokens/total': 7259648, 'tokens/trainable': 109672, 'epoch': '0.9375'}
-
 47%|████▋     | 240/512 [26:59<30:15,  6.67s/it]
 47%|████▋     | 241/512 [27:06<30:09,  6.68s/it]
                                                 
{'loss': '0.01761', 'grad_norm': '0.4365', 'learning_rate': '6.323e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 7289824, 'tokens/trainable': 110135, 'epoch': '0.9414'}
-
 47%|████▋     | 241/512 [27:06<30:09,  6.68s/it]
 47%|████▋     | 242/512 [27:13<30:05,  6.69s/it]
                                                 
{'loss': '0.0006151', 'grad_norm': '0.06252', 'learning_rate': '6.294e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 7320288, 'tokens/trainable': 110563, 'epoch': '0.9453'}
-
 47%|████▋     | 242/512 [27:13<30:05,  6.69s/it]
 47%|████▋     | 243/512 [27:19<29:58,  6.69s/it]
                                                 
{'loss': '0.003188', 'grad_norm': '0.1603', 'learning_rate': '6.266e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.12', 'tokens/total': 7350736, 'tokens/trainable': 111049, 'epoch': '0.9492'}
-
 47%|████▋     | 243/512 [27:19<29:58,  6.69s/it]
 48%|████▊     | 244/512 [27:26<30:00,  6.72s/it]
                                                 
{'loss': '0.0007044', 'grad_norm': '0.1788', 'learning_rate': '6.237e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 7381280, 'tokens/trainable': 111495, 'epoch': '0.9531'}
-
 48%|████▊     | 244/512 [27:26<30:00,  6.72s/it]
 48%|████▊     | 245/512 [27:33<29:51,  6.71s/it]
                                                 
{'loss': '0.0006958', 'grad_norm': '0.05235', 'learning_rate': '6.208e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.82', 'tokens/total': 7411632, 'tokens/trainable': 111968, 'epoch': '0.957'}
-
 48%|████▊     | 245/512 [27:33<29:51,  6.71s/it]
 48%|████▊     | 246/512 [27:40<29:38,  6.68s/it]
                                                 
{'loss': '0.008788', 'grad_norm': '0.1738', 'learning_rate': '6.18e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 7441904, 'tokens/trainable': 112416, 'epoch': '0.9609'}
-
 48%|████▊     | 246/512 [27:40<29:38,  6.68s/it]
 48%|████▊     | 247/512 [27:46<29:36,  6.71s/it]
                                                 
{'loss': '0.001352', 'grad_norm': '0.1043', 'learning_rate': '6.151e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 7472368, 'tokens/trainable': 112882, 'epoch': '0.9648'}
-
 48%|████▊     | 247/512 [27:46<29:36,  6.71s/it]
 48%|████▊     | 248/512 [27:53<29:30,  6.71s/it]
                                                 
{'loss': '0.001037', 'grad_norm': '0.0733', 'learning_rate': '6.122e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '40.07', 'tokens/total': 7502656, 'tokens/trainable': 113390, 'epoch': '0.9688'}
-
 48%|████▊     | 248/512 [27:53<29:30,  6.71s/it]
 49%|████▊     | 249/512 [28:00<29:22,  6.70s/it]
                                                 
{'loss': '0.01024', 'grad_norm': '0.7874', 'learning_rate': '6.093e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '37.84', 'tokens/total': 7532800, 'tokens/trainable': 113904, 'epoch': '0.9727'}
-
 49%|████▊     | 249/512 [28:00<29:22,  6.70s/it]
 49%|████▉     | 250/512 [28:06<29:17,  6.71s/it]
                                                 
{'loss': '0.0006508', 'grad_norm': '0.041', 'learning_rate': '6.065e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 7563264, 'tokens/trainable': 114351, 'epoch': '0.9766'}
-
 49%|████▉     | 250/512 [28:06<29:17,  6.71s/it]
 49%|████▉     | 251/512 [28:13<29:12,  6.71s/it]
                                                 
{'loss': '0.002258', 'grad_norm': '0.147', 'learning_rate': '6.036e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 7593840, 'tokens/trainable': 114842, 'epoch': '0.9805'}
-
 49%|████▉     | 251/512 [28:13<29:12,  6.71s/it]
 49%|████▉     | 252/512 [28:20<29:02,  6.70s/it]
                                                 
{'loss': '0.00276', 'grad_norm': '0.209', 'learning_rate': '6.007e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '28.92', 'tokens/total': 7624416, 'tokens/trainable': 115263, 'epoch': '0.9844'}
-
 49%|████▉     | 252/512 [28:20<29:02,  6.70s/it]
 49%|████▉     | 253/512 [28:27<28:54,  6.70s/it]
                                                 
{'loss': '0.001512', 'grad_norm': '0.1135', 'learning_rate': '5.978e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 7654688, 'tokens/trainable': 115703, 'epoch': '0.9883'}
-
 49%|████▉     | 253/512 [28:27<28:54,  6.70s/it]
 50%|████▉     | 254/512 [28:33<28:43,  6.68s/it]
                                                 
{'loss': '0.00226', 'grad_norm': '0.2045', 'learning_rate': '5.949e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 7685088, 'tokens/trainable': 116157, 'epoch': '0.9922'}
-
 50%|████▉     | 254/512 [28:33<28:43,  6.68s/it]
 50%|████▉     | 255/512 [28:40<28:33,  6.67s/it]
                                                 
{'loss': '0.01213', 'grad_norm': '0.662', 'learning_rate': '5.92e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 7715552, 'tokens/trainable': 116567, 'epoch': '0.9961'}
-
 50%|████▉     | 255/512 [28:40<28:33,  6.67s/it]
 50%|█████     | 256/512 [28:47<28:36,  6.71s/it]
                                                 
{'loss': '0.0002805', 'grad_norm': '0.01708', 'learning_rate': '5.891e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '31.26', 'tokens/total': 7745872, 'tokens/trainable': 117030, 'epoch': '1'}
-
 50%|█████     | 256/512 [28:47<28:36,  6.71s/it][2026-08-18 13:12:35,117] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
-
 50%|█████     | 257/512 [28:56<31:19,  7.37s/it]
                                                 
{'loss': '0.0001653', 'grad_norm': '0.007685', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 7776208, 'tokens/trainable': 117517, 'epoch': '1.004'}
-
 50%|█████     | 257/512 [28:56<31:19,  7.37s/it]
 50%|█████     | 258/512 [29:02<30:20,  7.17s/it]
                                                 
{'loss': '0.0004825', 'grad_norm': '0.04673', 'learning_rate': '5.834e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '38.44', 'tokens/total': 7806480, 'tokens/trainable': 118015, 'epoch': '1.008'}
-
 50%|█████     | 258/512 [29:02<30:20,  7.17s/it]
 51%|█████     | 259/512 [29:09<29:37,  7.03s/it]
                                                 
{'loss': '0.001111', 'grad_norm': '0.09396', 'learning_rate': '5.805e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 7836800, 'tokens/trainable': 118520, 'epoch': '1.012'}
-
 51%|█████     | 259/512 [29:09<29:37,  7.03s/it]
 51%|█████     | 260/512 [29:16<29:06,  6.93s/it]
                                                 
{'loss': '0.0003673', 'grad_norm': '0.02427', 'learning_rate': '5.776e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 7867040, 'tokens/trainable': 118992, 'epoch': '1.016'}
-
 51%|█████     | 260/512 [29:16<29:06,  6.93s/it]
 51%|█████     | 261/512 [29:22<28:41,  6.86s/it]
                                                 
{'loss': '0.002127', 'grad_norm': '0.1085', 'learning_rate': '5.747e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 7897408, 'tokens/trainable': 119438, 'epoch': '1.02'}
-
 51%|█████     | 261/512 [29:22<28:41,  6.86s/it]
 51%|█████     | 262/512 [29:29<28:18,  6.79s/it]
                                                 
{'loss': '0.001213', 'grad_norm': '0.1838', 'learning_rate': '5.718e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.26', 'tokens/total': 7927648, 'tokens/trainable': 119881, 'epoch': '1.023'}
-
 51%|█████     | 262/512 [29:29<28:18,  6.79s/it]
 51%|█████▏    | 263/512 [29:36<27:59,  6.74s/it]
                                                 
{'loss': '0.002099', 'grad_norm': '0.2827', 'learning_rate': '5.689e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '27.91', 'tokens/total': 7957968, 'tokens/trainable': 120285, 'epoch': '1.027'}
-
 51%|█████▏    | 263/512 [29:36<27:59,  6.74s/it]
 52%|█████▏    | 264/512 [29:42<27:43,  6.71s/it]
                                                 
{'loss': '0.0004398', 'grad_norm': '0.07407', 'learning_rate': '5.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.2', 'tokens/total': 7987888, 'tokens/trainable': 120755, 'epoch': '1.031'}
-
 52%|█████▏    | 264/512 [29:42<27:43,  6.71s/it]
 52%|█████▏    | 265/512 [29:49<27:32,  6.69s/it]
                                                 
{'loss': '0.0001488', 'grad_norm': '0.01879', 'learning_rate': '5.631e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.69', 'tokens/total': 8018048, 'tokens/trainable': 121194, 'epoch': '1.035'}
-
 52%|█████▏    | 265/512 [29:49<27:32,  6.69s/it]
 52%|█████▏    | 266/512 [29:56<27:24,  6.69s/it]
                                                 
{'loss': '0.0001396', 'grad_norm': '0.01209', 'learning_rate': '5.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 8048448, 'tokens/trainable': 121670, 'epoch': '1.039'}
-
 52%|█████▏    | 266/512 [29:56<27:24,  6.69s/it]
 52%|█████▏    | 267/512 [30:02<27:21,  6.70s/it]
                                                 
{'loss': '0.0001756', 'grad_norm': '0.03192', 'learning_rate': '5.573e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 8078848, 'tokens/trainable': 122142, 'epoch': '1.043'}
-
 52%|█████▏    | 267/512 [30:02<27:21,  6.70s/it]
 52%|█████▏    | 268/512 [30:09<27:19,  6.72s/it]
                                                 
{'loss': '0.0001036', 'grad_norm': '0.008872', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.41', 'tokens/total': 8109360, 'tokens/trainable': 122585, 'epoch': '1.047'}
-
 52%|█████▏    | 268/512 [30:09<27:19,  6.72s/it]
 53%|█████▎    | 269/512 [30:16<27:12,  6.72s/it]
                                                 
{'loss': '0.002882', 'grad_norm': '0.3732', 'learning_rate': '5.515e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '38.85', 'tokens/total': 8139888, 'tokens/trainable': 123091, 'epoch': '1.051'}
-
 53%|█████▎    | 269/512 [30:16<27:12,  6.72s/it]
 53%|█████▎    | 270/512 [30:23<27:14,  6.75s/it]
                                                 
{'loss': '0.0003205', 'grad_norm': '0.01947', 'learning_rate': '5.485e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '30.9', 'tokens/total': 8170480, 'tokens/trainable': 123505, 'epoch': '1.055'}
-
 53%|█████▎    | 270/512 [30:23<27:14,  6.75s/it]
 53%|█████▎    | 271/512 [30:29<26:28,  6.59s/it]
                                                 
{'loss': '0.0003346', 'grad_norm': '0.0526', 'learning_rate': '5.456e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 8198848, 'tokens/trainable': 123953, 'epoch': '1.059'}
-
 53%|█████▎    | 271/512 [30:29<26:28,  6.59s/it]
 53%|█████▎    | 272/512 [30:36<26:32,  6.64s/it]
                                                 
{'loss': '0.0004355', 'grad_norm': '0.09935', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '28.66', 'tokens/total': 8229200, 'tokens/trainable': 124400, 'epoch': '1.062'}
-
 53%|█████▎    | 272/512 [30:36<26:32,  6.64s/it]
 53%|█████▎    | 273/512 [30:42<26:30,  6.66s/it]
                                                 
{'loss': '0.00702', 'grad_norm': '0.3927', 'learning_rate': '5.398e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.05', 'tokens/total': 8259536, 'tokens/trainable': 124870, 'epoch': '1.066'}
-
 53%|█████▎    | 273/512 [30:42<26:30,  6.66s/it]
 54%|█████▎    | 274/512 [30:49<26:25,  6.66s/it]
                                                 
{'loss': '0.0003887', 'grad_norm': '0.0385', 'learning_rate': '5.369e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.07', 'tokens/total': 8289904, 'tokens/trainable': 125333, 'epoch': '1.07'}
-
 54%|█████▎    | 274/512 [30:49<26:25,  6.66s/it]
 54%|█████▎    | 275/512 [30:56<26:23,  6.68s/it]
                                                 
{'loss': '8.032e-05', 'grad_norm': '0.01037', 'learning_rate': '5.34e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 8320176, 'tokens/trainable': 125834, 'epoch': '1.074'}
-
 54%|█████▎    | 275/512 [30:56<26:23,  6.68s/it]
 54%|█████▍    | 276/512 [31:02<26:19,  6.69s/it]
                                                 
{'loss': '0.0002651', 'grad_norm': '0.05539', 'learning_rate': '5.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '37.97', 'tokens/total': 8350832, 'tokens/trainable': 126316, 'epoch': '1.078'}
-
 54%|█████▍    | 276/512 [31:02<26:19,  6.69s/it]
 54%|█████▍    | 277/512 [31:09<26:14,  6.70s/it]
                                                 
{'loss': '0.0007178', 'grad_norm': '0.2011', 'learning_rate': '5.282e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 8380976, 'tokens/trainable': 126798, 'epoch': '1.082'}
-
 54%|█████▍    | 277/512 [31:09<26:14,  6.70s/it]
 54%|█████▍    | 278/512 [31:16<26:00,  6.67s/it]
                                                 
{'loss': '4.906e-05', 'grad_norm': '0.005951', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.57', 'tokens/total': 8411072, 'tokens/trainable': 127223, 'epoch': '1.086'}
-
 54%|█████▍    | 278/512 [31:16<26:00,  6.67s/it]
 54%|█████▍    | 279/512 [31:22<25:45,  6.63s/it]
                                                 
{'loss': '8.996e-05', 'grad_norm': '0.009279', 'learning_rate': '5.224e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.01', 'tokens/total': 8440736, 'tokens/trainable': 127672, 'epoch': '1.09'}
-
 54%|█████▍    | 279/512 [31:22<25:45,  6.63s/it]
 55%|█████▍    | 280/512 [31:29<25:39,  6.64s/it]
                                                 
{'loss': '0.0006871', 'grad_norm': '0.1146', 'learning_rate': '5.195e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.07', 'tokens/total': 8470944, 'tokens/trainable': 128116, 'epoch': '1.094'}
-
 55%|█████▍    | 280/512 [31:29<25:39,  6.64s/it]
 55%|█████▍    | 281/512 [31:36<25:39,  6.66s/it]
                                                 
{'loss': '4.705e-05', 'grad_norm': '0.004336', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 8501104, 'tokens/trainable': 128589, 'epoch': '1.098'}
-
 55%|█████▍    | 281/512 [31:36<25:39,  6.66s/it]
 55%|█████▌    | 282/512 [31:42<25:34,  6.67s/it]
                                                 
{'loss': '0.01315', 'grad_norm': '0.2142', 'learning_rate': '5.138e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 8531696, 'tokens/trainable': 129036, 'epoch': '1.102'}
-
 55%|█████▌    | 282/512 [31:42<25:34,  6.67s/it]
 55%|█████▌    | 283/512 [31:49<25:31,  6.69s/it]
                                                 
{'loss': '1.559e-05', 'grad_norm': '0.0006189', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.16', 'tokens/total': 8561936, 'tokens/trainable': 129485, 'epoch': '1.105'}
-
 55%|█████▌    | 283/512 [31:49<25:31,  6.69s/it]
 55%|█████▌    | 284/512 [31:56<25:26,  6.69s/it]
                                                 
{'loss': '8.038e-05', 'grad_norm': '0.005335', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.49', 'tokens/total': 8592192, 'tokens/trainable': 129920, 'epoch': '1.109'}
-
 55%|█████▌    | 284/512 [31:56<25:26,  6.69s/it]
 56%|█████▌    | 285/512 [32:02<25:22,  6.71s/it]
                                                 
{'loss': '4.425e-05', 'grad_norm': '0.003026', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.06', 'tokens/total': 8622432, 'tokens/trainable': 130417, 'epoch': '1.113'}
-
 56%|█████▌    | 285/512 [32:02<25:22,  6.71s/it]
 56%|█████▌    | 286/512 [32:09<25:16,  6.71s/it]
                                                 
{'loss': '5.365e-05', 'grad_norm': '0.003212', 'learning_rate': '5.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.89', 'tokens/total': 8652768, 'tokens/trainable': 130903, 'epoch': '1.117'}
-
 56%|█████▌    | 286/512 [32:09<25:16,  6.71s/it]
 56%|█████▌    | 287/512 [32:16<25:08,  6.70s/it]
                                                 
{'loss': '9.199e-05', 'grad_norm': '0.01538', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 8683296, 'tokens/trainable': 131343, 'epoch': '1.121'}
-
 56%|█████▌    | 287/512 [32:16<25:08,  6.70s/it]
 56%|█████▋    | 288/512 [32:23<25:00,  6.70s/it]
                                                 
{'loss': '0.006935', 'grad_norm': '0.8839', 'learning_rate': '4.964e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.19', 'tokens/total': 8713504, 'tokens/trainable': 131763, 'epoch': '1.125'}
-
 56%|█████▋    | 288/512 [32:23<25:00,  6.70s/it][2026-08-18 13:16:11,062] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
-
 56%|█████▋    | 289/512 [32:31<27:23,  7.37s/it]
                                                 
{'loss': '2.326e-05', 'grad_norm': '0.001999', 'learning_rate': '4.935e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 8743888, 'tokens/trainable': 132229, 'epoch': '1.129'}
-
 56%|█████▋    | 289/512 [32:31<27:23,  7.37s/it]
 57%|█████▋    | 290/512 [32:38<26:30,  7.16s/it]
                                                 
{'loss': '1.431e-05', 'grad_norm': '0.0004976', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.54', 'tokens/train_per_sec_per_gpu': '35.46', 'tokens/total': 8774336, 'tokens/trainable': 132678, 'epoch': '1.133'}
-
 57%|█████▋    | 290/512 [32:38<26:30,  7.16s/it]
 57%|█████▋    | 291/512 [32:45<25:52,  7.03s/it]
                                                 
{'loss': '0.0001439', 'grad_norm': '0.01294', 'learning_rate': '4.878e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.39', 'tokens/total': 8804816, 'tokens/trainable': 133136, 'epoch': '1.137'}
-
 57%|█████▋    | 291/512 [32:45<25:52,  7.03s/it]
 57%|█████▋    | 292/512 [32:52<25:25,  6.93s/it]
                                                 
{'loss': '0.003074', 'grad_norm': '0.3451', 'learning_rate': '4.849e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '35.31', 'tokens/total': 8835024, 'tokens/trainable': 133608, 'epoch': '1.141'}
-
 57%|█████▋    | 292/512 [32:52<25:25,  6.93s/it]
 57%|█████▋    | 293/512 [32:58<25:01,  6.86s/it]
                                                 
{'loss': '0.0001658', 'grad_norm': '0.02419', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '40.44', 'tokens/total': 8865408, 'tokens/trainable': 134108, 'epoch': '1.145'}
-
 57%|█████▋    | 293/512 [32:58<25:01,  6.86s/it]
 57%|█████▋    | 294/512 [33:05<24:44,  6.81s/it]
                                                 
{'loss': '3.24e-05', 'grad_norm': '0.001724', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.72', 'tokens/total': 8895824, 'tokens/trainable': 134541, 'epoch': '1.148'}
-
 57%|█████▋    | 294/512 [33:05<24:44,  6.81s/it]
 58%|█████▊    | 295/512 [33:12<24:29,  6.77s/it]
                                                 
{'loss': '0.000106', 'grad_norm': '0.01028', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.51', 'tokens/total': 8926336, 'tokens/trainable': 134966, 'epoch': '1.152'}
-
 58%|█████▊    | 295/512 [33:12<24:29,  6.77s/it]
 58%|█████▊    | 296/512 [33:18<24:14,  6.73s/it]
                                                 
{'loss': '0.0002625', 'grad_norm': '0.02922', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '33.56', 'tokens/total': 8956656, 'tokens/trainable': 135402, 'epoch': '1.156'}
-
 58%|█████▊    | 296/512 [33:18<24:14,  6.73s/it]
 58%|█████▊    | 297/512 [33:25<23:58,  6.69s/it]
                                                 
{'loss': '0.004879', 'grad_norm': '0.4516', 'learning_rate': '4.706e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.05', 'tokens/total': 8986752, 'tokens/trainable': 135804, 'epoch': '1.16'}
-
 58%|█████▊    | 297/512 [33:25<23:58,  6.69s/it]
 58%|█████▊    | 298/512 [33:31<23:46,  6.67s/it]
                                                 
{'loss': '0.0001912', 'grad_norm': '0.0781', 'learning_rate': '4.677e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 9017024, 'tokens/trainable': 136231, 'epoch': '1.164'}
-
 58%|█████▊    | 298/512 [33:31<23:46,  6.67s/it]
 58%|█████▊    | 299/512 [33:38<23:41,  6.67s/it]
                                                 
{'loss': '6.207e-05', 'grad_norm': '0.008365', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '34.27', 'tokens/total': 9047424, 'tokens/trainable': 136705, 'epoch': '1.168'}
-
 58%|█████▊    | 299/512 [33:38<23:41,  6.67s/it]
 59%|█████▊    | 300/512 [33:45<23:31,  6.66s/it]
                                                 
{'loss': '0.008261', 'grad_norm': '0.3563', 'learning_rate': '4.62e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.86', 'tokens/total': 9077648, 'tokens/trainable': 137178, 'epoch': '1.172'}
-
 59%|█████▊    | 300/512 [33:45<23:31,  6.66s/it]
 59%|█████▉    | 301/512 [33:51<23:26,  6.67s/it]
                                                 
{'loss': '0.0006729', 'grad_norm': '0.1326', 'learning_rate': '4.592e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.42', 'tokens/total': 9108032, 'tokens/trainable': 137643, 'epoch': '1.176'}
-
 59%|█████▉    | 301/512 [33:51<23:26,  6.67s/it]
 59%|█████▉    | 302/512 [33:58<23:20,  6.67s/it]
                                                 
{'loss': '0.0008891', 'grad_norm': '0.1117', 'learning_rate': '4.563e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 9138240, 'tokens/trainable': 138078, 'epoch': '1.18'}
-
 59%|█████▉    | 302/512 [33:58<23:20,  6.67s/it]
 59%|█████▉    | 303/512 [34:05<23:11,  6.66s/it]
                                                 
{'loss': '0.004502', 'grad_norm': '0.6568', 'learning_rate': '4.535e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 9168352, 'tokens/trainable': 138515, 'epoch': '1.184'}
-
 59%|█████▉    | 303/512 [34:05<23:11,  6.66s/it]
 59%|█████▉    | 304/512 [34:12<23:10,  6.69s/it]
                                                 
{'loss': '9.651e-05', 'grad_norm': '0.007431', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.99', 'tokens/total': 9198864, 'tokens/trainable': 138948, 'epoch': '1.188'}
-
 59%|█████▉    | 304/512 [34:12<23:10,  6.69s/it]
 60%|█████▉    | 305/512 [34:18<23:03,  6.68s/it]
                                                 
{'loss': '5.174e-05', 'grad_norm': '0.003385', 'learning_rate': '4.478e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '38.17', 'tokens/total': 9229104, 'tokens/trainable': 139408, 'epoch': '1.191'}
-
 60%|█████▉    | 305/512 [34:18<23:03,  6.68s/it]
 60%|█████▉    | 306/512 [34:24<22:24,  6.53s/it]
                                                 
{'loss': '3.198e-05', 'grad_norm': '0.001723', 'learning_rate': '4.45e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '33.48', 'tokens/total': 9257344, 'tokens/trainable': 139840, 'epoch': '1.195'}
-
 60%|█████▉    | 306/512 [34:24<22:24,  6.53s/it]
 60%|█████▉    | 307/512 [34:31<22:27,  6.57s/it]
                                                 
{'loss': '0.0004305', 'grad_norm': '0.06396', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 9287520, 'tokens/trainable': 140281, 'epoch': '1.199'}
-
 60%|█████▉    | 307/512 [34:31<22:27,  6.57s/it]
 60%|██████    | 308/512 [34:38<22:27,  6.61s/it]
                                                 
{'loss': '0.000683', 'grad_norm': '0.1434', 'learning_rate': '4.394e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '37.44', 'tokens/total': 9317888, 'tokens/trainable': 140776, 'epoch': '1.203'}
-
 60%|██████    | 308/512 [34:38<22:27,  6.61s/it]
 60%|██████    | 309/512 [34:44<22:24,  6.63s/it]
                                                 
{'loss': '0.005062', 'grad_norm': '0.392', 'learning_rate': '4.366e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '36.89', 'tokens/total': 9348256, 'tokens/trainable': 141228, 'epoch': '1.207'}
-
 60%|██████    | 309/512 [34:44<22:24,  6.63s/it]
 61%|██████    | 310/512 [34:51<22:18,  6.63s/it]
                                                 
{'loss': '0.0001797', 'grad_norm': '0.01664', 'learning_rate': '4.338e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '37.42', 'tokens/total': 9378560, 'tokens/trainable': 141679, 'epoch': '1.211'}
-
 61%|██████    | 310/512 [34:51<22:18,  6.63s/it]
 61%|██████    | 311/512 [34:58<22:12,  6.63s/it]
                                                 
{'loss': '1.943e-05', 'grad_norm': '0.0008493', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '32.27', 'tokens/total': 9408848, 'tokens/trainable': 142122, 'epoch': '1.215'}
-
 61%|██████    | 311/512 [34:58<22:12,  6.63s/it]
 61%|██████    | 312/512 [35:04<22:06,  6.63s/it]
                                                 
{'loss': '0.0005122', 'grad_norm': '0.04301', 'learning_rate': '4.282e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.11', 'tokens/total': 9439344, 'tokens/trainable': 142561, 'epoch': '1.219'}
-
 61%|██████    | 312/512 [35:04<22:06,  6.63s/it]
 61%|██████    | 313/512 [35:11<22:04,  6.65s/it]
                                                 
{'loss': '0.0001515', 'grad_norm': '0.02346', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.66', 'tokens/total': 9469792, 'tokens/trainable': 143046, 'epoch': '1.223'}
-
 61%|██████    | 313/512 [35:11<22:04,  6.65s/it]
 61%|██████▏   | 314/512 [35:18<22:00,  6.67s/it]
                                                 
{'loss': '0.0006228', 'grad_norm': '0.06129', 'learning_rate': '4.226e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 9500128, 'tokens/trainable': 143493, 'epoch': '1.227'}
-
 61%|██████▏   | 314/512 [35:18<22:00,  6.67s/it]
 62%|██████▏   | 315/512 [35:24<21:51,  6.66s/it]
                                                 
{'loss': '7.468e-05', 'grad_norm': '0.007639', 'learning_rate': '4.198e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 9530480, 'tokens/trainable': 143956, 'epoch': '1.23'}
-
 62%|██████▏   | 315/512 [35:24<21:51,  6.66s/it]
 62%|██████▏   | 316/512 [35:31<21:47,  6.67s/it]
                                                 
{'loss': '0.00437', 'grad_norm': '0.3535', 'learning_rate': '4.17e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 9560912, 'tokens/trainable': 144411, 'epoch': '1.234'}
-
 62%|██████▏   | 316/512 [35:31<21:47,  6.67s/it]
 62%|██████▏   | 317/512 [35:38<21:44,  6.69s/it]
                                                 
{'loss': '0.01097', 'grad_norm': '0.502', 'learning_rate': '4.143e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '38.48', 'tokens/total': 9591344, 'tokens/trainable': 144892, 'epoch': '1.238'}
-
 62%|██████▏   | 317/512 [35:38<21:44,  6.69s/it]
 62%|██████▏   | 318/512 [35:44<21:35,  6.68s/it]
                                                 
{'loss': '4.459e-05', 'grad_norm': '0.002735', 'learning_rate': '4.115e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '32.6', 'tokens/total': 9621392, 'tokens/trainable': 145339, 'epoch': '1.242'}
-
 62%|██████▏   | 318/512 [35:44<21:35,  6.68s/it]
 62%|██████▏   | 319/512 [35:51<21:31,  6.69s/it]
                                                 
{'loss': '8.017e-05', 'grad_norm': '0.01343', 'learning_rate': '4.087e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 9651728, 'tokens/trainable': 145811, 'epoch': '1.246'}
-
 62%|██████▏   | 319/512 [35:51<21:31,  6.69s/it]
 62%|██████▎   | 320/512 [35:58<21:27,  6.71s/it]
                                                 
{'loss': '0.0001716', 'grad_norm': '0.02255', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '38.76', 'tokens/total': 9682224, 'tokens/trainable': 146314, 'epoch': '1.25'}
-
 62%|██████▎   | 320/512 [35:58<21:27,  6.71s/it][2026-08-18 13:19:46,437] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
-
 63%|██████▎   | 321/512 [36:06<23:06,  7.26s/it]
                                                 
{'loss': '0.0005239', 'grad_norm': '0.0436', 'learning_rate': '4.032e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '31.85', 'tokens/total': 9712656, 'tokens/trainable': 146781, 'epoch': '1.254'}
-
 63%|██████▎   | 321/512 [36:06<23:06,  7.26s/it]
 63%|██████▎   | 322/512 [36:13<22:23,  7.07s/it]
                                                 
{'loss': '0.0001891', 'grad_norm': '0.02378', 'learning_rate': '4.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '34.89', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 9743088, 'tokens/trainable': 147225, 'epoch': '1.258'}
-
 63%|██████▎   | 322/512 [36:13<22:23,  7.07s/it]
 63%|██████▎   | 323/512 [36:20<21:58,  6.98s/it]
                                                 
{'loss': '8.73e-05', 'grad_norm': '0.01257', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.17', 'tokens/train_per_sec_per_gpu': '34.82', 'tokens/total': 9773808, 'tokens/trainable': 147673, 'epoch': '1.262'}
-
 63%|██████▎   | 323/512 [36:20<21:58,  6.98s/it]
 63%|██████▎   | 324/512 [36:26<21:29,  6.86s/it]
                                                 
{'loss': '0.002842', 'grad_norm': '0.2107', 'learning_rate': '3.95e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 9804016, 'tokens/trainable': 148103, 'epoch': '1.266'}
-
 63%|██████▎   | 324/512 [36:26<21:29,  6.86s/it]
 63%|██████▎   | 325/512 [36:33<21:11,  6.80s/it]
                                                 
{'loss': '0.0001613', 'grad_norm': '0.02561', 'learning_rate': '3.923e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '31.73', 'tokens/total': 9834592, 'tokens/trainable': 148535, 'epoch': '1.27'}
-
 63%|██████▎   | 325/512 [36:33<21:11,  6.80s/it]
 64%|██████▎   | 326/512 [36:40<20:58,  6.77s/it]
                                                 
{'loss': '0.0003376', 'grad_norm': '0.03319', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '34.07', 'tokens/total': 9864848, 'tokens/trainable': 148999, 'epoch': '1.273'}
-
 64%|██████▎   | 326/512 [36:40<20:58,  6.77s/it]
 64%|██████▍   | 327/512 [36:46<20:46,  6.74s/it]
                                                 
{'loss': '0.0007673', 'grad_norm': '0.07407', 'learning_rate': '3.869e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 9895168, 'tokens/trainable': 149431, 'epoch': '1.277'}
-
 64%|██████▍   | 327/512 [36:46<20:46,  6.74s/it]
 64%|██████▍   | 328/512 [36:53<20:35,  6.71s/it]
                                                 
{'loss': '0.0007135', 'grad_norm': '0.05904', 'learning_rate': '3.842e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 9925696, 'tokens/trainable': 149886, 'epoch': '1.281'}
-
 64%|██████▍   | 328/512 [36:53<20:35,  6.71s/it]
 64%|██████▍   | 329/512 [37:00<20:21,  6.68s/it]
                                                 
{'loss': '7.994e-05', 'grad_norm': '0.00763', 'learning_rate': '3.815e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.98', 'tokens/total': 9955648, 'tokens/trainable': 150288, 'epoch': '1.285'}
-
 64%|██████▍   | 329/512 [37:00<20:21,  6.68s/it]
 64%|██████▍   | 330/512 [37:06<20:16,  6.69s/it]
                                                 
{'loss': '0.0001955', 'grad_norm': '0.02474', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 9985856, 'tokens/trainable': 150733, 'epoch': '1.289'}
-
 64%|██████▍   | 330/512 [37:06<20:16,  6.69s/it]
 65%|██████▍   | 331/512 [37:13<20:11,  6.69s/it]
                                                 
{'loss': '0.0001031', 'grad_norm': '0.006353', 'learning_rate': '3.761e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 10016208, 'tokens/trainable': 151207, 'epoch': '1.293'}
-
 65%|██████▍   | 331/512 [37:13<20:11,  6.69s/it]
 65%|██████▍   | 332/512 [37:20<20:06,  6.70s/it]
                                                 
{'loss': '0.001474', 'grad_norm': '0.162', 'learning_rate': '3.734e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.5', 'tokens/total': 10046512, 'tokens/trainable': 151694, 'epoch': '1.297'}
-
 65%|██████▍   | 332/512 [37:20<20:06,  6.70s/it]
 65%|██████▌   | 333/512 [37:27<20:03,  6.72s/it]
                                                 
{'loss': '0.000419', 'grad_norm': '0.04182', 'learning_rate': '3.708e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '38.27', 'tokens/total': 10076944, 'tokens/trainable': 152188, 'epoch': '1.301'}
-
 65%|██████▌   | 333/512 [37:27<20:03,  6.72s/it]
 65%|██████▌   | 334/512 [37:33<19:53,  6.71s/it]
                                                 
{'loss': '3.35e-05', 'grad_norm': '0.001125', 'learning_rate': '3.681e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.6', 'tokens/total': 10107296, 'tokens/trainable': 152596, 'epoch': '1.305'}
-
 65%|██████▌   | 334/512 [37:33<19:53,  6.71s/it]
 65%|██████▌   | 335/512 [37:40<19:46,  6.70s/it]
                                                 
{'loss': '4.685e-05', 'grad_norm': '0.003708', 'learning_rate': '3.655e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 10137568, 'tokens/trainable': 153052, 'epoch': '1.309'}
-
 65%|██████▌   | 335/512 [37:40<19:46,  6.70s/it]
 66%|██████▌   | 336/512 [37:47<19:40,  6.71s/it]
                                                 
{'loss': '2.363e-05', 'grad_norm': '0.001753', 'learning_rate': '3.628e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.19', 'tokens/total': 10167952, 'tokens/trainable': 153491, 'epoch': '1.312'}
-
 66%|██████▌   | 336/512 [37:47<19:40,  6.71s/it]
 66%|██████▌   | 337/512 [37:53<19:36,  6.72s/it]
                                                 
{'loss': '0.0006341', 'grad_norm': '0.06262', 'learning_rate': '3.602e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '40.28', 'tokens/total': 10198512, 'tokens/trainable': 153983, 'epoch': '1.316'}
-
 66%|██████▌   | 337/512 [37:53<19:36,  6.72s/it]
 66%|██████▌   | 338/512 [38:00<19:29,  6.72s/it]
                                                 
{'loss': '0.0001541', 'grad_norm': '0.01662', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 10228752, 'tokens/trainable': 154464, 'epoch': '1.32'}
-
 66%|██████▌   | 338/512 [38:00<19:29,  6.72s/it]
 66%|██████▌   | 339/512 [38:07<19:21,  6.71s/it]
                                                 
{'loss': '8.605e-06', 'grad_norm': '0.0004508', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.73', 'tokens/total': 10259104, 'tokens/trainable': 154924, 'epoch': '1.324'}
-
 66%|██████▌   | 339/512 [38:07<19:21,  6.71s/it]
 66%|██████▋   | 340/512 [38:14<19:12,  6.70s/it]
                                                 
{'loss': '4.251e-05', 'grad_norm': '0.002413', 'learning_rate': '3.523e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.76', 'tokens/total': 10289520, 'tokens/trainable': 155373, 'epoch': '1.328'}
-
 66%|██████▋   | 340/512 [38:14<19:12,  6.70s/it]
 67%|██████▋   | 341/512 [38:20<19:08,  6.72s/it]
                                                 
{'loss': '0.0004415', 'grad_norm': '0.0478', 'learning_rate': '3.497e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 10320016, 'tokens/trainable': 155873, 'epoch': '1.332'}
-
 67%|██████▋   | 341/512 [38:20<19:08,  6.72s/it]
 67%|██████▋   | 342/512 [38:27<18:58,  6.70s/it]
                                                 
{'loss': '3.104e-05', 'grad_norm': '0.001358', 'learning_rate': '3.471e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.34', 'tokens/total': 10350368, 'tokens/trainable': 156278, 'epoch': '1.336'}
-
 67%|██████▋   | 342/512 [38:27<18:58,  6.70s/it]
 67%|██████▋   | 343/512 [38:34<18:52,  6.70s/it]
                                                 
{'loss': '9.817e-05', 'grad_norm': '0.0139', 'learning_rate': '3.445e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.31', 'tokens/total': 10380624, 'tokens/trainable': 156733, 'epoch': '1.34'}
-
 67%|██████▋   | 343/512 [38:34<18:52,  6.70s/it]
 67%|██████▋   | 344/512 [38:40<18:48,  6.72s/it]
                                                 
{'loss': '4.33e-05', 'grad_norm': '0.003103', 'learning_rate': '3.42e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 10411024, 'tokens/trainable': 157203, 'epoch': '1.344'}
-
 67%|██████▋   | 344/512 [38:40<18:48,  6.72s/it]
 67%|██████▋   | 345/512 [38:47<18:42,  6.72s/it]
                                                 
{'loss': '0.000103', 'grad_norm': '0.009028', 'learning_rate': '3.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 10441248, 'tokens/trainable': 157707, 'epoch': '1.348'}
-
 67%|██████▋   | 345/512 [38:47<18:42,  6.72s/it]
 68%|██████▊   | 346/512 [38:54<18:39,  6.74s/it]
                                                 
{'loss': '4.62e-05', 'grad_norm': '0.003629', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '34.96', 'tokens/total': 10471712, 'tokens/trainable': 158180, 'epoch': '1.352'}
-
 68%|██████▊   | 346/512 [38:54<18:39,  6.74s/it]
 68%|██████▊   | 347/512 [39:01<18:31,  6.74s/it]
                                                 
{'loss': '0.0009502', 'grad_norm': '0.08972', 'learning_rate': '3.343e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 10502288, 'tokens/trainable': 158646, 'epoch': '1.355'}
-
 68%|██████▊   | 347/512 [39:01<18:31,  6.74s/it]
 68%|██████▊   | 348/512 [39:07<18:18,  6.70s/it]
                                                 
{'loss': '7.784e-05', 'grad_norm': '0.008904', 'learning_rate': '3.317e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '35.59', 'tokens/total': 10532304, 'tokens/trainable': 159115, 'epoch': '1.359'}
-
 68%|██████▊   | 348/512 [39:07<18:18,  6.70s/it]
 68%|██████▊   | 349/512 [39:14<18:10,  6.69s/it]
                                                 
{'loss': '0.0002163', 'grad_norm': '0.02332', 'learning_rate': '3.292e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 10562608, 'tokens/trainable': 159538, 'epoch': '1.363'}
-
 68%|██████▊   | 349/512 [39:14<18:10,  6.69s/it]
 68%|██████▊   | 350/512 [39:21<18:04,  6.69s/it]
                                                 
{'loss': '0.0004488', 'grad_norm': '0.04788', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 10592992, 'tokens/trainable': 160016, 'epoch': '1.367'}
-
 68%|██████▊   | 350/512 [39:21<18:04,  6.69s/it]
 69%|██████▊   | 351/512 [39:27<17:56,  6.69s/it]
                                                 
{'loss': '5.017e-05', 'grad_norm': '0.008666', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 10623328, 'tokens/trainable': 160475, 'epoch': '1.371'}
-
 69%|██████▊   | 351/512 [39:27<17:56,  6.69s/it]
 69%|██████▉   | 352/512 [39:34<17:50,  6.69s/it]
                                                 
{'loss': '0.0007272', 'grad_norm': '0.06453', 'learning_rate': '3.217e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 10653632, 'tokens/trainable': 160952, 'epoch': '1.375'}
-
 69%|██████▉   | 352/512 [39:34<17:50,  6.69s/it][2026-08-18 13:23:22,545] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
-
 69%|██████▉   | 353/512 [39:43<19:15,  7.27s/it]
                                                 
{'loss': '0.007551', 'grad_norm': '0.3706', 'learning_rate': '3.192e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 10684176, 'tokens/trainable': 161420, 'epoch': '1.379'}
-
 69%|██████▉   | 353/512 [39:43<19:15,  7.27s/it]
 69%|██████▉   | 354/512 [39:49<18:38,  7.08s/it]
                                                 
{'loss': '2.345e-05', 'grad_norm': '0.003138', 'learning_rate': '3.167e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 10714352, 'tokens/trainable': 161868, 'epoch': '1.383'}
-
 69%|██████▉   | 354/512 [39:49<18:38,  7.08s/it]
 69%|██████▉   | 355/512 [39:56<18:12,  6.96s/it]
                                                 
{'loss': '2.583e-05', 'grad_norm': '0.00214', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 10744464, 'tokens/trainable': 162318, 'epoch': '1.387'}
-
 69%|██████▉   | 355/512 [39:56<18:12,  6.96s/it]
 70%|██████▉   | 356/512 [40:03<17:55,  6.89s/it]
                                                 
{'loss': '3.729e-05', 'grad_norm': '0.002815', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 10774976, 'tokens/trainable': 162789, 'epoch': '1.391'}
-
 70%|██████▉   | 356/512 [40:03<17:55,  6.89s/it]
 70%|██████▉   | 357/512 [40:09<17:40,  6.84s/it]
                                                 
{'loss': '0.002785', 'grad_norm': '0.1642', 'learning_rate': '3.093e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 10805600, 'tokens/trainable': 163254, 'epoch': '1.395'}
-
 70%|██████▉   | 357/512 [40:09<17:40,  6.84s/it]
 70%|██████▉   | 358/512 [40:16<17:26,  6.79s/it]
                                                 
{'loss': '0.0002488', 'grad_norm': '0.06654', 'learning_rate': '3.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.41', 'tokens/total': 10835920, 'tokens/trainable': 163715, 'epoch': '1.398'}
-
 70%|██████▉   | 358/512 [40:16<17:26,  6.79s/it]
 70%|███████   | 359/512 [40:23<17:15,  6.77s/it]
                                                 
{'loss': '7.841e-05', 'grad_norm': '0.006618', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.4', 'tokens/total': 10866048, 'tokens/trainable': 164197, 'epoch': '1.402'}
-
 70%|███████   | 359/512 [40:23<17:15,  6.77s/it]
 70%|███████   | 360/512 [40:30<17:06,  6.75s/it]
                                                 
{'loss': '0.0001049', 'grad_norm': '0.01041', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.6', 'tokens/total': 10896288, 'tokens/trainable': 164661, 'epoch': '1.406'}
-
 70%|███████   | 360/512 [40:30<17:06,  6.75s/it]
 71%|███████   | 361/512 [40:36<16:56,  6.73s/it]
                                                 
{'loss': '4.034e-05', 'grad_norm': '0.002426', 'learning_rate': '2.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 10926816, 'tokens/trainable': 165111, 'epoch': '1.41'}
-
 71%|███████   | 361/512 [40:36<16:56,  6.73s/it]
 71%|███████   | 362/512 [40:42<16:24,  6.56s/it]
                                                 
{'loss': '0.0001499', 'grad_norm': '0.01567', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '30.34', 'tokens/total': 10954928, 'tokens/trainable': 165552, 'epoch': '1.414'}
-
 71%|███████   | 362/512 [40:42<16:24,  6.56s/it]
 71%|███████   | 363/512 [40:49<16:24,  6.61s/it]
                                                 
{'loss': '0.001498', 'grad_norm': '0.2696', 'learning_rate': '2.947e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 10985328, 'tokens/trainable': 166026, 'epoch': '1.418'}
-
 71%|███████   | 363/512 [40:49<16:24,  6.61s/it]
 71%|███████   | 364/512 [40:56<16:22,  6.64s/it]
                                                 
{'loss': '0.0003051', 'grad_norm': '0.03586', 'learning_rate': '2.924e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '37.86', 'tokens/total': 11015664, 'tokens/trainable': 166486, 'epoch': '1.422'}
-
 71%|███████   | 364/512 [40:56<16:22,  6.64s/it]
 71%|███████▏  | 365/512 [41:02<16:18,  6.66s/it]
                                                 
{'loss': '2.413e-05', 'grad_norm': '0.001931', 'learning_rate': '2.9e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 11046256, 'tokens/trainable': 166959, 'epoch': '1.426'}
-
 71%|███████▏  | 365/512 [41:03<16:18,  6.66s/it]
 71%|███████▏  | 366/512 [41:09<16:10,  6.65s/it]
                                                 
{'loss': '2.824e-05', 'grad_norm': '0.005941', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '33.64', 'tokens/total': 11076288, 'tokens/trainable': 167394, 'epoch': '1.43'}
-
 71%|███████▏  | 366/512 [41:09<16:10,  6.65s/it]
 72%|███████▏  | 367/512 [41:16<16:03,  6.64s/it]
                                                 
{'loss': '1.603e-05', 'grad_norm': '0.002126', 'learning_rate': '2.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '28.5', 'tokens/total': 11106416, 'tokens/trainable': 167832, 'epoch': '1.434'}
-
 72%|███████▏  | 367/512 [41:16<16:03,  6.64s/it]
 72%|███████▏  | 368/512 [41:22<15:58,  6.65s/it]
                                                 
{'loss': '0.000104', 'grad_norm': '0.251', 'learning_rate': '2.829e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.86', 'tokens/total': 11136848, 'tokens/trainable': 168242, 'epoch': '1.438'}
-
 72%|███████▏  | 368/512 [41:22<15:58,  6.65s/it]
 72%|███████▏  | 369/512 [41:29<15:52,  6.66s/it]
                                                 
{'loss': '2.642e-05', 'grad_norm': '0.001904', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.63', 'tokens/total': 11167376, 'tokens/trainable': 168715, 'epoch': '1.441'}
-
 72%|███████▏  | 369/512 [41:29<15:52,  6.66s/it]
 72%|███████▏  | 370/512 [41:36<15:48,  6.68s/it]
                                                 
{'loss': '2.997e-05', 'grad_norm': '0.002815', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.7', 'tokens/total': 11197776, 'tokens/trainable': 169147, 'epoch': '1.445'}
-
 72%|███████▏  | 370/512 [41:36<15:48,  6.68s/it]
 72%|███████▏  | 371/512 [41:43<15:41,  6.68s/it]
                                                 
{'loss': '1.342e-05', 'grad_norm': '0.001082', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 11227840, 'tokens/trainable': 169628, 'epoch': '1.449'}
-
 72%|███████▏  | 371/512 [41:43<15:41,  6.68s/it]
 73%|███████▎  | 372/512 [41:49<15:36,  6.69s/it]
                                                 
{'loss': '6.656e-05', 'grad_norm': '0.005368', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 11258176, 'tokens/trainable': 170085, 'epoch': '1.453'}
-
 73%|███████▎  | 372/512 [41:49<15:36,  6.69s/it]
 73%|███████▎  | 373/512 [41:56<15:28,  6.68s/it]
                                                 
{'loss': '0.0003471', 'grad_norm': '0.04757', 'learning_rate': '2.714e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 11288336, 'tokens/trainable': 170584, 'epoch': '1.457'}
-
 73%|███████▎  | 373/512 [41:56<15:28,  6.68s/it]
 73%|███████▎  | 374/512 [42:03<15:21,  6.68s/it]
                                                 
{'loss': '0.000288', 'grad_norm': '0.04089', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 11318848, 'tokens/trainable': 171058, 'epoch': '1.461'}
-
 73%|███████▎  | 374/512 [42:03<15:21,  6.68s/it]
 73%|███████▎  | 375/512 [42:09<15:16,  6.69s/it]
                                                 
{'loss': '2.691e-05', 'grad_norm': '0.003122', 'learning_rate': '2.668e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 11349168, 'tokens/trainable': 171518, 'epoch': '1.465'}
-
 73%|███████▎  | 375/512 [42:09<15:16,  6.69s/it]
 73%|███████▎  | 376/512 [42:16<15:08,  6.68s/it]
                                                 
{'loss': '8.402e-05', 'grad_norm': '0.01218', 'learning_rate': '2.646e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '40.48', 'tokens/total': 11379328, 'tokens/trainable': 172007, 'epoch': '1.469'}
-
 73%|███████▎  | 376/512 [42:16<15:08,  6.68s/it]
 74%|███████▎  | 377/512 [42:23<15:00,  6.67s/it]
                                                 
{'loss': '0.00133', 'grad_norm': '0.2414', 'learning_rate': '2.624e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.71', 'tokens/total': 11409568, 'tokens/trainable': 172468, 'epoch': '1.473'}
-
 74%|███████▎  | 377/512 [42:23<15:00,  6.67s/it]
 74%|███████▍  | 378/512 [42:29<14:54,  6.68s/it]
                                                 
{'loss': '0.0002016', 'grad_norm': '0.02171', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.64', 'tokens/total': 11439968, 'tokens/trainable': 172923, 'epoch': '1.477'}
-
 74%|███████▍  | 378/512 [42:29<14:54,  6.68s/it]
 74%|███████▍  | 379/512 [42:36<14:47,  6.67s/it]
                                                 
{'loss': '0.01576', 'grad_norm': '0.3541', 'learning_rate': '2.579e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 11470496, 'tokens/trainable': 173370, 'epoch': '1.48'}
-
 74%|███████▍  | 379/512 [42:36<14:47,  6.67s/it]
 74%|███████▍  | 380/512 [42:43<14:40,  6.67s/it]
                                                 
{'loss': '2.579e-05', 'grad_norm': '0.005766', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 11500720, 'tokens/trainable': 173837, 'epoch': '1.484'}
-
 74%|███████▍  | 380/512 [42:43<14:40,  6.67s/it]
 74%|███████▍  | 381/512 [42:49<14:33,  6.67s/it]
                                                 
{'loss': '5.822e-05', 'grad_norm': '0.005899', 'learning_rate': '2.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.56', 'tokens/total': 11531280, 'tokens/trainable': 174294, 'epoch': '1.488'}
-
 74%|███████▍  | 381/512 [42:49<14:33,  6.67s/it]
 75%|███████▍  | 382/512 [42:56<14:25,  6.66s/it]
                                                 
{'loss': '4.018e-05', 'grad_norm': '0.005452', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.78', 'tokens/total': 11561264, 'tokens/trainable': 174701, 'epoch': '1.492'}
-
 75%|███████▍  | 382/512 [42:56<14:25,  6.66s/it]
 75%|███████▍  | 383/512 [43:03<14:19,  6.66s/it]
                                                 
{'loss': '3.404e-05', 'grad_norm': '0.00383', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.04', 'tokens/total': 11591616, 'tokens/trainable': 175131, 'epoch': '1.496'}
-
 75%|███████▍  | 383/512 [43:03<14:19,  6.66s/it]
 75%|███████▌  | 384/512 [43:09<14:13,  6.67s/it]
                                                 
{'loss': '3.562e-05', 'grad_norm': '0.004227', 'learning_rate': '2.47e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 11622000, 'tokens/trainable': 175586, 'epoch': '1.5'}
-
 75%|███████▌  | 384/512 [43:09<14:13,  6.67s/it][2026-08-18 13:26:57,790] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
-
 75%|███████▌  | 385/512 [43:18<15:18,  7.24s/it]
                                                 
{'loss': '0.0001464', 'grad_norm': '0.01336', 'learning_rate': '2.449e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 11652352, 'tokens/trainable': 176026, 'epoch': '1.504'}
-
 75%|███████▌  | 385/512 [43:18<15:18,  7.24s/it]
 75%|███████▌  | 386/512 [43:25<14:51,  7.08s/it]
                                                 
{'loss': '8.266e-05', 'grad_norm': '0.008831', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 11682736, 'tokens/trainable': 176512, 'epoch': '1.508'}
-
 75%|███████▌  | 386/512 [43:25<14:51,  7.08s/it]
 76%|███████▌  | 387/512 [43:31<14:31,  6.97s/it]
                                                 
{'loss': '0.0001227', 'grad_norm': '0.009311', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 11713136, 'tokens/trainable': 176943, 'epoch': '1.512'}
-
 76%|███████▌  | 387/512 [43:31<14:31,  6.97s/it]
 76%|███████▌  | 388/512 [43:38<14:12,  6.87s/it]
                                                 
{'loss': '2.322e-05', 'grad_norm': '0.001782', 'learning_rate': '2.385e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.23', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 11743808, 'tokens/trainable': 177370, 'epoch': '1.516'}
-
 76%|███████▌  | 388/512 [43:38<14:12,  6.87s/it]
 76%|███████▌  | 389/512 [43:45<13:58,  6.82s/it]
                                                 
{'loss': '0.0001264', 'grad_norm': '0.0236', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.58', 'tokens/train_per_sec_per_gpu': '35.14', 'tokens/total': 11774144, 'tokens/trainable': 177851, 'epoch': '1.52'}
-
 76%|███████▌  | 389/512 [43:45<13:58,  6.82s/it]
 76%|███████▌  | 390/512 [43:51<13:47,  6.78s/it]
                                                 
{'loss': '0.0004641', 'grad_norm': '0.04384', 'learning_rate': '2.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.68', 'tokens/total': 11804272, 'tokens/trainable': 178310, 'epoch': '1.523'}
-
 76%|███████▌  | 390/512 [43:51<13:47,  6.78s/it]
 76%|███████▋  | 391/512 [43:57<13:19,  6.60s/it]
                                                 
{'loss': '0.0003324', 'grad_norm': '0.06193', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.8', 'tokens/total': 11832432, 'tokens/trainable': 178736, 'epoch': '1.527'}
-
 76%|███████▋  | 391/512 [43:57<13:19,  6.60s/it]
 77%|███████▋  | 392/512 [44:04<13:18,  6.65s/it]
                                                 
{'loss': '0.0005994', 'grad_norm': '0.04984', 'learning_rate': '2.303e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 11862960, 'tokens/trainable': 179233, 'epoch': '1.531'}
-
 77%|███████▋  | 392/512 [44:04<13:18,  6.65s/it]
 77%|███████▋  | 393/512 [44:11<13:14,  6.68s/it]
                                                 
{'loss': '0.0001238', 'grad_norm': '0.01204', 'learning_rate': '2.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.49', 'tokens/total': 11893232, 'tokens/trainable': 179730, 'epoch': '1.535'}
-
 77%|███████▋  | 393/512 [44:11<13:14,  6.68s/it]
 77%|███████▋  | 394/512 [44:18<13:09,  6.69s/it]
                                                 
{'loss': '0.0001039', 'grad_norm': '0.008993', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.81', 'tokens/total': 11923664, 'tokens/trainable': 180227, 'epoch': '1.539'}
-
 77%|███████▋  | 394/512 [44:18<13:09,  6.69s/it]
 77%|███████▋  | 395/512 [44:24<13:04,  6.70s/it]
                                                 
{'loss': '0.003169', 'grad_norm': '0.2463', 'learning_rate': '2.242e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.85', 'tokens/total': 11954032, 'tokens/trainable': 180687, 'epoch': '1.543'}
-
 77%|███████▋  | 395/512 [44:24<13:04,  6.70s/it]
 77%|███████▋  | 396/512 [44:31<12:57,  6.71s/it]
                                                 
{'loss': '0.0005169', 'grad_norm': '0.06872', 'learning_rate': '2.222e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 11984512, 'tokens/trainable': 181141, 'epoch': '1.547'}
-
 77%|███████▋  | 396/512 [44:31<12:57,  6.71s/it]
 78%|███████▊  | 397/512 [44:38<12:53,  6.72s/it]
                                                 
{'loss': '0.000175', 'grad_norm': '0.01167', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 12014880, 'tokens/trainable': 181648, 'epoch': '1.551'}
-
 78%|███████▊  | 397/512 [44:38<12:53,  6.72s/it]
 78%|███████▊  | 398/512 [44:45<12:43,  6.70s/it]
                                                 
{'loss': '0.001466', 'grad_norm': '0.1461', 'learning_rate': '2.183e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 12044928, 'tokens/trainable': 182109, 'epoch': '1.555'}
-
 78%|███████▊  | 398/512 [44:45<12:43,  6.70s/it]
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
                                                 
{'loss': '7.929e-05', 'grad_norm': '0.01636', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.5', 'tokens/total': 12075344, 'tokens/trainable': 182559, 'epoch': '1.559'}
-
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
 78%|███████▊  | 400/512 [44:58<12:27,  6.68s/it]
                                                 
{'loss': '1.526e-05', 'grad_norm': '0.001048', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 12105616, 'tokens/trainable': 182985, 'epoch': '1.562'}
-
 78%|███████▊  | 400/512 [44:58<12:27,  6.68s/it]
 78%|███████▊  | 401/512 [45:05<12:20,  6.67s/it]
                                                 
{'loss': '1.217e-05', 'grad_norm': '0.0005901', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 12136080, 'tokens/trainable': 183440, 'epoch': '1.566'}
-
 78%|███████▊  | 401/512 [45:05<12:20,  6.67s/it]
 79%|███████▊  | 402/512 [45:11<12:14,  6.68s/it]
                                                 
{'loss': '0.000197', 'grad_norm': '0.01319', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '34.52', 'tokens/total': 12166240, 'tokens/trainable': 183917, 'epoch': '1.57'}
-
 79%|███████▊  | 402/512 [45:11<12:14,  6.68s/it]
 79%|███████▊  | 403/512 [45:18<12:07,  6.67s/it]
                                                 
{'loss': '0.0002239', 'grad_norm': '0.02566', 'learning_rate': '2.086e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '30.39', 'tokens/total': 12196544, 'tokens/trainable': 184348, 'epoch': '1.574'}
-
 79%|███████▊  | 403/512 [45:18<12:07,  6.67s/it]
 79%|███████▉  | 404/512 [45:25<12:00,  6.67s/it]
                                                 
{'loss': '8.35e-05', 'grad_norm': '0.008275', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 12226784, 'tokens/trainable': 184786, 'epoch': '1.578'}
-
 79%|███████▉  | 404/512 [45:25<12:00,  6.67s/it]
 79%|███████▉  | 405/512 [45:31<11:54,  6.68s/it]
                                                 
{'loss': '0.0001853', 'grad_norm': '0.03818', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 12257264, 'tokens/trainable': 185249, 'epoch': '1.582'}
-
 79%|███████▉  | 405/512 [45:31<11:54,  6.68s/it]
 79%|███████▉  | 406/512 [45:38<11:48,  6.68s/it]
                                                 
{'loss': '0.0001144', 'grad_norm': '0.007673', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 12287584, 'tokens/trainable': 185680, 'epoch': '1.586'}
-
 79%|███████▉  | 406/512 [45:38<11:48,  6.68s/it]
 79%|███████▉  | 407/512 [45:45<11:40,  6.67s/it]
                                                 
{'loss': '4.649e-05', 'grad_norm': '0.00316', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.72', 'tokens/total': 12317984, 'tokens/trainable': 186100, 'epoch': '1.59'}
-
 79%|███████▉  | 407/512 [45:45<11:40,  6.67s/it]
 80%|███████▉  | 408/512 [45:51<11:33,  6.67s/it]
                                                 
{'loss': '0.00274', 'grad_norm': '0.3274', 'learning_rate': '1.993e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.82', 'tokens/total': 12348144, 'tokens/trainable': 186565, 'epoch': '1.594'}
-
 80%|███████▉  | 408/512 [45:51<11:33,  6.67s/it]
 80%|███████▉  | 409/512 [45:58<11:26,  6.67s/it]
                                                 
{'loss': '1.297e-05', 'grad_norm': '0.0006676', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 12378544, 'tokens/trainable': 187017, 'epoch': '1.598'}
-
 80%|███████▉  | 409/512 [45:58<11:26,  6.67s/it]
 80%|████████  | 410/512 [46:05<11:21,  6.68s/it]
                                                 
{'loss': '3.858e-05', 'grad_norm': '0.003977', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 12408784, 'tokens/trainable': 187468, 'epoch': '1.602'}
-
 80%|████████  | 410/512 [46:05<11:21,  6.68s/it]
 80%|████████  | 411/512 [46:11<11:13,  6.67s/it]
                                                 
{'loss': '2.431e-05', 'grad_norm': '0.002458', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.61', 'tokens/total': 12438832, 'tokens/trainable': 187961, 'epoch': '1.605'}
-
 80%|████████  | 411/512 [46:11<11:13,  6.67s/it]
 80%|████████  | 412/512 [46:18<11:07,  6.68s/it]
                                                 
{'loss': '4.12e-05', 'grad_norm': '0.003755', 'learning_rate': '1.922e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 12469296, 'tokens/trainable': 188447, 'epoch': '1.609'}
-
 80%|████████  | 412/512 [46:18<11:07,  6.68s/it]
 81%|████████  | 413/512 [46:25<11:00,  6.67s/it]
                                                 
{'loss': '0.004735', 'grad_norm': '0.3583', 'learning_rate': '1.904e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 12499392, 'tokens/trainable': 188903, 'epoch': '1.613'}
-
 81%|████████  | 413/512 [46:25<11:00,  6.67s/it]
 81%|████████  | 414/512 [46:31<10:57,  6.70s/it]
                                                 
{'loss': '0.000194', 'grad_norm': '0.04949', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 12529952, 'tokens/trainable': 189398, 'epoch': '1.617'}
-
 81%|████████  | 414/512 [46:31<10:57,  6.70s/it]
 81%|████████  | 415/512 [46:38<10:50,  6.70s/it]
                                                 
{'loss': '2.004e-05', 'grad_norm': '0.003045', 'learning_rate': '1.87e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 12560240, 'tokens/trainable': 189850, 'epoch': '1.621'}
-
 81%|████████  | 415/512 [46:38<10:50,  6.70s/it]
 81%|████████▏ | 416/512 [46:45<10:43,  6.71s/it]
                                                 
{'loss': '0.00162', 'grad_norm': '0.2542', 'learning_rate': '1.853e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '32.64', 'tokens/total': 12590736, 'tokens/trainable': 190306, 'epoch': '1.625'}
-
 81%|████████▏ | 416/512 [46:45<10:43,  6.71s/it][2026-08-18 13:30:33,327] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
-
 81%|████████▏ | 417/512 [46:53<11:28,  7.25s/it]
                                                 
{'loss': '4.322e-05', 'grad_norm': '0.006711', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 12621168, 'tokens/trainable': 190746, 'epoch': '1.629'}
-
 81%|████████▏ | 417/512 [46:53<11:28,  7.25s/it]
 82%|████████▏ | 418/512 [47:00<11:04,  7.07s/it]
                                                 
{'loss': '4.74e-05', 'grad_norm': '0.004932', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '31.07', 'tokens/total': 12651456, 'tokens/trainable': 191156, 'epoch': '1.633'}
-
 82%|████████▏ | 418/512 [47:00<11:04,  7.07s/it]
 82%|████████▏ | 419/512 [47:07<10:46,  6.95s/it]
                                                 
{'loss': '7.545e-05', 'grad_norm': '0.007677', 'learning_rate': '1.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '29.47', 'tokens/total': 12681760, 'tokens/trainable': 191573, 'epoch': '1.637'}
-
 82%|████████▏ | 419/512 [47:07<10:46,  6.95s/it]
 82%|████████▏ | 420/512 [47:13<10:32,  6.87s/it]
                                                 
{'loss': '0.0001045', 'grad_norm': '0.01545', 'learning_rate': '1.786e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 12712144, 'tokens/trainable': 192014, 'epoch': '1.641'}
-
 82%|████████▏ | 420/512 [47:13<10:32,  6.87s/it]
 82%|████████▏ | 421/512 [47:20<10:19,  6.81s/it]
                                                 
{'loss': '0.002415', 'grad_norm': '0.1688', 'learning_rate': '1.77e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.82', 'tokens/total': 12742400, 'tokens/trainable': 192453, 'epoch': '1.645'}
-
 82%|████████▏ | 421/512 [47:20<10:19,  6.81s/it]
 82%|████████▏ | 422/512 [47:27<10:08,  6.76s/it]
                                                 
{'loss': '3.337e-05', 'grad_norm': '0.002461', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 12772288, 'tokens/trainable': 192886, 'epoch': '1.648'}
-
 82%|████████▏ | 422/512 [47:27<10:08,  6.76s/it]
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
                                                 
{'loss': '8.696e-06', 'grad_norm': '0.0005388', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '29.85', 'tokens/total': 12802880, 'tokens/trainable': 193323, 'epoch': '1.652'}
-
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
 83%|████████▎ | 424/512 [47:40<09:56,  6.78s/it]
                                                 
{'loss': '4.861e-05', 'grad_norm': '0.00501', 'learning_rate': '1.722e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.6', 'tokens/total': 12833456, 'tokens/trainable': 193753, 'epoch': '1.656'}
-
 83%|████████▎ | 424/512 [47:40<09:56,  6.78s/it]
 83%|████████▎ | 425/512 [47:47<09:47,  6.75s/it]
                                                 
{'loss': '3.262e-05', 'grad_norm': '0.001982', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 12863952, 'tokens/trainable': 194240, 'epoch': '1.66'}
-
 83%|████████▎ | 425/512 [47:47<09:47,  6.75s/it]
 83%|████████▎ | 426/512 [47:54<09:38,  6.73s/it]
                                                 
{'loss': '0.0005229', 'grad_norm': '0.05306', 'learning_rate': '1.69e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 12894352, 'tokens/trainable': 194680, 'epoch': '1.664'}
-
 83%|████████▎ | 426/512 [47:54<09:38,  6.73s/it]
 83%|████████▎ | 427/512 [48:00<09:31,  6.72s/it]
                                                 
{'loss': '1.543e-05', 'grad_norm': '0.001265', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '38.73', 'tokens/total': 12924688, 'tokens/trainable': 195186, 'epoch': '1.668'}
-
 83%|████████▎ | 427/512 [48:00<09:31,  6.72s/it]
 84%|████████▎ | 428/512 [48:07<09:25,  6.74s/it]
                                                 
{'loss': '2.778e-05', 'grad_norm': '0.001626', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.51', 'tokens/total': 12955072, 'tokens/trainable': 195693, 'epoch': '1.672'}
-
 84%|████████▎ | 428/512 [48:07<09:25,  6.74s/it]
 84%|████████▍ | 429/512 [48:14<09:17,  6.71s/it]
                                                 
{'loss': '6.972e-05', 'grad_norm': '0.01333', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.93', 'tokens/total': 12985328, 'tokens/trainable': 196130, 'epoch': '1.676'}
-
 84%|████████▍ | 429/512 [48:14<09:17,  6.71s/it]
 84%|████████▍ | 430/512 [48:20<09:08,  6.69s/it]
                                                 
{'loss': '4.5e-05', 'grad_norm': '0.003462', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.58', 'tokens/total': 13015600, 'tokens/trainable': 196530, 'epoch': '1.68'}
-
 84%|████████▍ | 430/512 [48:20<09:08,  6.69s/it]
 84%|████████▍ | 431/512 [48:27<09:01,  6.68s/it]
                                                 
{'loss': '0.0002949', 'grad_norm': '0.03424', 'learning_rate': '1.615e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.88', 'tokens/total': 13045728, 'tokens/trainable': 196974, 'epoch': '1.684'}
-
 84%|████████▍ | 431/512 [48:27<09:01,  6.68s/it]
 84%|████████▍ | 432/512 [48:34<08:54,  6.69s/it]
                                                 
{'loss': '7.788e-05', 'grad_norm': '0.00645', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 13075920, 'tokens/trainable': 197454, 'epoch': '1.688'}
-
 84%|████████▍ | 432/512 [48:34<08:54,  6.69s/it]
 85%|████████▍ | 433/512 [48:40<08:48,  6.69s/it]
                                                 
{'loss': '6.603e-06', 'grad_norm': '0.0002191', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '31.4', 'tokens/total': 13106128, 'tokens/trainable': 197902, 'epoch': '1.691'}
-
 85%|████████▍ | 433/512 [48:40<08:48,  6.69s/it]
 85%|████████▍ | 434/512 [48:47<08:40,  6.67s/it]
                                                 
{'loss': '6.426e-06', 'grad_norm': '0.0002612', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '36', 'tokens/total': 13136368, 'tokens/trainable': 198374, 'epoch': '1.695'}
-
 85%|████████▍ | 434/512 [48:47<08:40,  6.67s/it]
 85%|████████▍ | 435/512 [48:54<08:34,  6.68s/it]
                                                 
{'loss': '2.549e-05', 'grad_norm': '0.00164', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.54', 'tokens/total': 13166704, 'tokens/trainable': 198846, 'epoch': '1.699'}
-
 85%|████████▍ | 435/512 [48:54<08:34,  6.68s/it]
 85%|████████▌ | 436/512 [49:00<08:29,  6.70s/it]
                                                 
{'loss': '0.0002141', 'grad_norm': '0.01888', 'learning_rate': '1.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 13197344, 'tokens/trainable': 199320, 'epoch': '1.703'}
-
 85%|████████▌ | 436/512 [49:00<08:29,  6.70s/it]
 85%|████████▌ | 437/512 [49:07<08:21,  6.69s/it]
                                                 
{'loss': '2.316e-05', 'grad_norm': '0.001229', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.43', 'tokens/total': 13227664, 'tokens/trainable': 199777, 'epoch': '1.707'}
-
 85%|████████▌ | 437/512 [49:07<08:21,  6.69s/it]
 86%|████████▌ | 438/512 [49:14<08:13,  6.67s/it]
                                                 
{'loss': '4.594e-05', 'grad_norm': '0.005586', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '30.82', 'tokens/total': 13257632, 'tokens/trainable': 200194, 'epoch': '1.711'}
-
 86%|████████▌ | 438/512 [49:14<08:13,  6.67s/it]
 86%|████████▌ | 439/512 [49:21<08:09,  6.71s/it]
                                                 
{'loss': '2.285e-05', 'grad_norm': '0.001554', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 13288240, 'tokens/trainable': 200663, 'epoch': '1.715'}
-
 86%|████████▌ | 439/512 [49:21<08:09,  6.71s/it]
 86%|████████▌ | 440/512 [49:27<08:05,  6.74s/it]
                                                 
{'loss': '4.093e-05', 'grad_norm': '0.002979', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.79', 'tokens/total': 13318992, 'tokens/trainable': 201121, 'epoch': '1.719'}
-
 86%|████████▌ | 440/512 [49:27<08:05,  6.74s/it]
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
                                                 
{'loss': '7.297e-06', 'grad_norm': '0.001739', 'learning_rate': '1.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.74', 'tokens/total': 13349392, 'tokens/trainable': 201598, 'epoch': '1.723'}
-
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
 86%|████████▋ | 442/512 [49:41<07:50,  6.72s/it]
                                                 
{'loss': '1.716e-05', 'grad_norm': '0.0009943', 'learning_rate': '1.464e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 13379648, 'tokens/trainable': 202068, 'epoch': '1.727'}
-
 86%|████████▋ | 442/512 [49:41<07:50,  6.72s/it]
 87%|████████▋ | 443/512 [49:47<07:41,  6.70s/it]
                                                 
{'loss': '2.382e-05', 'grad_norm': '0.001147', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 13409936, 'tokens/trainable': 202524, 'epoch': '1.73'}
-
 87%|████████▋ | 443/512 [49:47<07:41,  6.70s/it]
 87%|████████▋ | 444/512 [49:54<07:35,  6.69s/it]
                                                 
{'loss': '6.557e-05', 'grad_norm': '0.006799', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 13440288, 'tokens/trainable': 202996, 'epoch': '1.734'}
-
 87%|████████▋ | 444/512 [49:54<07:35,  6.69s/it]
 87%|████████▋ | 445/512 [50:00<07:19,  6.56s/it]
                                                 
{'loss': '1.839e-05', 'grad_norm': '0.001165', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.14', 'tokens/total': 13468880, 'tokens/trainable': 203470, 'epoch': '1.738'}
-
 87%|████████▋ | 445/512 [50:00<07:19,  6.56s/it]
 87%|████████▋ | 446/512 [50:07<07:16,  6.61s/it]
                                                 
{'loss': '0.000164', 'grad_norm': '0.02089', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 13499184, 'tokens/trainable': 203919, 'epoch': '1.742'}
-
 87%|████████▋ | 446/512 [50:07<07:16,  6.61s/it]
 87%|████████▋ | 447/512 [50:14<07:10,  6.62s/it]
                                                 
{'loss': '0.001068', 'grad_norm': '0.09616', 'learning_rate': '1.402e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.3', 'tokens/total': 13529632, 'tokens/trainable': 204376, 'epoch': '1.746'}
-
 87%|████████▋ | 447/512 [50:14<07:10,  6.62s/it]
 88%|████████▊ | 448/512 [50:20<07:05,  6.64s/it]
                                                 
{'loss': '0.001196', 'grad_norm': '0.09744', 'learning_rate': '1.39e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32', 'tokens/total': 13560080, 'tokens/trainable': 204821, 'epoch': '1.75'}
-
 88%|████████▊ | 448/512 [50:20<07:05,  6.64s/it][2026-08-18 13:34:08,932] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
-
 88%|████████▊ | 449/512 [50:29<07:36,  7.24s/it]
                                                 
{'loss': '8.02e-06', 'grad_norm': '0.0007811', 'learning_rate': '1.378e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.63', 'tokens/total': 13590496, 'tokens/trainable': 205271, 'epoch': '1.754'}
-
 88%|████████▊ | 449/512 [50:29<07:36,  7.24s/it]
 88%|████████▊ | 450/512 [50:36<07:18,  7.07s/it]
                                                 
{'loss': '0.0001024', 'grad_norm': '0.03409', 'learning_rate': '1.367e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '30.95', 'tokens/total': 13620960, 'tokens/trainable': 205723, 'epoch': '1.758'}
-
 88%|████████▊ | 450/512 [50:36<07:18,  7.07s/it]
 88%|████████▊ | 451/512 [50:42<07:04,  6.96s/it]
                                                 
{'loss': '0.0002193', 'grad_norm': '0.02931', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.42', 'tokens/total': 13651344, 'tokens/trainable': 206169, 'epoch': '1.762'}
-
 88%|████████▊ | 451/512 [50:42<07:04,  6.96s/it]
 88%|████████▊ | 452/512 [50:49<06:54,  6.90s/it]
                                                 
{'loss': '8e-05', 'grad_norm': '0.01096', 'learning_rate': '1.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 13681792, 'tokens/trainable': 206592, 'epoch': '1.766'}
-
 88%|████████▊ | 452/512 [50:49<06:54,  6.90s/it]
 88%|████████▊ | 453/512 [50:56<06:43,  6.83s/it]
                                                 
{'loss': '3.79e-05', 'grad_norm': '0.004453', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 13712256, 'tokens/trainable': 207055, 'epoch': '1.77'}
-
 88%|████████▊ | 453/512 [50:56<06:43,  6.83s/it]
 89%|████████▊ | 454/512 [51:02<06:31,  6.76s/it]
                                                 
{'loss': '1.224e-05', 'grad_norm': '0.000565', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.65', 'tokens/total': 13742128, 'tokens/trainable': 207494, 'epoch': '1.773'}
-
 89%|████████▊ | 454/512 [51:02<06:31,  6.76s/it]
 89%|████████▉ | 455/512 [51:09<06:23,  6.72s/it]
                                                 
{'loss': '1.357e-05', 'grad_norm': '0.0007141', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 13772160, 'tokens/trainable': 207956, 'epoch': '1.777'}
-
 89%|████████▉ | 455/512 [51:09<06:23,  6.72s/it]
 89%|████████▉ | 456/512 [51:16<06:15,  6.71s/it]
                                                 
{'loss': '1.166e-05', 'grad_norm': '0.0008097', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 13802448, 'tokens/trainable': 208408, 'epoch': '1.781'}
-
 89%|████████▉ | 456/512 [51:16<06:15,  6.71s/it]
 89%|████████▉ | 457/512 [51:22<06:08,  6.70s/it]
                                                 
{'loss': '9.201e-06', 'grad_norm': '0.001265', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '36', 'tokens/total': 13832880, 'tokens/trainable': 208889, 'epoch': '1.785'}
-
 89%|████████▉ | 457/512 [51:22<06:08,  6.70s/it]
 89%|████████▉ | 458/512 [51:29<06:00,  6.68s/it]
                                                 
{'loss': '0.0001209', 'grad_norm': '0.01498', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.08', 'tokens/total': 13862880, 'tokens/trainable': 209336, 'epoch': '1.789'}
-
 89%|████████▉ | 458/512 [51:29<06:00,  6.68s/it]
 90%|████████▉ | 459/512 [51:36<05:54,  6.68s/it]
                                                 
{'loss': '3.241e-05', 'grad_norm': '0.004875', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.26', 'tokens/total': 13893440, 'tokens/trainable': 209797, 'epoch': '1.793'}
-
 90%|████████▉ | 459/512 [51:36<05:54,  6.68s/it]
 90%|████████▉ | 460/512 [51:42<05:47,  6.68s/it]
                                                 
{'loss': '1.98e-05', 'grad_norm': '0.0009628', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.12', 'tokens/total': 13923824, 'tokens/trainable': 210240, 'epoch': '1.797'}
-
 90%|████████▉ | 460/512 [51:42<05:47,  6.68s/it]
 90%|█████████ | 461/512 [51:49<05:41,  6.69s/it]
                                                 
{'loss': '0.0001068', 'grad_norm': '0.01991', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 13954208, 'tokens/trainable': 210702, 'epoch': '1.801'}
-
 90%|█████████ | 461/512 [51:49<05:41,  6.69s/it]
 90%|█████████ | 462/512 [51:56<05:34,  6.69s/it]
                                                 
{'loss': '1.108e-05', 'grad_norm': '0.000537', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 13984784, 'tokens/trainable': 211140, 'epoch': '1.805'}
-
 90%|█████████ | 462/512 [51:56<05:34,  6.69s/it]
 90%|█████████ | 463/512 [52:02<05:27,  6.68s/it]
                                                 
{'loss': '1.551e-05', 'grad_norm': '0.0008154', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.6', 'tokens/total': 14014896, 'tokens/trainable': 211575, 'epoch': '1.809'}
-
 90%|█████████ | 463/512 [52:02<05:27,  6.68s/it]
 91%|█████████ | 464/512 [52:09<05:21,  6.69s/it]
                                                 
{'loss': '0.0001314', 'grad_norm': '0.02876', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '39.03', 'tokens/total': 14045312, 'tokens/trainable': 212083, 'epoch': '1.812'}
-
 91%|█████████ | 464/512 [52:09<05:21,  6.69s/it]
 91%|█████████ | 465/512 [52:16<05:15,  6.71s/it]
                                                 
{'loss': '0.000297', 'grad_norm': '0.04559', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 14075840, 'tokens/trainable': 212558, 'epoch': '1.816'}
-
 91%|█████████ | 465/512 [52:16<05:15,  6.71s/it]
 91%|█████████ | 466/512 [52:23<05:07,  6.69s/it]
                                                 
{'loss': '9.938e-06', 'grad_norm': '0.0004123', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 14106016, 'tokens/trainable': 213014, 'epoch': '1.82'}
-
 91%|█████████ | 466/512 [52:23<05:07,  6.69s/it]
 91%|█████████ | 467/512 [52:29<05:01,  6.70s/it]
                                                 
{'loss': '1.558e-05', 'grad_norm': '0.001011', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.99', 'tokens/train_per_sec_per_gpu': '34.03', 'tokens/total': 14136512, 'tokens/trainable': 213465, 'epoch': '1.824'}
-
 91%|█████████ | 467/512 [52:29<05:01,  6.70s/it]
 91%|█████████▏| 468/512 [52:36<04:54,  6.70s/it]
                                                 
{'loss': '3.261e-05', 'grad_norm': '0.00431', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.62', 'tokens/total': 14166832, 'tokens/trainable': 213940, 'epoch': '1.828'}
-
 91%|█████████▏| 468/512 [52:36<04:54,  6.70s/it]
 92%|█████████▏| 469/512 [52:43<04:49,  6.72s/it]
                                                 
{'loss': '2.145e-05', 'grad_norm': '0.005317', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 14197360, 'tokens/trainable': 214415, 'epoch': '1.832'}
-
 92%|█████████▏| 469/512 [52:43<04:49,  6.72s/it]
 92%|█████████▏| 470/512 [52:50<04:42,  6.72s/it]
                                                 
{'loss': '5.694e-06', 'grad_norm': '0.0003265', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 14227600, 'tokens/trainable': 214880, 'epoch': '1.836'}
-
 92%|█████████▏| 470/512 [52:50<04:42,  6.72s/it]
 92%|█████████▏| 471/512 [52:56<04:35,  6.71s/it]
                                                 
{'loss': '0.0001062', 'grad_norm': '0.01372', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 14257888, 'tokens/trainable': 215334, 'epoch': '1.84'}
-
 92%|█████████▏| 471/512 [52:56<04:35,  6.71s/it]
 92%|█████████▏| 472/512 [53:03<04:28,  6.71s/it]
                                                 
{'loss': '4.587e-06', 'grad_norm': '0.0002091', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 14288208, 'tokens/trainable': 215822, 'epoch': '1.844'}
-
 92%|█████████▏| 472/512 [53:03<04:28,  6.71s/it]
 92%|█████████▏| 473/512 [53:10<04:21,  6.70s/it]
                                                 
{'loss': '0.0006915', 'grad_norm': '0.1067', 'learning_rate': '1.149e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '30.86', 'tokens/total': 14318448, 'tokens/trainable': 216239, 'epoch': '1.848'}
-
 92%|█████████▏| 473/512 [53:10<04:21,  6.70s/it]
 93%|█████████▎| 474/512 [53:16<04:14,  6.69s/it]
                                                 
{'loss': '6.862e-05', 'grad_norm': '0.01328', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.37', 'tokens/total': 14348912, 'tokens/trainable': 216691, 'epoch': '1.852'}
-
 93%|█████████▎| 474/512 [53:16<04:14,  6.69s/it]
 93%|█████████▎| 475/512 [53:23<04:07,  6.68s/it]
                                                 
{'loss': '8.334e-05', 'grad_norm': '0.007407', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 14379440, 'tokens/trainable': 217162, 'epoch': '1.855'}
-
 93%|█████████▎| 475/512 [53:23<04:07,  6.68s/it]
 93%|█████████▎| 476/512 [53:30<04:01,  6.71s/it]
                                                 
{'loss': '1.804e-05', 'grad_norm': '0.001777', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 14409840, 'tokens/trainable': 217638, 'epoch': '1.859'}
-
 93%|█████████▎| 476/512 [53:30<04:01,  6.71s/it]
 93%|█████████▎| 477/512 [53:36<03:54,  6.71s/it]
                                                 
{'loss': '9.578e-06', 'grad_norm': '0.0004483', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 14440256, 'tokens/trainable': 218053, 'epoch': '1.863'}
-
 93%|█████████▎| 477/512 [53:36<03:54,  6.71s/it]
 93%|█████████▎| 478/512 [53:43<03:47,  6.70s/it]
                                                 
{'loss': '1.095e-05', 'grad_norm': '0.000956', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 14470704, 'tokens/trainable': 218503, 'epoch': '1.867'}
-
 93%|█████████▎| 478/512 [53:43<03:47,  6.70s/it]
 94%|█████████▎| 479/512 [53:50<03:40,  6.70s/it]
                                                 
{'loss': '0.00123', 'grad_norm': '0.2087', 'learning_rate': '1.108e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.19', 'tokens/total': 14501072, 'tokens/trainable': 218964, 'epoch': '1.871'}
-
 94%|█████████▎| 479/512 [53:50<03:40,  6.70s/it]
 94%|█████████▍| 480/512 [53:56<03:33,  6.68s/it]
                                                 
{'loss': '0.0001162', 'grad_norm': '0.01922', 'learning_rate': '1.102e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 14531200, 'tokens/trainable': 219412, 'epoch': '1.875'}
-
 94%|█████████▍| 480/512 [53:56<03:33,  6.68s/it][2026-08-18 13:37:44,940] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
-
 94%|█████████▍| 481/512 [54:05<03:44,  7.24s/it]
                                                 
{'loss': '2.362e-05', 'grad_norm': '0.002075', 'learning_rate': '1.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '38.29', 'tokens/total': 14561152, 'tokens/trainable': 219886, 'epoch': '1.879'}
-
 94%|█████████▍| 481/512 [54:05<03:44,  7.24s/it]
 94%|█████████▍| 482/512 [54:12<03:32,  7.07s/it]
                                                 
{'loss': '5.528e-05', 'grad_norm': '0.01115', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 14591264, 'tokens/trainable': 220343, 'epoch': '1.883'}
-
 94%|█████████▍| 482/512 [54:12<03:32,  7.07s/it]
 94%|█████████▍| 483/512 [54:18<03:21,  6.96s/it]
                                                 
{'loss': '6.712e-05', 'grad_norm': '0.01091', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 14621568, 'tokens/trainable': 220825, 'epoch': '1.887'}
-
 94%|█████████▍| 483/512 [54:18<03:21,  6.96s/it]
 95%|█████████▍| 484/512 [54:25<03:12,  6.88s/it]
                                                 
{'loss': '0.002836', 'grad_norm': '0.3236', 'learning_rate': '1.079e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '32.52', 'tokens/total': 14652064, 'tokens/trainable': 221282, 'epoch': '1.891'}
-
 95%|█████████▍| 484/512 [54:25<03:12,  6.88s/it]
 95%|█████████▍| 485/512 [54:32<03:04,  6.83s/it]
                                                 
{'loss': '5.301e-06', 'grad_norm': '0.0002806', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.86', 'tokens/total': 14682608, 'tokens/trainable': 221780, 'epoch': '1.895'}
-
 95%|█████████▍| 485/512 [54:32<03:04,  6.83s/it]
 95%|█████████▍| 486/512 [54:39<02:57,  6.81s/it]
                                                 
{'loss': '3.665e-05', 'grad_norm': '0.009112', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 14713168, 'tokens/trainable': 222210, 'epoch': '1.898'}
-
 95%|█████████▍| 486/512 [54:39<02:57,  6.81s/it]
 95%|█████████▌| 487/512 [54:45<02:49,  6.77s/it]
                                                 
{'loss': '0.0002056', 'grad_norm': '0.04485', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.33', 'tokens/total': 14743536, 'tokens/trainable': 222646, 'epoch': '1.902'}
-
 95%|█████████▌| 487/512 [54:45<02:49,  6.77s/it]
 95%|█████████▌| 488/512 [54:52<02:41,  6.74s/it]
                                                 
{'loss': '0.008499', 'grad_norm': '0.2785', 'learning_rate': '1.058e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 14773728, 'tokens/trainable': 223066, 'epoch': '1.906'}
-
 95%|█████████▌| 488/512 [54:52<02:41,  6.74s/it]
 96%|█████████▌| 489/512 [54:59<02:34,  6.72s/it]
                                                 
{'loss': '9.457e-06', 'grad_norm': '0.0006374', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 14803744, 'tokens/trainable': 223512, 'epoch': '1.91'}
-
 96%|█████████▌| 489/512 [54:59<02:34,  6.72s/it]
 96%|█████████▌| 490/512 [55:05<02:24,  6.56s/it]
                                                 
{'loss': '6.928e-06', 'grad_norm': '0.000387', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.96', 'tokens/total': 14832112, 'tokens/trainable': 223956, 'epoch': '1.914'}
-
 96%|█████████▌| 490/512 [55:05<02:24,  6.56s/it]
 96%|█████████▌| 491/512 [55:11<02:15,  6.45s/it]
                                                 
{'loss': '2.272e-05', 'grad_norm': '0.001532', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 14860560, 'tokens/trainable': 224416, 'epoch': '1.918'}
-
 96%|█████████▌| 491/512 [55:11<02:15,  6.45s/it]
 96%|█████████▌| 492/512 [55:18<02:10,  6.52s/it]
                                                 
{'loss': '0.0001193', 'grad_norm': '0.01609', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 14890848, 'tokens/trainable': 224883, 'epoch': '1.922'}
-
 96%|█████████▌| 492/512 [55:18<02:10,  6.52s/it]
 96%|█████████▋| 493/512 [55:24<02:04,  6.56s/it]
                                                 
{'loss': '1.477e-05', 'grad_norm': '0.001622', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.47', 'tokens/total': 14921008, 'tokens/trainable': 225324, 'epoch': '1.926'}
-
 96%|█████████▋| 493/512 [55:24<02:04,  6.56s/it]
 96%|█████████▋| 494/512 [55:31<01:58,  6.60s/it]
                                                 
{'loss': '3.447e-05', 'grad_norm': '0.004443', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.9', 'tokens/total': 14951360, 'tokens/trainable': 225783, 'epoch': '1.93'}
-
 96%|█████████▋| 494/512 [55:31<01:58,  6.60s/it]
 97%|█████████▋| 495/512 [55:38<01:52,  6.63s/it]
                                                 
{'loss': '1.805e-05', 'grad_norm': '0.002437', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 14981552, 'tokens/trainable': 226281, 'epoch': '1.934'}
-
 97%|█████████▋| 495/512 [55:38<01:52,  6.63s/it]
 97%|█████████▋| 496/512 [55:44<01:45,  6.62s/it]
                                                 
{'loss': '1.692e-05', 'grad_norm': '0.00158', 'learning_rate': '1.027e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.72', 'tokens/total': 15011584, 'tokens/trainable': 226735, 'epoch': '1.938'}
-
 97%|█████████▋| 496/512 [55:44<01:45,  6.62s/it]
 97%|█████████▋| 497/512 [55:51<01:39,  6.64s/it]
                                                 
{'loss': '4.592e-05', 'grad_norm': '0.003621', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.31', 'tokens/total': 15042016, 'tokens/trainable': 227198, 'epoch': '1.941'}
-
 97%|█████████▋| 497/512 [55:51<01:39,  6.64s/it]
 97%|█████████▋| 498/512 [55:58<01:32,  6.63s/it]
                                                 
{'loss': '1.531e-05', 'grad_norm': '0.0005919', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.07', 'tokens/total': 15071984, 'tokens/trainable': 227623, 'epoch': '1.945'}
-
 97%|█████████▋| 498/512 [55:58<01:32,  6.63s/it]
 97%|█████████▋| 499/512 [56:04<01:26,  6.65s/it]
                                                 
{'loss': '1.349e-05', 'grad_norm': '0.001367', 'learning_rate': '1.018e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '32.15', 'tokens/total': 15102192, 'tokens/trainable': 228074, 'epoch': '1.949'}
-
 97%|█████████▋| 499/512 [56:04<01:26,  6.65s/it]
 98%|█████████▊| 500/512 [56:11<01:19,  6.66s/it]
                                                 
{'loss': '2.841e-05', 'grad_norm': '0.002793', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 15132560, 'tokens/trainable': 228541, 'epoch': '1.953'}
-
 98%|█████████▊| 500/512 [56:11<01:19,  6.66s/it]
 98%|█████████▊| 501/512 [56:18<01:13,  6.67s/it]
                                                 
{'loss': '3.183e-05', 'grad_norm': '0.002832', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.77', 'tokens/total': 15163152, 'tokens/trainable': 229031, 'epoch': '1.957'}
-
 98%|█████████▊| 501/512 [56:18<01:13,  6.67s/it]
 98%|█████████▊| 502/512 [56:24<01:06,  6.68s/it]
                                                 
{'loss': '2.238e-05', 'grad_norm': '0.003093', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.68', 'tokens/total': 15193456, 'tokens/trainable': 229512, 'epoch': '1.961'}
-
 98%|█████████▊| 502/512 [56:24<01:06,  6.68s/it]
 98%|█████████▊| 503/512 [56:31<01:00,  6.67s/it]
                                                 
{'loss': '0.0004141', 'grad_norm': '0.0639', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 15223680, 'tokens/trainable': 229989, 'epoch': '1.965'}
-
 98%|█████████▊| 503/512 [56:31<01:00,  6.67s/it]
 98%|█████████▊| 504/512 [56:38<00:53,  6.67s/it]
                                                 
{'loss': '4.398e-05', 'grad_norm': '0.003928', 'learning_rate': '1.008e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.81', 'tokens/total': 15254336, 'tokens/trainable': 230447, 'epoch': '1.969'}
-
 98%|█████████▊| 504/512 [56:38<00:53,  6.67s/it]
 99%|█████████▊| 505/512 [56:44<00:46,  6.68s/it]
                                                 
{'loss': '2.18e-05', 'grad_norm': '0.002279', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 15284496, 'tokens/trainable': 230879, 'epoch': '1.973'}
-
 99%|█████████▊| 505/512 [56:44<00:46,  6.68s/it]
 99%|█████████▉| 506/512 [56:51<00:40,  6.69s/it]
                                                 
{'loss': '2.01e-05', 'grad_norm': '0.002216', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 15314976, 'tokens/trainable': 231348, 'epoch': '1.977'}
-
 99%|█████████▉| 506/512 [56:51<00:40,  6.69s/it]
 99%|█████████▉| 507/512 [56:58<00:33,  6.69s/it]
                                                 
{'loss': '9.908e-05', 'grad_norm': '0.01093', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.22', 'tokens/total': 15345280, 'tokens/trainable': 231819, 'epoch': '1.98'}
-
 99%|█████████▉| 507/512 [56:58<00:33,  6.69s/it]
 99%|█████████▉| 508/512 [57:04<00:26,  6.67s/it]
                                                 
{'loss': '1.419e-05', 'grad_norm': '0.0008611', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.57', 'tokens/total': 15375424, 'tokens/trainable': 232269, 'epoch': '1.984'}
-
 99%|█████████▉| 508/512 [57:04<00:26,  6.67s/it]
 99%|█████████▉| 509/512 [57:11<00:20,  6.67s/it]
                                                 
{'loss': '0.001132', 'grad_norm': '0.1847', 'learning_rate': '1.001e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 15405872, 'tokens/trainable': 232716, 'epoch': '1.988'}
-
 99%|█████████▉| 509/512 [57:11<00:20,  6.67s/it]
100%|█████████▉| 510/512 [57:18<00:13,  6.68s/it]
                                                 
{'loss': '1.466e-05', 'grad_norm': '0.001229', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 15436128, 'tokens/trainable': 233178, 'epoch': '1.992'}
-
100%|█████████▉| 510/512 [57:18<00:13,  6.68s/it]
100%|█████████▉| 511/512 [57:24<00:06,  6.68s/it]
                                                 
{'loss': '2.096e-05', 'grad_norm': '0.003051', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 15466464, 'tokens/trainable': 233647, 'epoch': '1.996'}
-
100%|█████████▉| 511/512 [57:24<00:06,  6.68s/it]
100%|██████████| 512/512 [57:31<00:00,  6.70s/it]
                                                 
{'loss': '1.928e-05', 'grad_norm': '0.00193', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.42', 'tokens/total': 15496368, 'tokens/trainable': 234060, 'epoch': '2'}
-
100%|██████████| 512/512 [57:31<00:00,  6.70s/it][2026-08-18 13:41:19,700] [INFO] [axolotl.core.trainers.base._save:828] [PID:3104] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
-
                                                 
{'train_runtime': '3454', 'train_samples_per_second': '4.744', 'train_steps_per_second': '0.148', 'train_loss': '0.007604', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '38.44', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
-
100%|██████████| 512/512 [57:33<00:00,  6.70s/it]
100%|██████████| 512/512 [57:33<00:00,  6.75s/it]
-[2026-08-18 13:41:21,525] [INFO] [axolotl.train.save_trained_model:267] [PID:3104] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
-[2026-08-18 13:41:21,960] [INFO] [axolotl.train.save_trained_model:388] [PID:3104] Model successfully saved to /workspace/wave/training/checkpoints
+[2026-08-18 14:04:56,316] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:9922] after adapters 0.000GB ()
+[2026-08-18 14:04:57,312] [INFO] [axolotl.monkeypatch.lora_kernels.apply_lora_kernel_patches:418] [PID:9922] LoRA kernels: dropout=0.05 enabled
+[2026-08-18 14:05:01,950] [INFO] [axolotl.train.save_initial_configs:450] [PID:9922] Pre-saving adapter config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:01,950] [INFO] [axolotl.train.save_initial_configs:454] [PID:9922] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:02,286] [INFO] [axolotl.train.save_initial_configs:459] [PID:9922] Pre-saving model config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:02,290] [INFO] [axolotl.train.save_initial_configs:463] [PID:9922] Pre-saving processor to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:02,604] [INFO] [axolotl.train.execute_training:226] [PID:9922] Starting trainer...
+
  0%|          | 0/512 [00:00<?, ?it/s]
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
                                                 
{'loss': '0.1025', 'grad_norm': '0.8591', 'learning_rate': '0', 'ppl': '1.108', 'memory/max_active (GiB)': '32.98', 'memory/max_allocated (GiB)': '32.98', 'memory/device_reserved (GiB)': '34.94', 'tokens/train_per_sec_per_gpu': '30.01', 'tokens/total': 30464, 'tokens/trainable': 470, 'epoch': '0.003906'}
+
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
  0%|          | 2/512 [00:14<1:01:50,  7.28s/it]
                                                 
{'loss': '0.09678', 'grad_norm': '0.8034', 'learning_rate': '4e-06', 'ppl': '1.102', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.83', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 60800, 'tokens/trainable': 922, 'epoch': '0.007812'}
+
  0%|          | 2/512 [00:14<1:01:50,  7.28s/it]
  1%|          | 3/512 [00:21<59:21,  7.00s/it]  
                                               
{'loss': '0.1088', 'grad_norm': '1.01', 'learning_rate': '8e-06', 'ppl': '1.115', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '34.67', 'tokens/total': 91136, 'tokens/trainable': 1396, 'epoch': '0.01172'}
+
  1%|          | 3/512 [00:21<59:21,  7.00s/it]
  1%|          | 4/512 [00:28<58:17,  6.88s/it]
                                               
{'loss': '0.1031', 'grad_norm': '2.204', 'learning_rate': '1.2e-05', 'ppl': '1.109', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '38', 'tokens/total': 121552, 'tokens/trainable': 1850, 'epoch': '0.01562'}
+
  1%|          | 4/512 [00:28<58:17,  6.88s/it]
  1%|          | 5/512 [00:35<57:58,  6.86s/it]
                                               
{'loss': '0.1097', 'grad_norm': '2.576', 'learning_rate': '1.6e-05', 'ppl': '1.116', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 152304, 'tokens/trainable': 2302, 'epoch': '0.01953'}
+
  1%|          | 5/512 [00:35<57:58,  6.86s/it]
  1%|          | 6/512 [00:41<57:18,  6.80s/it]
                                               
{'loss': '0.08722', 'grad_norm': '1.498', 'learning_rate': '2e-05', 'ppl': '1.091', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '31.65', 'tokens/total': 182448, 'tokens/trainable': 2742, 'epoch': '0.02344'}
+
  1%|          | 6/512 [00:41<57:18,  6.80s/it]
  1%|▏         | 7/512 [00:48<56:56,  6.77s/it]
                                               
{'loss': '0.07384', 'grad_norm': '1.28', 'learning_rate': '2.4e-05', 'ppl': '1.077', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 212736, 'tokens/trainable': 3208, 'epoch': '0.02734'}
+
  1%|▏         | 7/512 [00:48<56:56,  6.77s/it]
  2%|▏         | 8/512 [00:55<56:34,  6.74s/it]
                                               
{'loss': '0.08001', 'grad_norm': '1.695', 'learning_rate': '2.8e-05', 'ppl': '1.083', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 243024, 'tokens/trainable': 3655, 'epoch': '0.03125'}
+
  2%|▏         | 8/512 [00:55<56:34,  6.74s/it]
  2%|▏         | 9/512 [01:01<55:00,  6.56s/it]
                                               
{'loss': '0.04736', 'grad_norm': '1.222', 'learning_rate': '3.2e-05', 'ppl': '1.048', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '35.78', 'tokens/total': 271264, 'tokens/trainable': 4091, 'epoch': '0.03516'}
+
  2%|▏         | 9/512 [01:01<55:00,  6.56s/it]
  2%|▏         | 10/512 [01:07<55:18,  6.61s/it]
                                                
{'loss': '0.0957', 'grad_norm': '2.902', 'learning_rate': '3.6e-05', 'ppl': '1.1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '34.98', 'tokens/total': 301520, 'tokens/trainable': 4553, 'epoch': '0.03906'}
+
  2%|▏         | 10/512 [01:07<55:18,  6.61s/it]
  2%|▏         | 11/512 [01:14<55:21,  6.63s/it]
                                                
{'loss': '0.08283', 'grad_norm': '2.177', 'learning_rate': '4e-05', 'ppl': '1.086', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.08', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 331808, 'tokens/trainable': 4992, 'epoch': '0.04297'}
+
  2%|▏         | 11/512 [01:14<55:21,  6.63s/it]
  2%|▏         | 12/512 [01:21<55:35,  6.67s/it]
                                                
{'loss': '0.1214', 'grad_norm': '3.152', 'learning_rate': '4.4e-05', 'ppl': '1.129', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.08', 'tokens/train_per_sec_per_gpu': '38.32', 'tokens/total': 362224, 'tokens/trainable': 5494, 'epoch': '0.04688'}
+
  2%|▏         | 12/512 [01:21<55:35,  6.67s/it]
  3%|▎         | 13/512 [01:28<55:17,  6.65s/it]
                                                
{'loss': '0.03794', 'grad_norm': '2.383', 'learning_rate': '4.8e-05', 'ppl': '1.039', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.04', 'tokens/total': 392432, 'tokens/trainable': 5933, 'epoch': '0.05078'}
+
  3%|▎         | 13/512 [01:28<55:17,  6.65s/it]
  3%|▎         | 14/512 [01:34<55:19,  6.67s/it]
                                                
{'loss': '0.06515', 'grad_norm': '3.259', 'learning_rate': '5.2e-05', 'ppl': '1.067', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '29.67', 'tokens/total': 422784, 'tokens/trainable': 6369, 'epoch': '0.05469'}
+
  3%|▎         | 14/512 [01:34<55:19,  6.67s/it]
  3%|▎         | 15/512 [01:41<55:11,  6.66s/it]
                                                
{'loss': '0.05657', 'grad_norm': '1.582', 'learning_rate': '5.6e-05', 'ppl': '1.058', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 453376, 'tokens/trainable': 6820, 'epoch': '0.05859'}
+
  3%|▎         | 15/512 [01:41<55:11,  6.66s/it]
  3%|▎         | 16/512 [01:48<55:02,  6.66s/it]
                                                
{'loss': '0.0607', 'grad_norm': '1.422', 'learning_rate': '6e-05', 'ppl': '1.063', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.85', 'tokens/total': 483504, 'tokens/trainable': 7258, 'epoch': '0.0625'}
+
  3%|▎         | 16/512 [01:48<55:02,  6.66s/it]
  3%|▎         | 17/512 [01:54<55:08,  6.68s/it]
                                                
{'loss': '0.05027', 'grad_norm': '1.307', 'learning_rate': '6.4e-05', 'ppl': '1.052', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 514032, 'tokens/trainable': 7710, 'epoch': '0.06641'}
+
  3%|▎         | 17/512 [01:54<55:08,  6.68s/it]
  4%|▎         | 18/512 [02:01<55:05,  6.69s/it]
                                                
{'loss': '0.0285', 'grad_norm': '0.6123', 'learning_rate': '6.8e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '31.76', 'tokens/total': 544432, 'tokens/trainable': 8174, 'epoch': '0.07031'}
+
  4%|▎         | 18/512 [02:01<55:05,  6.69s/it]
  4%|▎         | 19/512 [02:08<54:56,  6.69s/it]
                                                
{'loss': '0.03114', 'grad_norm': '0.6484', 'learning_rate': '7.2e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 574880, 'tokens/trainable': 8617, 'epoch': '0.07422'}
+
  4%|▎         | 19/512 [02:08<54:56,  6.69s/it]
  4%|▍         | 20/512 [02:14<54:45,  6.68s/it]
                                                
{'loss': '0.03075', 'grad_norm': '0.6737', 'learning_rate': '7.6e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.37', 'tokens/total': 605408, 'tokens/trainable': 9037, 'epoch': '0.07812'}
+
  4%|▍         | 20/512 [02:14<54:45,  6.68s/it]
  4%|▍         | 21/512 [02:21<54:40,  6.68s/it]
                                                
{'loss': '0.02451', 'grad_norm': '0.7465', 'learning_rate': '8e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.63', 'tokens/total': 635584, 'tokens/trainable': 9503, 'epoch': '0.08203'}
+
  4%|▍         | 21/512 [02:21<54:40,  6.68s/it]
  4%|▍         | 22/512 [02:28<54:38,  6.69s/it]
                                                
{'loss': '0.01763', 'grad_norm': '0.9436', 'learning_rate': '8.4e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.79', 'tokens/total': 665952, 'tokens/trainable': 9972, 'epoch': '0.08594'}
+
  4%|▍         | 22/512 [02:28<54:38,  6.69s/it]
  4%|▍         | 23/512 [02:34<54:39,  6.71s/it]
                                                
{'loss': '0.01396', 'grad_norm': '0.637', 'learning_rate': '8.8e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '37.54', 'tokens/total': 696240, 'tokens/trainable': 10447, 'epoch': '0.08984'}
+
  4%|▍         | 23/512 [02:34<54:39,  6.71s/it]
  5%|▍         | 24/512 [02:41<54:22,  6.69s/it]
                                                
{'loss': '0.03303', 'grad_norm': '1.067', 'learning_rate': '9.2e-05', 'ppl': '1.034', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '37.23', 'tokens/total': 726464, 'tokens/trainable': 10899, 'epoch': '0.09375'}
+
  5%|▍         | 24/512 [02:41<54:22,  6.69s/it]
  5%|▍         | 25/512 [02:48<54:19,  6.69s/it]
                                                
{'loss': '0.03084', 'grad_norm': '1.049', 'learning_rate': '9.6e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 756704, 'tokens/trainable': 11360, 'epoch': '0.09766'}
+
  5%|▍         | 25/512 [02:48<54:19,  6.69s/it]
  5%|▌         | 26/512 [02:54<54:12,  6.69s/it]
                                                
{'loss': '0.03408', 'grad_norm': '0.8108', 'learning_rate': '0.0001', 'ppl': '1.035', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '29.3', 'tokens/total': 786912, 'tokens/trainable': 11775, 'epoch': '0.1016'}
+
  5%|▌         | 26/512 [02:54<54:12,  6.69s/it]
  5%|▌         | 27/512 [03:01<52:53,  6.54s/it]
                                                
{'loss': '0.014', 'grad_norm': '0.507', 'learning_rate': '0.0001', 'ppl': '1.014', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.89', 'tokens/total': 815424, 'tokens/trainable': 12242, 'epoch': '0.1055'}
+
  5%|▌         | 27/512 [03:01<52:53,  6.54s/it]
  5%|▌         | 28/512 [03:07<53:04,  6.58s/it]
                                                
{'loss': '0.02229', 'grad_norm': '0.6185', 'learning_rate': '0.0001', 'ppl': '1.023', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.48', 'tokens/total': 845584, 'tokens/trainable': 12696, 'epoch': '0.1094'}
+
  5%|▌         | 28/512 [03:07<53:04,  6.58s/it]
  6%|▌         | 29/512 [03:14<53:15,  6.62s/it]
                                                
{'loss': '0.02593', 'grad_norm': '0.4989', 'learning_rate': '9.999e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 875808, 'tokens/trainable': 13153, 'epoch': '0.1133'}
+
  6%|▌         | 29/512 [03:14<53:15,  6.62s/it]
  6%|▌         | 30/512 [03:20<52:19,  6.51s/it]
                                                
{'loss': '0.01448', 'grad_norm': '0.3578', 'learning_rate': '9.999e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.7', 'tokens/total': 904096, 'tokens/trainable': 13624, 'epoch': '0.1172'}
+
  6%|▌         | 30/512 [03:20<52:19,  6.51s/it]
  6%|▌         | 31/512 [03:27<52:42,  6.57s/it]
                                                
{'loss': '0.01593', 'grad_norm': '0.4404', 'learning_rate': '9.998e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 934400, 'tokens/trainable': 14064, 'epoch': '0.1211'}
+
  6%|▌         | 31/512 [03:27<52:42,  6.57s/it]
  6%|▋         | 32/512 [03:34<52:58,  6.62s/it]
                                                
{'loss': '0.01941', 'grad_norm': '0.5914', 'learning_rate': '9.997e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.92', 'tokens/total': 964832, 'tokens/trainable': 14565, 'epoch': '0.125'}
+
  6%|▋         | 32/512 [03:34<52:58,  6.62s/it][2026-08-18 14:08:37,375] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
+
  6%|▋         | 33/512 [03:43<58:41,  7.35s/it]
                                                
{'loss': '0.007002', 'grad_norm': '0.2796', 'learning_rate': '9.995e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.77', 'tokens/total': 995040, 'tokens/trainable': 15005, 'epoch': '0.1289'}
+
  6%|▋         | 33/512 [03:43<58:41,  7.35s/it]
  7%|▋         | 34/512 [03:49<56:49,  7.13s/it]
                                                
{'loss': '0.02145', 'grad_norm': '1.176', 'learning_rate': '9.994e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 1024896, 'tokens/trainable': 15459, 'epoch': '0.1328'}
+
  7%|▋         | 34/512 [03:49<56:49,  7.13s/it]
  7%|▋         | 35/512 [03:56<55:36,  7.00s/it]
                                                
{'loss': '0.01894', 'grad_norm': '0.386', 'learning_rate': '9.992e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '38.21', 'tokens/total': 1054992, 'tokens/trainable': 15960, 'epoch': '0.1367'}
+
  7%|▋         | 35/512 [03:56<55:36,  7.00s/it]
  7%|▋         | 36/512 [04:03<54:40,  6.89s/it]
                                                
{'loss': '0.01381', 'grad_norm': '0.4804', 'learning_rate': '9.991e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.59', 'tokens/total': 1085280, 'tokens/trainable': 16428, 'epoch': '0.1406'}
+
  7%|▋         | 36/512 [04:03<54:40,  6.89s/it]
  7%|▋         | 37/512 [04:09<53:54,  6.81s/it]
                                                
{'loss': '0.008235', 'grad_norm': '0.3357', 'learning_rate': '9.989e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '31.89', 'tokens/total': 1115504, 'tokens/trainable': 16884, 'epoch': '0.1445'}
+
  7%|▋         | 37/512 [04:09<53:54,  6.81s/it]
  7%|▋         | 38/512 [04:16<53:34,  6.78s/it]
                                                
{'loss': '0.01956', 'grad_norm': '0.9475', 'learning_rate': '9.987e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '32.02', 'tokens/total': 1145712, 'tokens/trainable': 17341, 'epoch': '0.1484'}
+
  7%|▋         | 38/512 [04:16<53:34,  6.78s/it]
  8%|▊         | 39/512 [04:23<53:13,  6.75s/it]
                                                
{'loss': '0.01788', 'grad_norm': '1.102', 'learning_rate': '9.984e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 1176128, 'tokens/trainable': 17786, 'epoch': '0.1523'}
+
  8%|▊         | 39/512 [04:23<53:13,  6.75s/it]
  8%|▊         | 40/512 [04:30<53:05,  6.75s/it]
                                                
{'loss': '0.01934', 'grad_norm': '0.7563', 'learning_rate': '9.982e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 1206576, 'tokens/trainable': 18262, 'epoch': '0.1562'}
+
  8%|▊         | 40/512 [04:30<53:05,  6.75s/it]
  8%|▊         | 41/512 [04:36<52:54,  6.74s/it]
                                                
{'loss': '0.01162', 'grad_norm': '0.4158', 'learning_rate': '9.979e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 1236848, 'tokens/trainable': 18716, 'epoch': '0.1602'}
+
  8%|▊         | 41/512 [04:36<52:54,  6.74s/it]
  8%|▊         | 42/512 [04:43<52:41,  6.73s/it]
                                                
{'loss': '0.04662', 'grad_norm': '1.181', 'learning_rate': '9.976e-05', 'ppl': '1.048', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 1267088, 'tokens/trainable': 19157, 'epoch': '0.1641'}
+
  8%|▊         | 42/512 [04:43<52:41,  6.73s/it]
  8%|▊         | 43/512 [04:50<52:36,  6.73s/it]
                                                
{'loss': '0.02922', 'grad_norm': '0.7583', 'learning_rate': '9.973e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 1297360, 'tokens/trainable': 19647, 'epoch': '0.168'}
+
  8%|▊         | 43/512 [04:50<52:36,  6.73s/it]
  9%|▊         | 44/512 [04:56<52:25,  6.72s/it]
                                                
{'loss': '0.007799', 'grad_norm': '0.1886', 'learning_rate': '9.97e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 1327744, 'tokens/trainable': 20119, 'epoch': '0.1719'}
+
  9%|▊         | 44/512 [04:56<52:25,  6.72s/it]
  9%|▉         | 45/512 [05:03<52:14,  6.71s/it]
                                                
{'loss': '0.01204', 'grad_norm': '0.351', 'learning_rate': '9.966e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 1358096, 'tokens/trainable': 20560, 'epoch': '0.1758'}
+
  9%|▉         | 45/512 [05:03<52:14,  6.71s/it]
  9%|▉         | 46/512 [05:10<51:59,  6.69s/it]
                                                
{'loss': '0.01786', 'grad_norm': '0.5071', 'learning_rate': '9.963e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 1388240, 'tokens/trainable': 20992, 'epoch': '0.1797'}
+
  9%|▉         | 46/512 [05:10<51:59,  6.69s/it]
  9%|▉         | 47/512 [05:16<50:37,  6.53s/it]
                                                
{'loss': '0.01944', 'grad_norm': '0.557', 'learning_rate': '9.959e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.26', 'memory/max_allocated (GiB)': '33.26', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.55', 'tokens/total': 1416240, 'tokens/trainable': 21441, 'epoch': '0.1836'}
+
  9%|▉         | 47/512 [05:16<50:37,  6.53s/it]
  9%|▉         | 48/512 [05:23<50:51,  6.58s/it]
                                                
{'loss': '0.01302', 'grad_norm': '0.4206', 'learning_rate': '9.955e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.68', 'tokens/total': 1446880, 'tokens/trainable': 21901, 'epoch': '0.1875'}
+
  9%|▉         | 48/512 [05:23<50:51,  6.58s/it]
 10%|▉         | 49/512 [05:29<50:57,  6.60s/it]
                                                
{'loss': '0.007749', 'grad_norm': '0.304', 'learning_rate': '9.951e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 1477168, 'tokens/trainable': 22341, 'epoch': '0.1914'}
+
 10%|▉         | 49/512 [05:29<50:57,  6.60s/it]
 10%|▉         | 50/512 [05:36<51:07,  6.64s/it]
                                                
{'loss': '0.01445', 'grad_norm': '2.045', 'learning_rate': '9.946e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 1507712, 'tokens/trainable': 22833, 'epoch': '0.1953'}
+
 10%|▉         | 50/512 [05:36<51:07,  6.64s/it]
 10%|▉         | 51/512 [05:43<51:06,  6.65s/it]
                                                
{'loss': '0.01272', 'grad_norm': '0.9358', 'learning_rate': '9.942e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.72', 'tokens/total': 1537936, 'tokens/trainable': 23277, 'epoch': '0.1992'}
+
 10%|▉         | 51/512 [05:43<51:06,  6.65s/it]
 10%|█         | 52/512 [05:49<51:07,  6.67s/it]
                                                
{'loss': '0.005878', 'grad_norm': '0.2882', 'learning_rate': '9.937e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.63', 'tokens/total': 1568352, 'tokens/trainable': 23759, 'epoch': '0.2031'}
+
 10%|█         | 52/512 [05:49<51:07,  6.67s/it]
 10%|█         | 53/512 [05:56<51:10,  6.69s/it]
                                                
{'loss': '0.01452', 'grad_norm': '0.7578', 'learning_rate': '9.932e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 1598992, 'tokens/trainable': 24193, 'epoch': '0.207'}
+
 10%|█         | 53/512 [05:56<51:10,  6.69s/it]
 11%|█         | 54/512 [06:03<50:57,  6.68s/it]
                                                
{'loss': '0.006856', 'grad_norm': '0.5201', 'learning_rate': '9.927e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.59', 'tokens/total': 1629488, 'tokens/trainable': 24619, 'epoch': '0.2109'}
+
 11%|█         | 54/512 [06:03<50:57,  6.68s/it]
 11%|█         | 55/512 [06:09<50:53,  6.68s/it]
                                                
{'loss': '0.03797', 'grad_norm': '0.84', 'learning_rate': '9.921e-05', 'ppl': '1.039', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.85', 'tokens/total': 1659696, 'tokens/trainable': 25075, 'epoch': '0.2148'}
+
 11%|█         | 55/512 [06:09<50:53,  6.68s/it]
 11%|█         | 56/512 [06:16<50:43,  6.68s/it]
                                                
{'loss': '0.01281', 'grad_norm': '0.8204', 'learning_rate': '9.916e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.55', 'tokens/total': 1689872, 'tokens/trainable': 25519, 'epoch': '0.2188'}
+
 11%|█         | 56/512 [06:16<50:43,  6.68s/it]
 11%|█         | 57/512 [06:23<50:40,  6.68s/it]
                                                
{'loss': '0.00539', 'grad_norm': '0.2087', 'learning_rate': '9.91e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 1720144, 'tokens/trainable': 25966, 'epoch': '0.2227'}
+
 11%|█         | 57/512 [06:23<50:40,  6.68s/it]
 11%|█▏        | 58/512 [06:29<50:34,  6.68s/it]
                                                
{'loss': '0.009391', 'grad_norm': '0.4428', 'learning_rate': '9.904e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 1750448, 'tokens/trainable': 26423, 'epoch': '0.2266'}
+
 11%|█▏        | 58/512 [06:29<50:34,  6.68s/it]
 12%|█▏        | 59/512 [06:36<50:37,  6.70s/it]
                                                
{'loss': '0.01953', 'grad_norm': '0.7458', 'learning_rate': '9.898e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.51', 'tokens/total': 1781088, 'tokens/trainable': 26934, 'epoch': '0.2305'}
+
 12%|█▏        | 59/512 [06:36<50:37,  6.70s/it]
 12%|█▏        | 60/512 [06:43<50:30,  6.71s/it]
                                                
{'loss': '0.001454', 'grad_norm': '0.134', 'learning_rate': '9.892e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 1811344, 'tokens/trainable': 27393, 'epoch': '0.2344'}
+
 12%|█▏        | 60/512 [06:43<50:30,  6.71s/it]
 12%|█▏        | 61/512 [06:50<50:20,  6.70s/it]
                                                
{'loss': '0.01992', 'grad_norm': '1.039', 'learning_rate': '9.886e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 1841600, 'tokens/trainable': 27852, 'epoch': '0.2383'}
+
 12%|█▏        | 61/512 [06:50<50:20,  6.70s/it]
 12%|█▏        | 62/512 [06:56<50:11,  6.69s/it]
                                                
{'loss': '0.01588', 'grad_norm': '1.26', 'learning_rate': '9.879e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 1872048, 'tokens/trainable': 28292, 'epoch': '0.2422'}
+
 12%|█▏        | 62/512 [06:56<50:11,  6.69s/it]
 12%|█▏        | 63/512 [07:03<50:11,  6.71s/it]
                                                
{'loss': '0.002796', 'grad_norm': '0.1403', 'learning_rate': '9.872e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '37.36', 'tokens/total': 1902592, 'tokens/trainable': 28798, 'epoch': '0.2461'}
+
 12%|█▏        | 63/512 [07:03<50:11,  6.71s/it]
 12%|█▎        | 64/512 [07:10<50:05,  6.71s/it]
                                                
{'loss': '0.01712', 'grad_norm': '0.6248', 'learning_rate': '9.865e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.79', 'tokens/total': 1933088, 'tokens/trainable': 29283, 'epoch': '0.25'}
+
 12%|█▎        | 64/512 [07:10<50:05,  6.71s/it][2026-08-18 14:12:13,340] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
+
 13%|█▎        | 65/512 [07:18<53:59,  7.25s/it]
                                                
{'loss': '0.008029', 'grad_norm': '0.3513', 'learning_rate': '9.858e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.99', 'tokens/total': 1963296, 'tokens/trainable': 29745, 'epoch': '0.2539'}
+
 13%|█▎        | 65/512 [07:18<53:59,  7.25s/it]
 13%|█▎        | 66/512 [07:25<52:47,  7.10s/it]
                                                
{'loss': '0.01612', 'grad_norm': '0.2794', 'learning_rate': '9.851e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '34.87', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 1993760, 'tokens/trainable': 30186, 'epoch': '0.2578'}
+
 13%|█▎        | 66/512 [07:25<52:47,  7.10s/it]
 13%|█▎        | 67/512 [07:32<51:41,  6.97s/it]
                                                
{'loss': '0.009092', 'grad_norm': '0.3174', 'learning_rate': '9.844e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '33.36', 'tokens/total': 2024144, 'tokens/trainable': 30622, 'epoch': '0.2617'}
+
 13%|█▎        | 67/512 [07:32<51:41,  6.97s/it]
 13%|█▎        | 68/512 [07:38<50:58,  6.89s/it]
                                                
{'loss': '0.009083', 'grad_norm': '0.3539', 'learning_rate': '9.836e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.52', 'tokens/total': 2054608, 'tokens/trainable': 31097, 'epoch': '0.2656'}
+
 13%|█▎        | 68/512 [07:38<50:58,  6.89s/it]
 13%|█▎        | 69/512 [07:45<50:26,  6.83s/it]
                                                
{'loss': '0.03087', 'grad_norm': '0.6498', 'learning_rate': '9.828e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36.6', 'tokens/total': 2084912, 'tokens/trainable': 31595, 'epoch': '0.2695'}
+
 13%|█▎        | 69/512 [07:45<50:26,  6.83s/it]
 14%|█▎        | 70/512 [07:52<50:01,  6.79s/it]
                                                
{'loss': '0.003916', 'grad_norm': '0.1966', 'learning_rate': '9.82e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '30.75', 'tokens/total': 2115216, 'tokens/trainable': 32036, 'epoch': '0.2734'}
+
 14%|█▎        | 70/512 [07:52<50:01,  6.79s/it]
 14%|█▍        | 71/512 [07:58<49:37,  6.75s/it]
                                                
{'loss': '0.007547', 'grad_norm': '0.43', 'learning_rate': '9.812e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.09', 'tokens/total': 2145424, 'tokens/trainable': 32428, 'epoch': '0.2773'}
+
 14%|█▍        | 71/512 [07:58<49:37,  6.75s/it]
 14%|█▍        | 72/512 [08:05<49:23,  6.74s/it]
                                                
{'loss': '0.0004817', 'grad_norm': '0.02775', 'learning_rate': '9.803e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.11', 'tokens/total': 2175648, 'tokens/trainable': 32880, 'epoch': '0.2812'}
+
 14%|█▍        | 72/512 [08:05<49:23,  6.74s/it]
 14%|█▍        | 73/512 [08:12<49:10,  6.72s/it]
                                                
{'loss': '0.001214', 'grad_norm': '0.112', 'learning_rate': '9.795e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.51', 'tokens/total': 2205856, 'tokens/trainable': 33323, 'epoch': '0.2852'}
+
 14%|█▍        | 73/512 [08:12<49:10,  6.72s/it]
 14%|█▍        | 74/512 [08:18<48:57,  6.71s/it]
                                                
{'loss': '0.0004691', 'grad_norm': '0.02696', 'learning_rate': '9.786e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.86', 'tokens/total': 2236352, 'tokens/trainable': 33792, 'epoch': '0.2891'}
+
 14%|█▍        | 74/512 [08:18<48:57,  6.71s/it]
 15%|█▍        | 75/512 [08:25<48:45,  6.69s/it]
                                                
{'loss': '0.02965', 'grad_norm': '0.5172', 'learning_rate': '9.777e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 2266480, 'tokens/trainable': 34223, 'epoch': '0.293'}
+
 15%|█▍        | 75/512 [08:25<48:45,  6.69s/it]
 15%|█▍        | 76/512 [08:32<48:33,  6.68s/it]
                                                
{'loss': '0.00698', 'grad_norm': '0.8409', 'learning_rate': '9.768e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.61', 'tokens/total': 2296496, 'tokens/trainable': 34649, 'epoch': '0.2969'}
+
 15%|█▍        | 76/512 [08:32<48:33,  6.68s/it]
 15%|█▌        | 77/512 [08:38<48:22,  6.67s/it]
                                                
{'loss': '0.00887', 'grad_norm': '3.494', 'learning_rate': '9.759e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.64', 'tokens/total': 2327040, 'tokens/trainable': 35094, 'epoch': '0.3008'}
+
 15%|█▌        | 77/512 [08:38<48:22,  6.67s/it]
 15%|█▌        | 78/512 [08:45<48:15,  6.67s/it]
                                                
{'loss': '0.03207', 'grad_norm': '0.9529', 'learning_rate': '9.749e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.82', 'tokens/total': 2357328, 'tokens/trainable': 35534, 'epoch': '0.3047'}
+
 15%|█▌        | 78/512 [08:45<48:15,  6.67s/it]
 15%|█▌        | 79/512 [08:52<48:12,  6.68s/it]
                                                
{'loss': '0.01475', 'grad_norm': '0.5782', 'learning_rate': '9.74e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.82', 'tokens/total': 2387664, 'tokens/trainable': 36029, 'epoch': '0.3086'}
+
 15%|█▌        | 79/512 [08:52<48:12,  6.68s/it]
 16%|█▌        | 80/512 [08:58<48:01,  6.67s/it]
                                                
{'loss': '0.002579', 'grad_norm': '0.1509', 'learning_rate': '9.73e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 2418000, 'tokens/trainable': 36469, 'epoch': '0.3125'}
+
 16%|█▌        | 80/512 [08:58<48:01,  6.67s/it]
 16%|█▌        | 81/512 [09:05<48:02,  6.69s/it]
                                                
{'loss': '0.004292', 'grad_norm': '0.19', 'learning_rate': '9.72e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.75', 'tokens/total': 2448720, 'tokens/trainable': 36906, 'epoch': '0.3164'}
+
 16%|█▌        | 81/512 [09:05<48:02,  6.69s/it]
 16%|█▌        | 82/512 [09:12<47:54,  6.69s/it]
                                                
{'loss': '0.01853', 'grad_norm': '0.4506', 'learning_rate': '9.71e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '38.95', 'tokens/total': 2479248, 'tokens/trainable': 37390, 'epoch': '0.3203'}
+
 16%|█▌        | 82/512 [09:12<47:54,  6.69s/it]
 16%|█▌        | 83/512 [09:19<47:48,  6.69s/it]
                                                
{'loss': '0.01482', 'grad_norm': '0.4981', 'learning_rate': '9.699e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.8', 'tokens/total': 2509408, 'tokens/trainable': 37838, 'epoch': '0.3242'}
+
 16%|█▌        | 83/512 [09:19<47:48,  6.69s/it]
 16%|█▋        | 84/512 [09:25<47:41,  6.69s/it]
                                                
{'loss': '0.003186', 'grad_norm': '0.1638', 'learning_rate': '9.689e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 2539776, 'tokens/trainable': 38310, 'epoch': '0.3281'}
+
 16%|█▋        | 84/512 [09:25<47:41,  6.69s/it]
 17%|█▋        | 85/512 [09:32<47:33,  6.68s/it]
                                                
{'loss': '0.0208', 'grad_norm': '0.4073', 'learning_rate': '9.678e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.03', 'tokens/total': 2569840, 'tokens/trainable': 38789, 'epoch': '0.332'}
+
 17%|█▋        | 85/512 [09:32<47:33,  6.68s/it]
 17%|█▋        | 86/512 [09:39<47:30,  6.69s/it]
                                                
{'loss': '0.01293', 'grad_norm': '0.3687', 'learning_rate': '9.667e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '36.36', 'tokens/total': 2600192, 'tokens/trainable': 39277, 'epoch': '0.3359'}
+
 17%|█▋        | 86/512 [09:39<47:30,  6.69s/it]
 17%|█▋        | 87/512 [09:45<47:24,  6.69s/it]
                                                
{'loss': '0.01383', 'grad_norm': '0.3674', 'learning_rate': '9.656e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.68', 'tokens/total': 2630640, 'tokens/trainable': 39737, 'epoch': '0.3398'}
+
 17%|█▋        | 87/512 [09:45<47:24,  6.69s/it]
 17%|█▋        | 88/512 [09:52<47:13,  6.68s/it]
                                                
{'loss': '0.008718', 'grad_norm': '0.2704', 'learning_rate': '9.645e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '29.6', 'tokens/total': 2660832, 'tokens/trainable': 40179, 'epoch': '0.3438'}
+
 17%|█▋        | 88/512 [09:52<47:13,  6.68s/it]
 17%|█▋        | 89/512 [09:59<47:09,  6.69s/it]
                                                
{'loss': '0.01938', 'grad_norm': '3.375', 'learning_rate': '9.633e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.66', 'tokens/total': 2691328, 'tokens/trainable': 40649, 'epoch': '0.3477'}
+
 17%|█▋        | 89/512 [09:59<47:09,  6.69s/it]
 18%|█▊        | 90/512 [10:05<46:01,  6.54s/it]
                                                
{'loss': '0.02013', 'grad_norm': '0.469', 'learning_rate': '9.622e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 2719696, 'tokens/trainable': 41080, 'epoch': '0.3516'}
+
 18%|█▊        | 90/512 [10:05<46:01,  6.54s/it]
 18%|█▊        | 91/512 [10:12<46:20,  6.60s/it]
                                                
{'loss': '0.002885', 'grad_norm': '0.08706', 'learning_rate': '9.61e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '38.11', 'tokens/total': 2750272, 'tokens/trainable': 41599, 'epoch': '0.3555'}
+
 18%|█▊        | 91/512 [10:12<46:20,  6.60s/it]
 18%|█▊        | 92/512 [10:18<46:23,  6.63s/it]
                                                
{'loss': '0.008225', 'grad_norm': '8.198', 'learning_rate': '9.598e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 2780656, 'tokens/trainable': 42076, 'epoch': '0.3594'}
+
 18%|█▊        | 92/512 [10:18<46:23,  6.63s/it]
 18%|█▊        | 93/512 [10:25<46:23,  6.64s/it]
                                                
{'loss': '0.004234', 'grad_norm': '0.1494', 'learning_rate': '9.586e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.71', 'tokens/total': 2811088, 'tokens/trainable': 42522, 'epoch': '0.3633'}
+
 18%|█▊        | 93/512 [10:25<46:23,  6.64s/it]
 18%|█▊        | 94/512 [10:32<46:22,  6.66s/it]
                                                
{'loss': '0.00949', 'grad_norm': '0.154', 'learning_rate': '9.574e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.4', 'tokens/total': 2841520, 'tokens/trainable': 42974, 'epoch': '0.3672'}
+
 18%|█▊        | 94/512 [10:32<46:22,  6.66s/it]
 19%|█▊        | 95/512 [10:38<46:07,  6.64s/it]
                                                
{'loss': '0.02414', 'grad_norm': '0.5275', 'learning_rate': '9.562e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '30.67', 'tokens/total': 2871600, 'tokens/trainable': 43380, 'epoch': '0.3711'}
+
 19%|█▊        | 95/512 [10:38<46:07,  6.64s/it]
 19%|█▉        | 96/512 [10:45<46:13,  6.67s/it]
                                                
{'loss': '0.009927', 'grad_norm': '0.5183', 'learning_rate': '9.549e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 2902144, 'tokens/trainable': 43865, 'epoch': '0.375'}
+
 19%|█▉        | 96/512 [10:45<46:13,  6.67s/it][2026-08-18 14:15:48,624] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
+
 19%|█▉        | 97/512 [10:54<50:11,  7.26s/it]
                                                
{'loss': '0.01449', 'grad_norm': '0.5579', 'learning_rate': '9.536e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 2932272, 'tokens/trainable': 44328, 'epoch': '0.3789'}
+
 19%|█▉        | 97/512 [10:54<50:11,  7.26s/it]
 19%|█▉        | 98/512 [11:00<48:41,  7.06s/it]
                                                
{'loss': '0.002155', 'grad_norm': '0.1052', 'learning_rate': '9.523e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '30.95', 'tokens/total': 2962032, 'tokens/trainable': 44778, 'epoch': '0.3828'}
+
 19%|█▉        | 98/512 [11:00<48:41,  7.06s/it]
 19%|█▉        | 99/512 [11:07<47:44,  6.94s/it]
                                                
{'loss': '0.01703', 'grad_norm': '0.7545', 'learning_rate': '9.51e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.12', 'tokens/total': 2992256, 'tokens/trainable': 45225, 'epoch': '0.3867'}
+
 19%|█▉        | 99/512 [11:07<47:44,  6.94s/it]
 20%|█▉        | 100/512 [11:14<47:07,  6.86s/it]
                                                 
{'loss': '0.006046', 'grad_norm': '0.3857', 'learning_rate': '9.497e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.75', 'tokens/total': 3022608, 'tokens/trainable': 45678, 'epoch': '0.3906'}
+
 20%|█▉        | 100/512 [11:14<47:07,  6.86s/it]
 20%|█▉        | 101/512 [11:20<46:35,  6.80s/it]
                                                 
{'loss': '0.005754', 'grad_norm': '0.2159', 'learning_rate': '9.484e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '32.72', 'tokens/total': 3052992, 'tokens/trainable': 46125, 'epoch': '0.3945'}
+
 20%|█▉        | 101/512 [11:20<46:35,  6.80s/it]
 20%|█▉        | 102/512 [11:27<46:25,  6.79s/it]
                                                 
{'loss': '0.006994', 'grad_norm': '1.125', 'learning_rate': '9.47e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '39.6', 'tokens/total': 3083392, 'tokens/trainable': 46612, 'epoch': '0.3984'}
+
 20%|█▉        | 102/512 [11:27<46:25,  6.79s/it]
 20%|██        | 103/512 [11:34<46:14,  6.78s/it]
                                                 
{'loss': '0.01372', 'grad_norm': '0.2677', 'learning_rate': '9.456e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '32.41', 'tokens/total': 3113744, 'tokens/trainable': 47059, 'epoch': '0.4023'}
+
 20%|██        | 103/512 [11:34<46:14,  6.78s/it]
 20%|██        | 104/512 [11:41<46:17,  6.81s/it]
                                                 
{'loss': '0.01295', 'grad_norm': '0.3259', 'learning_rate': '9.442e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 3144272, 'tokens/trainable': 47536, 'epoch': '0.4062'}
+
 20%|██        | 104/512 [11:41<46:17,  6.81s/it]
 21%|██        | 105/512 [11:47<46:03,  6.79s/it]
                                                 
{'loss': '0.02244', 'grad_norm': '0.5319', 'learning_rate': '9.428e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 3174512, 'tokens/trainable': 48037, 'epoch': '0.4102'}
+
 21%|██        | 105/512 [11:47<46:03,  6.79s/it]
 21%|██        | 106/512 [11:54<45:47,  6.77s/it]
                                                 
{'loss': '0.0158', 'grad_norm': '0.3957', 'learning_rate': '9.414e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 3204896, 'tokens/trainable': 48491, 'epoch': '0.4141'}
+
 21%|██        | 106/512 [11:54<45:47,  6.77s/it]
 21%|██        | 107/512 [12:01<45:42,  6.77s/it]
                                                 
{'loss': '0.01632', 'grad_norm': '0.4242', 'learning_rate': '9.4e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 3235360, 'tokens/trainable': 48940, 'epoch': '0.418'}
+
 21%|██        | 107/512 [12:01<45:42,  6.77s/it]
 21%|██        | 108/512 [12:08<45:30,  6.76s/it]
                                                 
{'loss': '0.01079', 'grad_norm': '0.3206', 'learning_rate': '9.385e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 3265552, 'tokens/trainable': 49389, 'epoch': '0.4219'}
+
 21%|██        | 108/512 [12:08<45:30,  6.76s/it]
 21%|██▏       | 109/512 [12:14<44:10,  6.58s/it]
                                                 
{'loss': '0.008147', 'grad_norm': '0.1722', 'learning_rate': '9.37e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.91', 'tokens/total': 3293856, 'tokens/trainable': 49838, 'epoch': '0.4258'}
+
 21%|██▏       | 109/512 [12:14<44:10,  6.58s/it]
 21%|██▏       | 110/512 [12:20<44:21,  6.62s/it]
                                                 
{'loss': '0.01213', 'grad_norm': '0.3818', 'learning_rate': '9.355e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 3323984, 'tokens/trainable': 50310, 'epoch': '0.4297'}
+
 21%|██▏       | 110/512 [12:20<44:21,  6.62s/it]
 22%|██▏       | 111/512 [12:27<44:21,  6.64s/it]
                                                 
{'loss': '0.01525', 'grad_norm': '0.4984', 'learning_rate': '9.34e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.55', 'tokens/total': 3354320, 'tokens/trainable': 50755, 'epoch': '0.4336'}
+
 22%|██▏       | 111/512 [12:27<44:21,  6.64s/it]
 22%|██▏       | 112/512 [12:34<44:26,  6.67s/it]
                                                 
{'loss': '0.01127', 'grad_norm': '0.439', 'learning_rate': '9.325e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.5', 'tokens/total': 3384880, 'tokens/trainable': 51213, 'epoch': '0.4375'}
+
 22%|██▏       | 112/512 [12:34<44:26,  6.67s/it]
 22%|██▏       | 113/512 [12:41<44:16,  6.66s/it]
                                                 
{'loss': '0.002892', 'grad_norm': '0.567', 'learning_rate': '9.31e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.77', 'tokens/total': 3415104, 'tokens/trainable': 51660, 'epoch': '0.4414'}
+
 22%|██▏       | 113/512 [12:41<44:16,  6.66s/it]
 22%|██▏       | 114/512 [12:47<44:16,  6.67s/it]
                                                 
{'loss': '0.002557', 'grad_norm': '0.1595', 'learning_rate': '9.294e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.34', 'tokens/total': 3445504, 'tokens/trainable': 52133, 'epoch': '0.4453'}
+
 22%|██▏       | 114/512 [12:47<44:16,  6.67s/it]
 22%|██▏       | 115/512 [12:54<44:17,  6.69s/it]
                                                 
{'loss': '0.02446', 'grad_norm': '0.7951', 'learning_rate': '9.278e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.25', 'tokens/total': 3475744, 'tokens/trainable': 52580, 'epoch': '0.4492'}
+
 22%|██▏       | 115/512 [12:54<44:17,  6.69s/it]
 23%|██▎       | 116/512 [13:01<44:16,  6.71s/it]
                                                 
{'loss': '0.004517', 'grad_norm': '0.3832', 'learning_rate': '9.263e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.74', 'tokens/total': 3506016, 'tokens/trainable': 53033, 'epoch': '0.4531'}
+
 23%|██▎       | 116/512 [13:01<44:16,  6.71s/it]
 23%|██▎       | 117/512 [13:07<44:12,  6.71s/it]
                                                 
{'loss': '0.02197', 'grad_norm': '0.3037', 'learning_rate': '9.247e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.22', 'tokens/total': 3536256, 'tokens/trainable': 53517, 'epoch': '0.457'}
+
 23%|██▎       | 117/512 [13:07<44:12,  6.71s/it]
 23%|██▎       | 118/512 [13:14<43:54,  6.69s/it]
                                                 
{'loss': '0.009211', 'grad_norm': '1.39', 'learning_rate': '9.23e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.59', 'tokens/total': 3566480, 'tokens/trainable': 53967, 'epoch': '0.4609'}
+
 23%|██▎       | 118/512 [13:14<43:54,  6.69s/it]
 23%|██▎       | 119/512 [13:21<43:47,  6.69s/it]
                                                 
{'loss': '0.002145', 'grad_norm': '0.1572', 'learning_rate': '9.214e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.33', 'tokens/total': 3596624, 'tokens/trainable': 54430, 'epoch': '0.4648'}
+
 23%|██▎       | 119/512 [13:21<43:47,  6.69s/it]
 23%|██▎       | 120/512 [13:27<43:40,  6.69s/it]
                                                 
{'loss': '0.007881', 'grad_norm': '0.4', 'learning_rate': '9.198e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 3627168, 'tokens/trainable': 54896, 'epoch': '0.4688'}
+
 23%|██▎       | 120/512 [13:27<43:40,  6.69s/it]
 24%|██▎       | 121/512 [13:34<43:38,  6.70s/it]
                                                 
{'loss': '0.01013', 'grad_norm': '0.3812', 'learning_rate': '9.181e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 3657632, 'tokens/trainable': 55379, 'epoch': '0.4727'}
+
 24%|██▎       | 121/512 [13:34<43:38,  6.70s/it]
 24%|██▍       | 122/512 [13:41<43:23,  6.68s/it]
                                                 
{'loss': '0.000794', 'grad_norm': '0.05129', 'learning_rate': '9.164e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 3687824, 'tokens/trainable': 55803, 'epoch': '0.4766'}
+
 24%|██▍       | 122/512 [13:41<43:23,  6.68s/it]
 24%|██▍       | 123/512 [13:47<43:16,  6.67s/it]
                                                 
{'loss': '0.002327', 'grad_norm': '0.1332', 'learning_rate': '9.147e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 3718320, 'tokens/trainable': 56249, 'epoch': '0.4805'}
+
 24%|██▍       | 123/512 [13:47<43:16,  6.67s/it]
 24%|██▍       | 124/512 [13:54<43:15,  6.69s/it]
                                                 
{'loss': '0.003759', 'grad_norm': '0.2302', 'learning_rate': '9.13e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.43', 'tokens/total': 3748672, 'tokens/trainable': 56713, 'epoch': '0.4844'}
+
 24%|██▍       | 124/512 [13:54<43:15,  6.69s/it]
 24%|██▍       | 125/512 [14:01<43:13,  6.70s/it]
                                                 
{'loss': '0.004737', 'grad_norm': '0.3058', 'learning_rate': '9.113e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '39.1', 'tokens/total': 3778976, 'tokens/trainable': 57196, 'epoch': '0.4883'}
+
 24%|██▍       | 125/512 [14:01<43:13,  6.70s/it]
 25%|██▍       | 126/512 [14:08<43:06,  6.70s/it]
                                                 
{'loss': '0.005812', 'grad_norm': '0.3715', 'learning_rate': '9.096e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 3809296, 'tokens/trainable': 57680, 'epoch': '0.4922'}
+
 25%|██▍       | 126/512 [14:08<43:06,  6.70s/it]
 25%|██▍       | 127/512 [14:14<42:50,  6.68s/it]
                                                 
{'loss': '0.003203', 'grad_norm': '0.1105', 'learning_rate': '9.078e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.88', 'tokens/total': 3839328, 'tokens/trainable': 58110, 'epoch': '0.4961'}
+
 25%|██▍       | 127/512 [14:14<42:50,  6.68s/it]
 25%|██▌       | 128/512 [14:21<42:44,  6.68s/it]
                                                 
{'loss': '0.001725', 'grad_norm': '0.09251', 'learning_rate': '9.06e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.74', 'tokens/total': 3869824, 'tokens/trainable': 58534, 'epoch': '0.5'}
+
 25%|██▌       | 128/512 [14:21<42:44,  6.68s/it][2026-08-18 14:19:24,541] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
+
 25%|██▌       | 129/512 [14:29<46:14,  7.25s/it]
                                                 
{'loss': '0.004688', 'grad_norm': '0.2276', 'learning_rate': '9.043e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 3900080, 'tokens/trainable': 58998, 'epoch': '0.5039'}
+
 25%|██▌       | 129/512 [14:29<46:14,  7.25s/it]
 25%|██▌       | 130/512 [14:36<45:04,  7.08s/it]
                                                 
{'loss': '0.008267', 'grad_norm': '0.2882', 'learning_rate': '9.025e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 3930560, 'tokens/trainable': 59448, 'epoch': '0.5078'}
+
 25%|██▌       | 130/512 [14:36<45:04,  7.08s/it]
 26%|██▌       | 131/512 [14:43<44:07,  6.95s/it]
                                                 
{'loss': '0.002837', 'grad_norm': '0.188', 'learning_rate': '9.007e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.92', 'tokens/total': 3960496, 'tokens/trainable': 59906, 'epoch': '0.5117'}
+
 26%|██▌       | 131/512 [14:43<44:07,  6.95s/it]
 26%|██▌       | 132/512 [14:50<43:30,  6.87s/it]
                                                 
{'loss': '0.01204', 'grad_norm': '0.675', 'learning_rate': '8.988e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 3990928, 'tokens/trainable': 60350, 'epoch': '0.5156'}
+
 26%|██▌       | 132/512 [14:50<43:30,  6.87s/it]
 26%|██▌       | 133/512 [14:56<43:03,  6.82s/it]
                                                 
{'loss': '0.01131', 'grad_norm': '0.5789', 'learning_rate': '8.97e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.16', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 4021264, 'tokens/trainable': 60831, 'epoch': '0.5195'}
+
 26%|██▌       | 133/512 [14:56<43:03,  6.82s/it]
 26%|██▌       | 134/512 [15:03<42:42,  6.78s/it]
                                                 
{'loss': '0.03088', 'grad_norm': '0.4706', 'learning_rate': '8.951e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 4051728, 'tokens/trainable': 61305, 'epoch': '0.5234'}
+
 26%|██▌       | 134/512 [15:03<42:42,  6.78s/it]
 26%|██▋       | 135/512 [15:10<42:29,  6.76s/it]
                                                 
{'loss': '0.005536', 'grad_norm': '0.1663', 'learning_rate': '8.933e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.62', 'tokens/total': 4081920, 'tokens/trainable': 61792, 'epoch': '0.5273'}
+
 26%|██▋       | 135/512 [15:10<42:29,  6.76s/it]
 27%|██▋       | 136/512 [15:16<42:18,  6.75s/it]
                                                 
{'loss': '0.002974', 'grad_norm': '0.107', 'learning_rate': '8.914e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.37', 'tokens/total': 4112304, 'tokens/trainable': 62253, 'epoch': '0.5312'}
+
 27%|██▋       | 136/512 [15:16<42:18,  6.75s/it]
 27%|██▋       | 137/512 [15:23<42:06,  6.74s/it]
                                                 
{'loss': '0.0117', 'grad_norm': '0.292', 'learning_rate': '8.895e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.65', 'tokens/total': 4142512, 'tokens/trainable': 62707, 'epoch': '0.5352'}
+
 27%|██▋       | 137/512 [15:23<42:06,  6.74s/it]
 27%|██▋       | 138/512 [15:30<41:49,  6.71s/it]
                                                 
{'loss': '0.01136', 'grad_norm': '0.3187', 'learning_rate': '8.876e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '29.32', 'tokens/total': 4172992, 'tokens/trainable': 63145, 'epoch': '0.5391'}
+
 27%|██▋       | 138/512 [15:30<41:49,  6.71s/it]
 27%|██▋       | 139/512 [15:36<41:40,  6.70s/it]
                                                 
{'loss': '0.006681', 'grad_norm': '0.1859', 'learning_rate': '8.856e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 4203136, 'tokens/trainable': 63587, 'epoch': '0.543'}
+
 27%|██▋       | 139/512 [15:36<41:40,  6.70s/it]
 27%|██▋       | 140/512 [15:43<41:32,  6.70s/it]
                                                 
{'loss': '0.01478', 'grad_norm': '0.276', 'learning_rate': '8.837e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '36.74', 'tokens/total': 4233456, 'tokens/trainable': 64052, 'epoch': '0.5469'}
+
 27%|██▋       | 140/512 [15:43<41:32,  6.70s/it]
 28%|██▊       | 141/512 [15:50<41:25,  6.70s/it]
                                                 
{'loss': '0.01677', 'grad_norm': '0.4988', 'learning_rate': '8.817e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 4263728, 'tokens/trainable': 64526, 'epoch': '0.5508'}
+
 28%|██▊       | 141/512 [15:50<41:25,  6.70s/it]
 28%|██▊       | 142/512 [15:57<41:32,  6.74s/it]
                                                 
{'loss': '0.006472', 'grad_norm': '0.1818', 'learning_rate': '8.798e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '34.58', 'tokens/total': 4294432, 'tokens/trainable': 64983, 'epoch': '0.5547'}
+
 28%|██▊       | 142/512 [15:57<41:32,  6.74s/it]
 28%|██▊       | 143/512 [16:03<41:31,  6.75s/it]
                                                 
{'loss': '0.006493', 'grad_norm': '0.2096', 'learning_rate': '8.778e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '37.58', 'tokens/total': 4324960, 'tokens/trainable': 65492, 'epoch': '0.5586'}
+
 28%|██▊       | 143/512 [16:03<41:31,  6.75s/it]
 28%|██▊       | 144/512 [16:10<41:12,  6.72s/it]
                                                 
{'loss': '0.003702', 'grad_norm': '0.1336', 'learning_rate': '8.758e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '35.72', 'tokens/total': 4355424, 'tokens/trainable': 65925, 'epoch': '0.5625'}
+
 28%|██▊       | 144/512 [16:10<41:12,  6.72s/it]
 28%|██▊       | 145/512 [16:17<41:02,  6.71s/it]
                                                 
{'loss': '0.00649', 'grad_norm': '0.164', 'learning_rate': '8.738e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '31.39', 'tokens/total': 4385712, 'tokens/trainable': 66383, 'epoch': '0.5664'}
+
 28%|██▊       | 145/512 [16:17<41:02,  6.71s/it]
 29%|██▊       | 146/512 [16:23<40:53,  6.70s/it]
                                                 
{'loss': '0.002659', 'grad_norm': '0.1365', 'learning_rate': '8.718e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '31.39', 'tokens/total': 4416016, 'tokens/trainable': 66840, 'epoch': '0.5703'}
+
 29%|██▊       | 146/512 [16:23<40:53,  6.70s/it]
 29%|██▊       | 147/512 [16:30<40:44,  6.70s/it]
                                                 
{'loss': '0.01334', 'grad_norm': '0.4221', 'learning_rate': '8.697e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.76', 'tokens/total': 4446368, 'tokens/trainable': 67297, 'epoch': '0.5742'}
+
 29%|██▊       | 147/512 [16:30<40:44,  6.70s/it]
 29%|██▉       | 148/512 [16:37<40:39,  6.70s/it]
                                                 
{'loss': '0.01279', 'grad_norm': '0.3735', 'learning_rate': '8.677e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '37.36', 'tokens/total': 4476880, 'tokens/trainable': 67769, 'epoch': '0.5781'}
+
 29%|██▉       | 148/512 [16:37<40:39,  6.70s/it]
 29%|██▉       | 149/512 [16:44<40:33,  6.70s/it]
                                                 
{'loss': '0.01249', 'grad_norm': '0.3093', 'learning_rate': '8.656e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.83', 'tokens/total': 4507232, 'tokens/trainable': 68257, 'epoch': '0.582'}
+
 29%|██▉       | 149/512 [16:44<40:33,  6.70s/it]
 29%|██▉       | 150/512 [16:50<40:20,  6.69s/it]
                                                 
{'loss': '0.003457', 'grad_norm': '0.2454', 'learning_rate': '8.635e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.52', 'tokens/total': 4537376, 'tokens/trainable': 68706, 'epoch': '0.5859'}
+
 29%|██▉       | 150/512 [16:50<40:20,  6.69s/it]
 29%|██▉       | 151/512 [16:57<40:11,  6.68s/it]
                                                 
{'loss': '0.009631', 'grad_norm': '0.3378', 'learning_rate': '8.615e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.3', 'tokens/total': 4567584, 'tokens/trainable': 69131, 'epoch': '0.5898'}
+
 29%|██▉       | 151/512 [16:57<40:11,  6.68s/it]
 30%|██▉       | 152/512 [17:03<40:00,  6.67s/it]
                                                 
{'loss': '0.02038', 'grad_norm': '0.348', 'learning_rate': '8.594e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.93', 'tokens/total': 4597792, 'tokens/trainable': 69586, 'epoch': '0.5938'}
+
 30%|██▉       | 152/512 [17:03<40:00,  6.67s/it]
 30%|██▉       | 153/512 [17:10<39:58,  6.68s/it]
                                                 
{'loss': '0.0008243', 'grad_norm': '0.03586', 'learning_rate': '8.572e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 4627936, 'tokens/trainable': 70061, 'epoch': '0.5977'}
+
 30%|██▉       | 153/512 [17:10<39:58,  6.68s/it]
 30%|███       | 154/512 [17:17<39:49,  6.67s/it]
                                                 
{'loss': '0.00389', 'grad_norm': '0.187', 'learning_rate': '8.551e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '27.95', 'tokens/total': 4658352, 'tokens/trainable': 70467, 'epoch': '0.6016'}
+
 30%|███       | 154/512 [17:17<39:49,  6.67s/it]
 30%|███       | 155/512 [17:24<39:46,  6.69s/it]
                                                 
{'loss': '0.005027', 'grad_norm': '0.2884', 'learning_rate': '8.53e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 4688896, 'tokens/trainable': 70933, 'epoch': '0.6055'}
+
 30%|███       | 155/512 [17:24<39:46,  6.69s/it]
 30%|███       | 156/512 [17:30<39:42,  6.69s/it]
                                                 
{'loss': '0.00246', 'grad_norm': '0.07868', 'learning_rate': '8.508e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 4719168, 'tokens/trainable': 71382, 'epoch': '0.6094'}
+
 30%|███       | 156/512 [17:30<39:42,  6.69s/it]
 31%|███       | 157/512 [17:37<39:24,  6.66s/it]
                                                 
{'loss': '0.001206', 'grad_norm': '0.05601', 'learning_rate': '8.487e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '30.83', 'tokens/total': 4749136, 'tokens/trainable': 71808, 'epoch': '0.6133'}
+
 31%|███       | 157/512 [17:37<39:24,  6.66s/it]
 31%|███       | 158/512 [17:44<39:21,  6.67s/it]
                                                 
{'loss': '0.007983', 'grad_norm': '0.3424', 'learning_rate': '8.465e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 4779472, 'tokens/trainable': 72249, 'epoch': '0.6172'}
+
 31%|███       | 158/512 [17:44<39:21,  6.67s/it]
 31%|███       | 159/512 [17:50<39:18,  6.68s/it]
                                                 
{'loss': '0.01179', 'grad_norm': '0.3605', 'learning_rate': '8.443e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 4809632, 'tokens/trainable': 72726, 'epoch': '0.6211'}
+
 31%|███       | 159/512 [17:50<39:18,  6.68s/it]
 31%|███▏      | 160/512 [17:57<39:11,  6.68s/it]
                                                 
{'loss': '0.01866', 'grad_norm': '0.3767', 'learning_rate': '8.421e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 4840112, 'tokens/trainable': 73148, 'epoch': '0.625'}
+
 31%|███▏      | 160/512 [17:57<39:11,  6.68s/it][2026-08-18 14:23:00,532] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
+
 31%|███▏      | 161/512 [18:06<42:32,  7.27s/it]
                                                 
{'loss': '0.007014', 'grad_norm': '0.2013', 'learning_rate': '8.399e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 4870656, 'tokens/trainable': 73600, 'epoch': '0.6289'}
+
 31%|███▏      | 161/512 [18:06<42:32,  7.27s/it]
 32%|███▏      | 162/512 [18:12<41:14,  7.07s/it]
                                                 
{'loss': '0.008246', 'grad_norm': '0.2462', 'learning_rate': '8.376e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.76', 'tokens/total': 4900608, 'tokens/trainable': 74068, 'epoch': '0.6328'}
+
 32%|███▏      | 162/512 [18:12<41:14,  7.07s/it]
 32%|███▏      | 163/512 [18:19<40:24,  6.95s/it]
                                                 
{'loss': '0.007023', 'grad_norm': '0.2069', 'learning_rate': '8.354e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.75', 'tokens/total': 4930752, 'tokens/trainable': 74527, 'epoch': '0.6367'}
+
 32%|███▏      | 163/512 [18:19<40:24,  6.95s/it]
 32%|███▏      | 164/512 [18:25<39:46,  6.86s/it]
                                                 
{'loss': '0.008201', 'grad_norm': '0.1888', 'learning_rate': '8.332e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '31.87', 'tokens/total': 4961008, 'tokens/trainable': 74992, 'epoch': '0.6406'}
+
 32%|███▏      | 164/512 [18:25<39:46,  6.86s/it]
 32%|███▏      | 165/512 [18:32<39:22,  6.81s/it]
                                                 
{'loss': '0.004637', 'grad_norm': '0.1746', 'learning_rate': '8.309e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 4991200, 'tokens/trainable': 75442, 'epoch': '0.6445'}
+
 32%|███▏      | 165/512 [18:32<39:22,  6.81s/it]
 32%|███▏      | 166/512 [18:39<39:06,  6.78s/it]
                                                 
{'loss': '0.007356', 'grad_norm': '0.2601', 'learning_rate': '8.286e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.27', 'tokens/total': 5021600, 'tokens/trainable': 75887, 'epoch': '0.6484'}
+
 32%|███▏      | 166/512 [18:39<39:06,  6.78s/it]
 33%|███▎      | 167/512 [18:46<38:50,  6.76s/it]
                                                 
{'loss': '0.01041', 'grad_norm': '0.254', 'learning_rate': '8.263e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 5052032, 'tokens/trainable': 76331, 'epoch': '0.6523'}
+
 33%|███▎      | 167/512 [18:46<38:50,  6.76s/it]
 33%|███▎      | 168/512 [18:52<37:44,  6.58s/it]
                                                 
{'loss': '0.002282', 'grad_norm': '0.1233', 'learning_rate': '8.24e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.92', 'tokens/total': 5080256, 'tokens/trainable': 76745, 'epoch': '0.6562'}
+
 33%|███▎      | 168/512 [18:52<37:44,  6.58s/it]
 33%|███▎      | 169/512 [18:58<37:45,  6.61s/it]
                                                 
{'loss': '0.003871', 'grad_norm': '0.11', 'learning_rate': '8.217e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 5110784, 'tokens/trainable': 77201, 'epoch': '0.6602'}
+
 33%|███▎      | 169/512 [18:58<37:45,  6.61s/it]
 33%|███▎      | 170/512 [19:05<37:46,  6.63s/it]
                                                 
{'loss': '0.0008502', 'grad_norm': '0.0368', 'learning_rate': '8.194e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.4', 'tokens/total': 5141200, 'tokens/trainable': 77655, 'epoch': '0.6641'}
+
 33%|███▎      | 170/512 [19:05<37:46,  6.63s/it]
 33%|███▎      | 171/512 [19:12<37:45,  6.64s/it]
                                                 
{'loss': '0.001807', 'grad_norm': '0.1271', 'learning_rate': '8.171e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '39.23', 'tokens/total': 5171760, 'tokens/trainable': 78122, 'epoch': '0.668'}
+
 33%|███▎      | 171/512 [19:12<37:45,  6.64s/it]
 34%|███▎      | 172/512 [19:18<37:45,  6.66s/it]
                                                 
{'loss': '0.001766', 'grad_norm': '0.1453', 'learning_rate': '8.147e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.53', 'tokens/total': 5202272, 'tokens/trainable': 78576, 'epoch': '0.6719'}
+
 34%|███▎      | 172/512 [19:19<37:45,  6.66s/it]
 34%|███▍      | 173/512 [19:25<37:42,  6.67s/it]
                                                 
{'loss': '0.002328', 'grad_norm': '0.2125', 'learning_rate': '8.124e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 5232800, 'tokens/trainable': 79059, 'epoch': '0.6758'}
+
 34%|███▍      | 173/512 [19:25<37:42,  6.67s/it]
 34%|███▍      | 174/512 [19:32<37:27,  6.65s/it]
                                                 
{'loss': '0.00771', 'grad_norm': '0.6374', 'learning_rate': '8.1e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '29.72', 'tokens/total': 5262672, 'tokens/trainable': 79498, 'epoch': '0.6797'}
+
 34%|███▍      | 174/512 [19:32<37:27,  6.65s/it]
 34%|███▍      | 175/512 [19:38<37:19,  6.65s/it]
                                                 
{'loss': '0.0007811', 'grad_norm': '0.06158', 'learning_rate': '8.076e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '30.45', 'tokens/total': 5293072, 'tokens/trainable': 79922, 'epoch': '0.6836'}
+
 34%|███▍      | 175/512 [19:38<37:19,  6.65s/it]
 34%|███▍      | 176/512 [19:45<36:26,  6.51s/it]
                                                 
{'loss': '0.0007725', 'grad_norm': '0.04236', 'learning_rate': '8.053e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 5321520, 'tokens/trainable': 80383, 'epoch': '0.6875'}
+
 34%|███▍      | 176/512 [19:45<36:26,  6.51s/it]
 35%|███▍      | 177/512 [19:51<36:37,  6.56s/it]
                                                 
{'loss': '0.00232', 'grad_norm': '0.08921', 'learning_rate': '8.029e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 5351904, 'tokens/trainable': 80824, 'epoch': '0.6914'}
+
 35%|███▍      | 177/512 [19:51<36:37,  6.56s/it]
 35%|███▍      | 178/512 [19:58<36:39,  6.58s/it]
                                                 
{'loss': '0.008408', 'grad_norm': '0.6793', 'learning_rate': '8.005e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '32.23', 'tokens/total': 5382432, 'tokens/trainable': 81243, 'epoch': '0.6953'}
+
 35%|███▍      | 178/512 [19:58<36:39,  6.58s/it]
 35%|███▍      | 179/512 [20:05<36:42,  6.61s/it]
                                                 
{'loss': '0.0004587', 'grad_norm': '0.05572', 'learning_rate': '7.98e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 5412960, 'tokens/trainable': 81716, 'epoch': '0.6992'}
+
 35%|███▍      | 179/512 [20:05<36:42,  6.61s/it]
 35%|███▌      | 180/512 [20:11<36:46,  6.65s/it]
                                                 
{'loss': '0.00478', 'grad_norm': '0.3616', 'learning_rate': '7.956e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 5443232, 'tokens/trainable': 82181, 'epoch': '0.7031'}
+
 35%|███▌      | 180/512 [20:11<36:46,  6.65s/it]
 35%|███▌      | 181/512 [20:18<35:53,  6.51s/it]
                                                 
{'loss': '0.004499', 'grad_norm': '0.3551', 'learning_rate': '7.932e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.29', 'memory/max_allocated (GiB)': '33.29', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.65', 'tokens/total': 5471440, 'tokens/trainable': 82626, 'epoch': '0.707'}
+
 35%|███▌      | 181/512 [20:18<35:53,  6.51s/it]
 36%|███▌      | 182/512 [20:24<36:04,  6.56s/it]
                                                 
{'loss': '0.0001246', 'grad_norm': '0.007323', 'learning_rate': '7.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 5501744, 'tokens/trainable': 83089, 'epoch': '0.7109'}
+
 36%|███▌      | 182/512 [20:24<36:04,  6.56s/it]
 36%|███▌      | 183/512 [20:31<36:11,  6.60s/it]
                                                 
{'loss': '0.000127', 'grad_norm': '0.005601', 'learning_rate': '7.883e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 5532272, 'tokens/trainable': 83590, 'epoch': '0.7148'}
+
 36%|███▌      | 183/512 [20:31<36:11,  6.60s/it]
 36%|███▌      | 184/512 [20:38<36:17,  6.64s/it]
                                                 
{'loss': '0.0002659', 'grad_norm': '0.03299', 'learning_rate': '7.858e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 5562784, 'tokens/trainable': 84046, 'epoch': '0.7188'}
+
 36%|███▌      | 184/512 [20:38<36:17,  6.64s/it]
 36%|███▌      | 185/512 [20:44<36:10,  6.64s/it]
                                                 
{'loss': '0.0002492', 'grad_norm': '0.0247', 'learning_rate': '7.833e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 5592800, 'tokens/trainable': 84517, 'epoch': '0.7227'}
+
 36%|███▌      | 185/512 [20:44<36:10,  6.64s/it]
 36%|███▋      | 186/512 [20:51<36:12,  6.66s/it]
                                                 
{'loss': '0.01673', 'grad_norm': '0.6944', 'learning_rate': '7.808e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 5623088, 'tokens/trainable': 84974, 'epoch': '0.7266'}
+
 36%|███▋      | 186/512 [20:51<36:12,  6.66s/it]
 37%|███▋      | 187/512 [20:58<36:12,  6.68s/it]
                                                 
{'loss': '0.0001727', 'grad_norm': '0.01172', 'learning_rate': '7.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.07', 'tokens/total': 5653344, 'tokens/trainable': 85460, 'epoch': '0.7305'}
+
 37%|███▋      | 187/512 [20:58<36:12,  6.68s/it]
 37%|███▋      | 188/512 [21:04<36:05,  6.68s/it]
                                                 
{'loss': '0.004549', 'grad_norm': '0.3487', 'learning_rate': '7.758e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 5683600, 'tokens/trainable': 85939, 'epoch': '0.7344'}
+
 37%|███▋      | 188/512 [21:04<36:05,  6.68s/it]
 37%|███▋      | 189/512 [21:11<35:06,  6.52s/it]
                                                 
{'loss': '0.00147', 'grad_norm': '0.0911', 'learning_rate': '7.733e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '38.43', 'tokens/total': 5711952, 'tokens/trainable': 86377, 'epoch': '0.7383'}
+
 37%|███▋      | 189/512 [21:11<35:06,  6.52s/it]
 37%|███▋      | 190/512 [21:17<35:17,  6.58s/it]
                                                 
{'loss': '0.01117', 'grad_norm': '0.6382', 'learning_rate': '7.708e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 5742336, 'tokens/trainable': 86834, 'epoch': '0.7422'}
+
 37%|███▋      | 190/512 [21:17<35:17,  6.58s/it]
 37%|███▋      | 191/512 [21:24<35:17,  6.60s/it]
                                                 
{'loss': '0.002249', 'grad_norm': '0.1397', 'learning_rate': '7.683e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 5772560, 'tokens/trainable': 87265, 'epoch': '0.7461'}
+
 37%|███▋      | 191/512 [21:24<35:17,  6.60s/it]
 38%|███▊      | 192/512 [21:31<35:28,  6.65s/it]
                                                 
{'loss': '0.0009342', 'grad_norm': '0.09119', 'learning_rate': '7.657e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 5803136, 'tokens/trainable': 87747, 'epoch': '0.75'}
+
 38%|███▊      | 192/512 [21:31<35:28,  6.65s/it][2026-08-18 14:26:34,276] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
+
 38%|███▊      | 193/512 [21:39<38:27,  7.23s/it]
                                                 
{'loss': '0.001603', 'grad_norm': '0.1008', 'learning_rate': '7.632e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '36.21', 'tokens/total': 5833440, 'tokens/trainable': 88239, 'epoch': '0.7539'}
+
 38%|███▊      | 193/512 [21:39<38:27,  7.23s/it]
 38%|███▊      | 194/512 [21:46<37:30,  7.08s/it]
                                                 
{'loss': '0.0009742', 'grad_norm': '0.05941', 'learning_rate': '7.606e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 5863552, 'tokens/trainable': 88718, 'epoch': '0.7578'}
+
 38%|███▊      | 194/512 [21:46<37:30,  7.08s/it]
 38%|███▊      | 195/512 [21:53<36:42,  6.95s/it]
                                                 
{'loss': '0.001538', 'grad_norm': '0.1252', 'learning_rate': '7.58e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '31.29', 'tokens/total': 5893680, 'tokens/trainable': 89129, 'epoch': '0.7617'}
+
 38%|███▊      | 195/512 [21:53<36:42,  6.95s/it]
 38%|███▊      | 196/512 [21:59<36:08,  6.86s/it]
                                                 
{'loss': '0.003383', 'grad_norm': '0.3236', 'learning_rate': '7.555e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '32.29', 'tokens/total': 5923744, 'tokens/trainable': 89567, 'epoch': '0.7656'}
+
 38%|███▊      | 196/512 [21:59<36:08,  6.86s/it]
 38%|███▊      | 197/512 [22:06<35:41,  6.80s/it]
                                                 
{'loss': '0.009017', 'grad_norm': '0.3034', 'learning_rate': '7.529e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 5954128, 'tokens/trainable': 89988, 'epoch': '0.7695'}
+
 38%|███▊      | 197/512 [22:06<35:41,  6.80s/it]
 39%|███▊      | 198/512 [22:13<35:24,  6.77s/it]
                                                 
{'loss': '0.01389', 'grad_norm': '0.5301', 'learning_rate': '7.503e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 5984352, 'tokens/trainable': 90434, 'epoch': '0.7734'}
+
 39%|███▊      | 198/512 [22:13<35:24,  6.77s/it]
 39%|███▉      | 199/512 [22:19<35:12,  6.75s/it]
                                                 
{'loss': '0.0002993', 'grad_norm': '0.02488', 'learning_rate': '7.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.6', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 6014704, 'tokens/trainable': 90913, 'epoch': '0.7773'}
+
 39%|███▉      | 199/512 [22:19<35:12,  6.75s/it]
 39%|███▉      | 200/512 [22:26<34:55,  6.72s/it]
                                                 
{'loss': '0.001377', 'grad_norm': '0.07655', 'learning_rate': '7.451e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.66', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 6045056, 'tokens/trainable': 91355, 'epoch': '0.7812'}
+
 39%|███▉      | 200/512 [22:26<34:55,  6.72s/it]
 39%|███▉      | 201/512 [22:33<34:56,  6.74s/it]
                                                 
{'loss': '0.001292', 'grad_norm': '0.08619', 'learning_rate': '7.424e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.66', 'tokens/train_per_sec_per_gpu': '37.38', 'tokens/total': 6075504, 'tokens/trainable': 91856, 'epoch': '0.7852'}
+
 39%|███▉      | 201/512 [22:33<34:56,  6.74s/it]
 39%|███▉      | 202/512 [22:39<34:42,  6.72s/it]
                                                 
{'loss': '0.0007855', 'grad_norm': '0.07494', 'learning_rate': '7.398e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 6105904, 'tokens/trainable': 92311, 'epoch': '0.7891'}
+
 39%|███▉      | 202/512 [22:39<34:42,  6.72s/it]
 40%|███▉      | 203/512 [22:46<34:34,  6.71s/it]
                                                 
{'loss': '0.002631', 'grad_norm': '0.1252', 'learning_rate': '7.372e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.73', 'tokens/total': 6136272, 'tokens/trainable': 92764, 'epoch': '0.793'}
+
 40%|███▉      | 203/512 [22:46<34:34,  6.71s/it]
 40%|███▉      | 204/512 [22:53<34:19,  6.69s/it]
                                                 
{'loss': '0.004027', 'grad_norm': '0.1421', 'learning_rate': '7.345e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 6166336, 'tokens/trainable': 93227, 'epoch': '0.7969'}
+
 40%|███▉      | 204/512 [22:53<34:19,  6.69s/it]
 40%|████      | 205/512 [22:59<34:09,  6.67s/it]
                                                 
{'loss': '0.01281', 'grad_norm': '0.9565', 'learning_rate': '7.319e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 6196720, 'tokens/trainable': 93675, 'epoch': '0.8008'}
+
 40%|████      | 205/512 [22:59<34:09,  6.67s/it]
 40%|████      | 206/512 [23:06<34:09,  6.70s/it]
                                                 
{'loss': '0.003737', 'grad_norm': '0.131', 'learning_rate': '7.292e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.3', 'tokens/total': 6227424, 'tokens/trainable': 94120, 'epoch': '0.8047'}
+
 40%|████      | 206/512 [23:06<34:09,  6.70s/it]
 40%|████      | 207/512 [23:13<33:58,  6.68s/it]
                                                 
{'loss': '0.001748', 'grad_norm': '0.0838', 'learning_rate': '7.266e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 6257616, 'tokens/trainable': 94557, 'epoch': '0.8086'}
+
 40%|████      | 207/512 [23:13<33:58,  6.68s/it]
 41%|████      | 208/512 [23:20<33:54,  6.69s/it]
                                                 
{'loss': '0.001855', 'grad_norm': '0.05349', 'learning_rate': '7.239e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '40.66', 'tokens/total': 6288000, 'tokens/trainable': 95035, 'epoch': '0.8125'}
+
 41%|████      | 208/512 [23:20<33:54,  6.69s/it]
 41%|████      | 209/512 [23:26<33:46,  6.69s/it]
                                                 
{'loss': '0.002475', 'grad_norm': '0.1101', 'learning_rate': '7.212e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '37.59', 'tokens/total': 6318336, 'tokens/trainable': 95517, 'epoch': '0.8164'}
+
 41%|████      | 209/512 [23:26<33:46,  6.69s/it]
 41%|████      | 210/512 [23:33<33:39,  6.69s/it]
                                                 
{'loss': '0.001984', 'grad_norm': '0.08477', 'learning_rate': '7.185e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 6348656, 'tokens/trainable': 96026, 'epoch': '0.8203'}
+
 41%|████      | 210/512 [23:33<33:39,  6.69s/it]
 41%|████      | 211/512 [23:40<33:29,  6.68s/it]
                                                 
{'loss': '0.001189', 'grad_norm': '0.06026', 'learning_rate': '7.158e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.64', 'tokens/total': 6379040, 'tokens/trainable': 96477, 'epoch': '0.8242'}
+
 41%|████      | 211/512 [23:40<33:29,  6.68s/it]
 41%|████▏     | 212/512 [23:46<33:21,  6.67s/it]
                                                 
{'loss': '0.001786', 'grad_norm': '0.07388', 'learning_rate': '7.131e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.17', 'tokens/total': 6409312, 'tokens/trainable': 96951, 'epoch': '0.8281'}
+
 41%|████▏     | 212/512 [23:46<33:21,  6.67s/it]
 42%|████▏     | 213/512 [23:53<33:13,  6.67s/it]
                                                 
{'loss': '0.00198', 'grad_norm': '0.1013', 'learning_rate': '7.104e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '30.81', 'tokens/total': 6439520, 'tokens/trainable': 97350, 'epoch': '0.832'}
+
 42%|████▏     | 213/512 [23:53<33:13,  6.67s/it]
 42%|████▏     | 214/512 [24:00<33:05,  6.66s/it]
                                                 
{'loss': '0.0001322', 'grad_norm': '0.009713', 'learning_rate': '7.077e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '29.91', 'tokens/total': 6469888, 'tokens/trainable': 97792, 'epoch': '0.8359'}
+
 42%|████▏     | 214/512 [24:00<33:05,  6.66s/it]
 42%|████▏     | 215/512 [24:06<33:02,  6.67s/it]
                                                 
{'loss': '0.00157', 'grad_norm': '0.1457', 'learning_rate': '7.05e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 6500128, 'tokens/trainable': 98257, 'epoch': '0.8398'}
+
 42%|████▏     | 215/512 [24:06<33:02,  6.67s/it]
 42%|████▏     | 216/512 [24:13<33:02,  6.70s/it]
                                                 
{'loss': '0.0003705', 'grad_norm': '0.02122', 'learning_rate': '7.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 6530832, 'tokens/trainable': 98731, 'epoch': '0.8438'}
+
 42%|████▏     | 216/512 [24:13<33:02,  6.70s/it]
 42%|████▏     | 217/512 [24:20<32:54,  6.69s/it]
                                                 
{'loss': '0.003762', 'grad_norm': '0.2245', 'learning_rate': '6.995e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.5', 'tokens/total': 6561248, 'tokens/trainable': 99212, 'epoch': '0.8477'}
+
 42%|████▏     | 217/512 [24:20<32:54,  6.69s/it]
 43%|████▎     | 218/512 [24:26<32:41,  6.67s/it]
                                                 
{'loss': '0.001126', 'grad_norm': '0.6341', 'learning_rate': '6.968e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 6591648, 'tokens/trainable': 99654, 'epoch': '0.8516'}
+
 43%|████▎     | 218/512 [24:26<32:41,  6.67s/it]
 43%|████▎     | 219/512 [24:33<32:39,  6.69s/it]
                                                 
{'loss': '0.001478', 'grad_norm': '0.1625', 'learning_rate': '6.94e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 6622368, 'tokens/trainable': 100086, 'epoch': '0.8555'}
+
 43%|████▎     | 219/512 [24:33<32:39,  6.69s/it]
 43%|████▎     | 220/512 [24:40<32:30,  6.68s/it]
                                                 
{'loss': '0.0001898', 'grad_norm': '0.01793', 'learning_rate': '6.913e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.34', 'tokens/total': 6652848, 'tokens/trainable': 100524, 'epoch': '0.8594'}
+
 43%|████▎     | 220/512 [24:40<32:30,  6.68s/it]
 43%|████▎     | 221/512 [24:46<32:30,  6.70s/it]
                                                 
{'loss': '0.0008955', 'grad_norm': '0.1249', 'learning_rate': '6.885e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.46', 'tokens/total': 6683392, 'tokens/trainable': 100996, 'epoch': '0.8633'}
+
 43%|████▎     | 221/512 [24:46<32:30,  6.70s/it]
 43%|████▎     | 222/512 [24:53<32:19,  6.69s/it]
                                                 
{'loss': '0.0001049', 'grad_norm': '0.01509', 'learning_rate': '6.857e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 6713712, 'tokens/trainable': 101449, 'epoch': '0.8672'}
+
 43%|████▎     | 222/512 [24:53<32:19,  6.69s/it]
 44%|████▎     | 223/512 [25:00<32:08,  6.67s/it]
                                                 
{'loss': '0.01887', 'grad_norm': '0.7131', 'learning_rate': '6.83e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.14', 'tokens/total': 6744176, 'tokens/trainable': 101905, 'epoch': '0.8711'}
+
 44%|████▎     | 223/512 [25:00<32:08,  6.67s/it]
 44%|████▍     | 224/512 [25:06<32:06,  6.69s/it]
                                                 
{'loss': '7.858e-05', 'grad_norm': '0.008568', 'learning_rate': '6.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.43', 'tokens/total': 6774416, 'tokens/trainable': 102411, 'epoch': '0.875'}
+
 44%|████▍     | 224/512 [25:06<32:06,  6.69s/it][2026-08-18 14:30:10,034] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
+
 44%|████▍     | 225/512 [25:15<34:39,  7.24s/it]
                                                 
{'loss': '0.01011', 'grad_norm': '0.6807', 'learning_rate': '6.774e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 6804592, 'tokens/trainable': 102882, 'epoch': '0.8789'}
+
 44%|████▍     | 225/512 [25:15<34:39,  7.24s/it]
 44%|████▍     | 226/512 [25:22<33:41,  7.07s/it]
                                                 
{'loss': '5.371e-05', 'grad_norm': '0.004204', 'learning_rate': '6.746e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '34.69', 'tokens/total': 6834976, 'tokens/trainable': 103332, 'epoch': '0.8828'}
+
 44%|████▍     | 226/512 [25:22<33:41,  7.07s/it]
 44%|████▍     | 227/512 [25:28<33:03,  6.96s/it]
                                                 
{'loss': '0.0003807', 'grad_norm': '0.0261', 'learning_rate': '6.718e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 6865408, 'tokens/trainable': 103790, 'epoch': '0.8867'}
+
 44%|████▍     | 227/512 [25:28<33:03,  6.96s/it]
 45%|████▍     | 228/512 [25:35<32:33,  6.88s/it]
                                                 
{'loss': '6.497e-05', 'grad_norm': '0.002807', 'learning_rate': '6.69e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 6895664, 'tokens/trainable': 104246, 'epoch': '0.8906'}
+
 45%|████▍     | 228/512 [25:35<32:33,  6.88s/it]
 45%|████▍     | 229/512 [25:42<32:09,  6.82s/it]
                                                 
{'loss': '0.01367', 'grad_norm': '0.8277', 'learning_rate': '6.662e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.72', 'tokens/total': 6926016, 'tokens/trainable': 104707, 'epoch': '0.8945'}
+
 45%|████▍     | 229/512 [25:42<32:09,  6.82s/it]
 45%|████▍     | 230/512 [25:48<31:53,  6.79s/it]
                                                 
{'loss': '0.0005469', 'grad_norm': '0.08311', 'learning_rate': '6.634e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.84', 'tokens/total': 6956384, 'tokens/trainable': 105167, 'epoch': '0.8984'}
+
 45%|████▍     | 230/512 [25:48<31:53,  6.79s/it]
 45%|████▌     | 231/512 [25:55<31:33,  6.74s/it]
                                                 
{'loss': '0.0001304', 'grad_norm': '0.008128', 'learning_rate': '6.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 6986768, 'tokens/trainable': 105576, 'epoch': '0.9023'}
+
 45%|████▌     | 231/512 [25:55<31:33,  6.74s/it]
 45%|████▌     | 232/512 [26:02<31:20,  6.72s/it]
                                                 
{'loss': '0.0003516', 'grad_norm': '0.03313', 'learning_rate': '6.578e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.76', 'tokens/total': 7017232, 'tokens/trainable': 106043, 'epoch': '0.9062'}
+
 45%|████▌     | 232/512 [26:02<31:20,  6.72s/it]
 46%|████▌     | 233/512 [26:08<31:11,  6.71s/it]
                                                 
{'loss': '0.001037', 'grad_norm': '0.0869', 'learning_rate': '6.55e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 7047568, 'tokens/trainable': 106506, 'epoch': '0.9102'}
+
 46%|████▌     | 233/512 [26:08<31:11,  6.71s/it]
 46%|████▌     | 234/512 [26:15<30:59,  6.69s/it]
                                                 
{'loss': '0.001943', 'grad_norm': '0.129', 'learning_rate': '6.522e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 7077760, 'tokens/trainable': 106951, 'epoch': '0.9141'}
+
 46%|████▌     | 234/512 [26:15<30:59,  6.69s/it]
 46%|████▌     | 235/512 [26:22<30:46,  6.66s/it]
                                                 
{'loss': '0.002509', 'grad_norm': '0.2505', 'learning_rate': '6.493e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.73', 'tokens/total': 7107760, 'tokens/trainable': 107382, 'epoch': '0.918'}
+
 46%|████▌     | 235/512 [26:22<30:46,  6.66s/it]
 46%|████▌     | 236/512 [26:28<30:42,  6.68s/it]
                                                 
{'loss': '0.008919', 'grad_norm': '0.5062', 'learning_rate': '6.465e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.82', 'tokens/total': 7138144, 'tokens/trainable': 107814, 'epoch': '0.9219'}
+
 46%|████▌     | 236/512 [26:28<30:42,  6.68s/it]
 46%|████▋     | 237/512 [26:35<30:37,  6.68s/it]
                                                 
{'loss': '0.004196', 'grad_norm': '0.3209', 'learning_rate': '6.437e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 7168352, 'tokens/trainable': 108274, 'epoch': '0.9258'}
+
 46%|████▋     | 237/512 [26:35<30:37,  6.68s/it]
 46%|████▋     | 238/512 [26:42<30:32,  6.69s/it]
                                                 
{'loss': '0.003744', 'grad_norm': '0.3213', 'learning_rate': '6.408e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.09', 'tokens/total': 7198624, 'tokens/trainable': 108714, 'epoch': '0.9297'}
+
 46%|████▋     | 238/512 [26:42<30:32,  6.69s/it]
 47%|████▋     | 239/512 [26:48<30:27,  6.69s/it]
                                                 
{'loss': '0.0003515', 'grad_norm': '0.02197', 'learning_rate': '6.38e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 7229216, 'tokens/trainable': 109220, 'epoch': '0.9336'}
+
 47%|████▋     | 239/512 [26:48<30:27,  6.69s/it]
 47%|████▋     | 240/512 [26:55<30:20,  6.69s/it]
                                                 
{'loss': '0.001268', 'grad_norm': '0.1164', 'learning_rate': '6.351e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32', 'tokens/total': 7259648, 'tokens/trainable': 109672, 'epoch': '0.9375'}
+
 47%|████▋     | 240/512 [26:55<30:20,  6.69s/it]
 47%|████▋     | 241/512 [27:02<30:14,  6.69s/it]
                                                 
{'loss': '0.002464', 'grad_norm': '0.1797', 'learning_rate': '6.323e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.09', 'tokens/total': 7289824, 'tokens/trainable': 110135, 'epoch': '0.9414'}
+
 47%|████▋     | 241/512 [27:02<30:14,  6.69s/it]
 47%|████▋     | 242/512 [27:09<30:08,  6.70s/it]
                                                 
{'loss': '0.01327', 'grad_norm': '0.5711', 'learning_rate': '6.294e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.6', 'tokens/total': 7320288, 'tokens/trainable': 110563, 'epoch': '0.9453'}
+
 47%|████▋     | 242/512 [27:09<30:08,  6.70s/it]
 47%|████▋     | 243/512 [27:15<30:04,  6.71s/it]
                                                 
{'loss': '0.01847', 'grad_norm': '0.782', 'learning_rate': '6.266e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.84', 'tokens/total': 7350736, 'tokens/trainable': 111049, 'epoch': '0.9492'}
+
 47%|████▋     | 243/512 [27:15<30:04,  6.71s/it]
 48%|████▊     | 244/512 [27:22<30:06,  6.74s/it]
                                                 
{'loss': '0.003869', 'grad_norm': '0.2531', 'learning_rate': '6.237e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 7381280, 'tokens/trainable': 111495, 'epoch': '0.9531'}
+
 48%|████▊     | 244/512 [27:22<30:06,  6.74s/it]
 48%|████▊     | 245/512 [27:29<29:58,  6.74s/it]
                                                 
{'loss': '0.001095', 'grad_norm': '0.09388', 'learning_rate': '6.208e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.68', 'tokens/total': 7411632, 'tokens/trainable': 111968, 'epoch': '0.957'}
+
 48%|████▊     | 245/512 [27:29<29:58,  6.74s/it]
 48%|████▊     | 246/512 [27:35<29:44,  6.71s/it]
                                                 
{'loss': '0.0006574', 'grad_norm': '0.06202', 'learning_rate': '6.18e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 7441904, 'tokens/trainable': 112416, 'epoch': '0.9609'}
+
 48%|████▊     | 246/512 [27:35<29:44,  6.71s/it]
 48%|████▊     | 247/512 [27:42<29:41,  6.72s/it]
                                                 
{'loss': '0.002766', 'grad_norm': '0.2164', 'learning_rate': '6.151e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 7472368, 'tokens/trainable': 112882, 'epoch': '0.9648'}
+
 48%|████▊     | 247/512 [27:42<29:41,  6.72s/it]
 48%|████▊     | 248/512 [27:49<29:34,  6.72s/it]
                                                 
{'loss': '0.001269', 'grad_norm': '0.04909', 'learning_rate': '6.122e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '40.02', 'tokens/total': 7502656, 'tokens/trainable': 113390, 'epoch': '0.9688'}
+
 48%|████▊     | 248/512 [27:49<29:34,  6.72s/it]
 49%|████▊     | 249/512 [27:56<29:26,  6.72s/it]
                                                 
{'loss': '0.0115', 'grad_norm': '0.2183', 'learning_rate': '6.093e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 7532800, 'tokens/trainable': 113904, 'epoch': '0.9727'}
+
 49%|████▊     | 249/512 [27:56<29:26,  6.72s/it]
 49%|████▉     | 250/512 [28:02<29:20,  6.72s/it]
                                                 
{'loss': '0.0009117', 'grad_norm': '0.05114', 'learning_rate': '6.065e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 7563264, 'tokens/trainable': 114351, 'epoch': '0.9766'}
+
 49%|████▉     | 250/512 [28:02<29:20,  6.72s/it]
 49%|████▉     | 251/512 [28:09<29:14,  6.72s/it]
                                                 
{'loss': '0.001624', 'grad_norm': '0.1029', 'learning_rate': '6.036e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 7593840, 'tokens/trainable': 114842, 'epoch': '0.9805'}
+
 49%|████▉     | 251/512 [28:09<29:14,  6.72s/it]
 49%|████▉     | 252/512 [28:16<29:04,  6.71s/it]
                                                 
{'loss': '0.001105', 'grad_norm': '0.0578', 'learning_rate': '6.007e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '28.88', 'tokens/total': 7624416, 'tokens/trainable': 115263, 'epoch': '0.9844'}
+
 49%|████▉     | 252/512 [28:16<29:04,  6.71s/it]
 49%|████▉     | 253/512 [28:22<28:57,  6.71s/it]
                                                 
{'loss': '0.0003236', 'grad_norm': '0.0175', 'learning_rate': '5.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '32.66', 'tokens/total': 7654688, 'tokens/trainable': 115703, 'epoch': '0.9883'}
+
 49%|████▉     | 253/512 [28:22<28:57,  6.71s/it]
 50%|████▉     | 254/512 [28:29<28:46,  6.69s/it]
                                                 
{'loss': '0.003511', 'grad_norm': '0.2686', 'learning_rate': '5.949e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.26', 'tokens/total': 7685088, 'tokens/trainable': 116157, 'epoch': '0.9922'}
+
 50%|████▉     | 254/512 [28:29<28:46,  6.69s/it]
 50%|████▉     | 255/512 [28:36<28:37,  6.68s/it]
                                                 
{'loss': '0.002814', 'grad_norm': '0.2503', 'learning_rate': '5.92e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '31.95', 'tokens/total': 7715552, 'tokens/trainable': 116567, 'epoch': '0.9961'}
+
 50%|████▉     | 255/512 [28:36<28:37,  6.68s/it]
 50%|█████     | 256/512 [28:43<28:43,  6.73s/it]
                                                 
{'loss': '0.0002322', 'grad_norm': '0.01063', 'learning_rate': '5.891e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '31.11', 'tokens/total': 7745872, 'tokens/trainable': 117030, 'epoch': '1'}
+
 50%|█████     | 256/512 [28:43<28:43,  6.73s/it][2026-08-18 14:33:46,258] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
+
 50%|█████     | 257/512 [28:52<31:38,  7.44s/it]
                                                 
{'loss': '8.973e-05', 'grad_norm': '0.004323', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.73', 'tokens/total': 7776208, 'tokens/trainable': 117517, 'epoch': '1.004'}
+
 50%|█████     | 257/512 [28:52<31:38,  7.44s/it]
 50%|█████     | 258/512 [28:58<30:34,  7.22s/it]
                                                 
{'loss': '0.001043', 'grad_norm': '0.04856', 'learning_rate': '5.834e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '38.33', 'tokens/total': 7806480, 'tokens/trainable': 118015, 'epoch': '1.008'}
+
 50%|█████     | 258/512 [28:58<30:34,  7.22s/it]
 51%|█████     | 259/512 [29:05<29:48,  7.07s/it]
                                                 
{'loss': '0.0002961', 'grad_norm': '0.01527', 'learning_rate': '5.805e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '36.2', 'tokens/total': 7836800, 'tokens/trainable': 118520, 'epoch': '1.012'}
+
 51%|█████     | 259/512 [29:05<29:48,  7.07s/it]
 51%|█████     | 260/512 [29:12<29:15,  6.97s/it]
                                                 
{'loss': '0.001453', 'grad_norm': '0.1239', 'learning_rate': '5.776e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 7867040, 'tokens/trainable': 118992, 'epoch': '1.016'}
+
 51%|█████     | 260/512 [29:12<29:15,  6.97s/it]
 51%|█████     | 261/512 [29:19<28:49,  6.89s/it]
                                                 
{'loss': '0.0002353', 'grad_norm': '0.008926', 'learning_rate': '5.747e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.65', 'tokens/total': 7897408, 'tokens/trainable': 119438, 'epoch': '1.02'}
+
 51%|█████     | 261/512 [29:19<28:49,  6.89s/it]
 51%|█████     | 262/512 [29:25<28:26,  6.82s/it]
                                                 
{'loss': '0.001416', 'grad_norm': '0.08671', 'learning_rate': '5.718e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 7927648, 'tokens/trainable': 119881, 'epoch': '1.023'}
+
 51%|█████     | 262/512 [29:25<28:26,  6.82s/it]
 51%|█████▏    | 263/512 [29:32<28:08,  6.78s/it]
                                                 
{'loss': '0.0004944', 'grad_norm': '0.03644', 'learning_rate': '5.689e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '27.72', 'tokens/total': 7957968, 'tokens/trainable': 120285, 'epoch': '1.027'}
+
 51%|█████▏    | 263/512 [29:32<28:08,  6.78s/it]
 52%|█████▏    | 264/512 [29:39<27:52,  6.74s/it]
                                                 
{'loss': '0.005234', 'grad_norm': '0.4136', 'learning_rate': '5.66e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 7987888, 'tokens/trainable': 120755, 'epoch': '1.031'}
+
 52%|█████▏    | 264/512 [29:39<27:52,  6.74s/it]
 52%|█████▏    | 265/512 [29:45<27:39,  6.72s/it]
                                                 
{'loss': '0.0001969', 'grad_norm': '0.01111', 'learning_rate': '5.631e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 8018048, 'tokens/trainable': 121194, 'epoch': '1.035'}
+
 52%|█████▏    | 265/512 [29:45<27:39,  6.72s/it]
 52%|█████▏    | 266/512 [29:52<27:32,  6.72s/it]
                                                 
{'loss': '0.005357', 'grad_norm': '0.2199', 'learning_rate': '5.602e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.54', 'tokens/total': 8048448, 'tokens/trainable': 121670, 'epoch': '1.039'}
+
 52%|█████▏    | 266/512 [29:52<27:32,  6.72s/it]
 52%|█████▏    | 267/512 [29:59<27:28,  6.73s/it]
                                                 
{'loss': '0.002412', 'grad_norm': '0.151', 'learning_rate': '5.573e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.58', 'tokens/total': 8078848, 'tokens/trainable': 122142, 'epoch': '1.043'}
+
 52%|█████▏    | 267/512 [29:59<27:28,  6.73s/it]
 52%|█████▏    | 268/512 [30:06<27:24,  6.74s/it]
                                                 
{'loss': '4.619e-05', 'grad_norm': '0.002497', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.36', 'tokens/total': 8109360, 'tokens/trainable': 122585, 'epoch': '1.047'}
+
 52%|█████▏    | 268/512 [30:06<27:24,  6.74s/it]
 53%|█████▎    | 269/512 [30:12<27:17,  6.74s/it]
                                                 
{'loss': '0.0002445', 'grad_norm': '0.0187', 'learning_rate': '5.515e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '38.78', 'tokens/total': 8139888, 'tokens/trainable': 123091, 'epoch': '1.051'}
+
 53%|█████▎    | 269/512 [30:12<27:17,  6.74s/it]
 53%|█████▎    | 270/512 [30:19<27:17,  6.77s/it]
                                                 
{'loss': '0.000507', 'grad_norm': '0.04384', 'learning_rate': '5.485e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '30.91', 'tokens/total': 8170480, 'tokens/trainable': 123505, 'epoch': '1.055'}
+
 53%|█████▎    | 270/512 [30:19<27:17,  6.77s/it]
 53%|█████▎    | 271/512 [30:25<26:31,  6.61s/it]
                                                 
{'loss': '0.0002814', 'grad_norm': '0.01803', 'learning_rate': '5.456e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.31', 'tokens/total': 8198848, 'tokens/trainable': 123953, 'epoch': '1.059'}
+
 53%|█████▎    | 271/512 [30:25<26:31,  6.61s/it]
 53%|█████▎    | 272/512 [30:32<26:35,  6.65s/it]
                                                 
{'loss': '0.0001505', 'grad_norm': '0.01605', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '28.61', 'tokens/total': 8229200, 'tokens/trainable': 124400, 'epoch': '1.062'}
+
 53%|█████▎    | 272/512 [30:32<26:35,  6.65s/it]
 53%|█████▎    | 273/512 [30:39<26:32,  6.66s/it]
                                                 
{'loss': '7.019e-05', 'grad_norm': '0.004128', 'learning_rate': '5.398e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 8259536, 'tokens/trainable': 124870, 'epoch': '1.066'}
+
 53%|█████▎    | 273/512 [30:39<26:32,  6.66s/it]
 54%|█████▎    | 274/512 [30:45<26:26,  6.67s/it]
                                                 
{'loss': '0.0002327', 'grad_norm': '0.01637', 'learning_rate': '5.369e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 8289904, 'tokens/trainable': 125333, 'epoch': '1.07'}
+
 54%|█████▎    | 274/512 [30:45<26:26,  6.67s/it]
 54%|█████▎    | 275/512 [30:52<26:25,  6.69s/it]
                                                 
{'loss': '2.309e-05', 'grad_norm': '0.0009976', 'learning_rate': '5.34e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 8320176, 'tokens/trainable': 125834, 'epoch': '1.074'}
+
 54%|█████▎    | 275/512 [30:52<26:25,  6.69s/it]
 54%|█████▍    | 276/512 [30:59<26:19,  6.69s/it]
                                                 
{'loss': '2.291e-05', 'grad_norm': '0.002002', 'learning_rate': '5.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.02', 'tokens/total': 8350832, 'tokens/trainable': 126316, 'epoch': '1.078'}
+
 54%|█████▍    | 276/512 [30:59<26:19,  6.69s/it]
 54%|█████▍    | 277/512 [31:06<26:14,  6.70s/it]
                                                 
{'loss': '6.588e-05', 'grad_norm': '0.005128', 'learning_rate': '5.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.33', 'tokens/total': 8380976, 'tokens/trainable': 126798, 'epoch': '1.082'}
+
 54%|█████▍    | 277/512 [31:06<26:14,  6.70s/it]
 54%|█████▍    | 278/512 [31:12<25:59,  6.66s/it]
                                                 
{'loss': '2.662e-05', 'grad_norm': '0.001301', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.64', 'tokens/total': 8411072, 'tokens/trainable': 127223, 'epoch': '1.086'}
+
 54%|█████▍    | 278/512 [31:12<25:59,  6.66s/it]
 54%|█████▍    | 279/512 [31:19<25:41,  6.62s/it]
                                                 
{'loss': '2.588e-05', 'grad_norm': '0.001754', 'learning_rate': '5.224e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 8440736, 'tokens/trainable': 127672, 'epoch': '1.09'}
+
 54%|█████▍    | 279/512 [31:19<25:41,  6.62s/it]
 55%|█████▍    | 280/512 [31:25<25:36,  6.62s/it]
                                                 
{'loss': '8.053e-05', 'grad_norm': '0.005083', 'learning_rate': '5.195e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 8470944, 'tokens/trainable': 128116, 'epoch': '1.094'}
+
 55%|█████▍    | 280/512 [31:25<25:36,  6.62s/it]
 55%|█████▍    | 281/512 [31:32<25:35,  6.65s/it]
                                                 
{'loss': '2.603e-05', 'grad_norm': '0.001439', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.82', 'tokens/total': 8501104, 'tokens/trainable': 128589, 'epoch': '1.098'}
+
 55%|█████▍    | 281/512 [31:32<25:35,  6.65s/it]
 55%|█████▌    | 282/512 [31:39<25:30,  6.66s/it]
                                                 
{'loss': '0.008352', 'grad_norm': '0.7785', 'learning_rate': '5.138e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 8531696, 'tokens/trainable': 129036, 'epoch': '1.102'}
+
 55%|█████▌    | 282/512 [31:39<25:30,  6.66s/it]
 55%|█████▌    | 283/512 [31:45<25:27,  6.67s/it]
                                                 
{'loss': '3.522e-05', 'grad_norm': '0.002119', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 8561936, 'tokens/trainable': 129485, 'epoch': '1.105'}
+
 55%|█████▌    | 283/512 [31:45<25:27,  6.67s/it]
 55%|█████▌    | 284/512 [31:52<25:23,  6.68s/it]
                                                 
{'loss': '0.0002057', 'grad_norm': '0.01936', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.55', 'tokens/total': 8592192, 'tokens/trainable': 129920, 'epoch': '1.109'}
+
 55%|█████▌    | 284/512 [31:52<25:23,  6.68s/it]
 56%|█████▌    | 285/512 [31:59<25:19,  6.69s/it]
                                                 
{'loss': '0.0002226', 'grad_norm': '0.03079', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.14', 'tokens/total': 8622432, 'tokens/trainable': 130417, 'epoch': '1.113'}
+
 56%|█████▌    | 285/512 [31:59<25:19,  6.69s/it]
 56%|█████▌    | 286/512 [32:06<25:13,  6.70s/it]
                                                 
{'loss': '0.002872', 'grad_norm': '0.2944', 'learning_rate': '5.022e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 8652768, 'tokens/trainable': 130903, 'epoch': '1.117'}
+
 56%|█████▌    | 286/512 [32:06<25:13,  6.70s/it]
 56%|█████▌    | 287/512 [32:12<25:05,  6.69s/it]
                                                 
{'loss': '2.243e-05', 'grad_norm': '0.001503', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.5', 'tokens/total': 8683296, 'tokens/trainable': 131343, 'epoch': '1.121'}
+
 56%|█████▌    | 287/512 [32:12<25:05,  6.69s/it]
 56%|█████▋    | 288/512 [32:19<24:57,  6.69s/it]
                                                 
{'loss': '0.002656', 'grad_norm': '0.2223', 'learning_rate': '4.964e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.23', 'tokens/total': 8713504, 'tokens/trainable': 131763, 'epoch': '1.125'}
+
 56%|█████▋    | 288/512 [32:19<24:57,  6.69s/it][2026-08-18 14:37:22,492] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
+
 56%|█████▋    | 289/512 [32:28<27:16,  7.34s/it]
                                                 
{'loss': '4.355e-05', 'grad_norm': '0.019', 'learning_rate': '4.935e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 8743888, 'tokens/trainable': 132229, 'epoch': '1.129'}
+
 56%|█████▋    | 289/512 [32:28<27:16,  7.34s/it]
 57%|█████▋    | 290/512 [32:34<26:28,  7.16s/it]
                                                 
{'loss': '2.45e-05', 'grad_norm': '0.00189', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.54', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 8774336, 'tokens/trainable': 132678, 'epoch': '1.133'}
+
 57%|█████▋    | 290/512 [32:34<26:28,  7.16s/it]
 57%|█████▋    | 291/512 [32:41<25:53,  7.03s/it]
                                                 
{'loss': '2.067e-05', 'grad_norm': '0.0009668', 'learning_rate': '4.878e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.28', 'tokens/total': 8804816, 'tokens/trainable': 133136, 'epoch': '1.137'}
+
 57%|█████▋    | 291/512 [32:41<25:53,  7.03s/it]
 57%|█████▋    | 292/512 [32:48<25:26,  6.94s/it]
                                                 
{'loss': '2.282e-05', 'grad_norm': '0.001047', 'learning_rate': '4.849e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 8835024, 'tokens/trainable': 133608, 'epoch': '1.141'}
+
 57%|█████▋    | 292/512 [32:48<25:26,  6.94s/it]
 57%|█████▋    | 293/512 [32:55<25:02,  6.86s/it]
                                                 
{'loss': '0.0002425', 'grad_norm': '0.0286', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '40.43', 'tokens/total': 8865408, 'tokens/trainable': 134108, 'epoch': '1.145'}
+
 57%|█████▋    | 293/512 [32:55<25:02,  6.86s/it]
 57%|█████▋    | 294/512 [33:01<24:46,  6.82s/it]
                                                 
{'loss': '3.146e-05', 'grad_norm': '0.002422', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.68', 'tokens/total': 8895824, 'tokens/trainable': 134541, 'epoch': '1.148'}
+
 57%|█████▋    | 294/512 [33:01<24:46,  6.82s/it]
 58%|█████▊    | 295/512 [33:08<24:31,  6.78s/it]
                                                 
{'loss': '5.829e-05', 'grad_norm': '0.003736', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.44', 'tokens/total': 8926336, 'tokens/trainable': 134966, 'epoch': '1.152'}
+
 58%|█████▊    | 295/512 [33:08<24:31,  6.78s/it]
 58%|█████▊    | 296/512 [33:15<24:16,  6.74s/it]
                                                 
{'loss': '0.0002222', 'grad_norm': '0.02378', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '33.53', 'tokens/total': 8956656, 'tokens/trainable': 135402, 'epoch': '1.156'}
+
 58%|█████▊    | 296/512 [33:15<24:16,  6.74s/it]
 58%|█████▊    | 297/512 [33:21<24:00,  6.70s/it]
                                                 
{'loss': '0.0005197', 'grad_norm': '0.03584', 'learning_rate': '4.706e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.02', 'tokens/total': 8986752, 'tokens/trainable': 135804, 'epoch': '1.16'}
+
 58%|█████▊    | 297/512 [33:21<24:00,  6.70s/it]
 58%|█████▊    | 298/512 [33:28<23:48,  6.67s/it]
                                                 
{'loss': '0.0007586', 'grad_norm': '0.1083', 'learning_rate': '4.677e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '31.13', 'tokens/total': 9017024, 'tokens/trainable': 136231, 'epoch': '1.164'}
+
 58%|█████▊    | 298/512 [33:28<23:48,  6.67s/it]
 58%|█████▊    | 299/512 [33:35<23:43,  6.68s/it]
                                                 
{'loss': '1.649e-05', 'grad_norm': '0.0007778', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 9047424, 'tokens/trainable': 136705, 'epoch': '1.168'}
+
 58%|█████▊    | 299/512 [33:35<23:43,  6.68s/it]
 59%|█████▊    | 300/512 [33:41<23:34,  6.67s/it]
                                                 
{'loss': '3.358e-05', 'grad_norm': '0.002182', 'learning_rate': '4.62e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.73', 'tokens/total': 9077648, 'tokens/trainable': 137178, 'epoch': '1.172'}
+
 59%|█████▊    | 300/512 [33:41<23:34,  6.67s/it]
 59%|█████▉    | 301/512 [33:48<23:30,  6.68s/it]
                                                 
{'loss': '9.5e-05', 'grad_norm': '0.01777', 'learning_rate': '4.592e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 9108032, 'tokens/trainable': 137643, 'epoch': '1.176'}
+
 59%|█████▉    | 301/512 [33:48<23:30,  6.68s/it]
 59%|█████▉    | 302/512 [33:55<23:23,  6.68s/it]
                                                 
{'loss': '0.001084', 'grad_norm': '0.08603', 'learning_rate': '4.563e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '31.91', 'tokens/total': 9138240, 'tokens/trainable': 138078, 'epoch': '1.18'}
+
 59%|█████▉    | 302/512 [33:55<23:23,  6.68s/it]
 59%|█████▉    | 303/512 [34:01<23:15,  6.67s/it]
                                                 
{'loss': '3.201e-05', 'grad_norm': '0.001786', 'learning_rate': '4.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '35', 'tokens/total': 9168352, 'tokens/trainable': 138515, 'epoch': '1.184'}
+
 59%|█████▉    | 303/512 [34:01<23:15,  6.67s/it]
 59%|█████▉    | 304/512 [34:08<23:11,  6.69s/it]
                                                 
{'loss': '2.666e-05', 'grad_norm': '0.001189', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '35.13', 'tokens/total': 9198864, 'tokens/trainable': 138948, 'epoch': '1.188'}
+
 59%|█████▉    | 304/512 [34:08<23:11,  6.69s/it]
 60%|█████▉    | 305/512 [34:15<23:05,  6.69s/it]
                                                 
{'loss': '0.0008552', 'grad_norm': '0.08659', 'learning_rate': '4.478e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '38.05', 'tokens/total': 9229104, 'tokens/trainable': 139408, 'epoch': '1.191'}
+
 60%|█████▉    | 305/512 [34:15<23:05,  6.69s/it]
 60%|█████▉    | 306/512 [34:21<22:26,  6.54s/it]
                                                 
{'loss': '2.396e-05', 'grad_norm': '0.001476', 'learning_rate': '4.45e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '33.38', 'tokens/total': 9257344, 'tokens/trainable': 139840, 'epoch': '1.195'}
+
 60%|█████▉    | 306/512 [34:21<22:26,  6.54s/it]
 60%|█████▉    | 307/512 [34:28<22:29,  6.58s/it]
                                                 
{'loss': '0.0001232', 'grad_norm': '0.01435', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 9287520, 'tokens/trainable': 140281, 'epoch': '1.199'}
+
 60%|█████▉    | 307/512 [34:28<22:29,  6.58s/it]
 60%|██████    | 308/512 [34:34<22:30,  6.62s/it]
                                                 
{'loss': '2.204e-05', 'grad_norm': '0.001902', 'learning_rate': '4.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '37.31', 'tokens/total': 9317888, 'tokens/trainable': 140776, 'epoch': '1.203'}
+
 60%|██████    | 308/512 [34:34<22:30,  6.62s/it]
 60%|██████    | 309/512 [34:41<22:27,  6.64s/it]
                                                 
{'loss': '2.58e-05', 'grad_norm': '0.001532', 'learning_rate': '4.366e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 9348256, 'tokens/trainable': 141228, 'epoch': '1.207'}
+
 60%|██████    | 309/512 [34:41<22:27,  6.64s/it]
 61%|██████    | 310/512 [34:48<22:22,  6.65s/it]
                                                 
{'loss': '6.094e-05', 'grad_norm': '0.006136', 'learning_rate': '4.338e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '37.19', 'tokens/total': 9378560, 'tokens/trainable': 141679, 'epoch': '1.211'}
+
 61%|██████    | 310/512 [34:48<22:22,  6.65s/it]
 61%|██████    | 311/512 [34:54<22:15,  6.65s/it]
                                                 
{'loss': '2.005e-05', 'grad_norm': '0.001599', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '32.25', 'tokens/total': 9408848, 'tokens/trainable': 142122, 'epoch': '1.215'}
+
 61%|██████    | 311/512 [34:54<22:15,  6.65s/it]
 61%|██████    | 312/512 [35:01<22:10,  6.65s/it]
                                                 
{'loss': '1.628e-05', 'grad_norm': '0.0007549', 'learning_rate': '4.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 9439344, 'tokens/trainable': 142561, 'epoch': '1.219'}
+
 61%|██████    | 312/512 [35:01<22:10,  6.65s/it]
 61%|██████    | 313/512 [35:08<22:07,  6.67s/it]
                                                 
{'loss': '1.512e-05', 'grad_norm': '0.0008527', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 9469792, 'tokens/trainable': 143046, 'epoch': '1.223'}
+
 61%|██████    | 313/512 [35:08<22:07,  6.67s/it]
 61%|██████▏   | 314/512 [35:14<22:01,  6.67s/it]
                                                 
{'loss': '0.0001308', 'grad_norm': '0.02262', 'learning_rate': '4.226e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 9500128, 'tokens/trainable': 143493, 'epoch': '1.227'}
+
 61%|██████▏   | 314/512 [35:14<22:01,  6.67s/it]
 62%|██████▏   | 315/512 [35:21<21:54,  6.67s/it]
                                                 
{'loss': '1.917e-05', 'grad_norm': '0.001019', 'learning_rate': '4.198e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.12', 'tokens/total': 9530480, 'tokens/trainable': 143956, 'epoch': '1.23'}
+
 62%|██████▏   | 315/512 [35:21<21:54,  6.67s/it]
 62%|██████▏   | 316/512 [35:28<21:48,  6.67s/it]
                                                 
{'loss': '0.0001811', 'grad_norm': '0.02337', 'learning_rate': '4.17e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 9560912, 'tokens/trainable': 144411, 'epoch': '1.234'}
+
 62%|██████▏   | 316/512 [35:28<21:48,  6.67s/it]
 62%|██████▏   | 317/512 [35:34<21:43,  6.68s/it]
                                                 
{'loss': '0.0001281', 'grad_norm': '0.0114', 'learning_rate': '4.143e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '38.66', 'tokens/total': 9591344, 'tokens/trainable': 144892, 'epoch': '1.238'}
+
 62%|██████▏   | 317/512 [35:34<21:43,  6.68s/it]
 62%|██████▏   | 318/512 [35:41<21:34,  6.67s/it]
                                                 
{'loss': '0.006938', 'grad_norm': '0.5837', 'learning_rate': '4.115e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 9621392, 'tokens/trainable': 145339, 'epoch': '1.242'}
+
 62%|██████▏   | 318/512 [35:41<21:34,  6.67s/it]
 62%|██████▏   | 319/512 [35:48<21:31,  6.69s/it]
                                                 
{'loss': '1.689e-05', 'grad_norm': '0.000792', 'learning_rate': '4.087e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '36.38', 'tokens/total': 9651728, 'tokens/trainable': 145811, 'epoch': '1.246'}
+
 62%|██████▏   | 319/512 [35:48<21:31,  6.69s/it]
 62%|██████▎   | 320/512 [35:55<21:28,  6.71s/it]
                                                 
{'loss': '0.0002538', 'grad_norm': '0.02599', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '38.62', 'tokens/total': 9682224, 'tokens/trainable': 146314, 'epoch': '1.25'}
+
 62%|██████▎   | 320/512 [35:55<21:28,  6.71s/it][2026-08-18 14:40:58,131] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
+
 63%|██████▎   | 321/512 [36:03<23:10,  7.28s/it]
                                                 
{'loss': '1.948e-05', 'grad_norm': '0.001073', 'learning_rate': '4.032e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '31.72', 'tokens/total': 9712656, 'tokens/trainable': 146781, 'epoch': '1.254'}
+
 63%|██████▎   | 321/512 [36:03<23:10,  7.28s/it]
 63%|██████▎   | 322/512 [36:10<22:26,  7.09s/it]
                                                 
{'loss': '0.0008847', 'grad_norm': '0.08435', 'learning_rate': '4.005e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '34.89', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 9743088, 'tokens/trainable': 147225, 'epoch': '1.258'}
+
 63%|██████▎   | 322/512 [36:10<22:26,  7.09s/it]
 63%|██████▎   | 323/512 [36:17<22:01,  6.99s/it]
                                                 
{'loss': '0.0004002', 'grad_norm': '0.09092', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.17', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 9773808, 'tokens/trainable': 147673, 'epoch': '1.262'}
+
 63%|██████▎   | 323/512 [36:17<22:01,  6.99s/it]
 63%|██████▎   | 324/512 [36:23<21:31,  6.87s/it]
                                                 
{'loss': '0.004064', 'grad_norm': '0.386', 'learning_rate': '3.95e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.23', 'tokens/total': 9804016, 'tokens/trainable': 148103, 'epoch': '1.266'}
+
 63%|██████▎   | 324/512 [36:23<21:31,  6.87s/it]
 63%|██████▎   | 325/512 [36:30<21:12,  6.80s/it]
                                                 
{'loss': '9.267e-06', 'grad_norm': '0.0004181', 'learning_rate': '3.923e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '31.74', 'tokens/total': 9834592, 'tokens/trainable': 148535, 'epoch': '1.27'}
+
 63%|██████▎   | 325/512 [36:30<21:12,  6.80s/it]
 64%|██████▎   | 326/512 [36:36<20:59,  6.77s/it]
                                                 
{'loss': '0.0002955', 'grad_norm': '0.03044', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 9864848, 'tokens/trainable': 148999, 'epoch': '1.273'}
+
 64%|██████▎   | 326/512 [36:36<20:59,  6.77s/it]
 64%|██████▍   | 327/512 [36:43<20:47,  6.74s/it]
                                                 
{'loss': '9.424e-06', 'grad_norm': '0.0003461', 'learning_rate': '3.869e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 9895168, 'tokens/trainable': 149431, 'epoch': '1.277'}
+
 64%|██████▍   | 327/512 [36:43<20:47,  6.74s/it]
 64%|██████▍   | 328/512 [36:50<20:35,  6.71s/it]
                                                 
{'loss': '0.004382', 'grad_norm': '0.3263', 'learning_rate': '3.842e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 9925696, 'tokens/trainable': 149886, 'epoch': '1.281'}
+
 64%|██████▍   | 328/512 [36:50<20:35,  6.71s/it]
 64%|██████▍   | 329/512 [36:56<20:21,  6.67s/it]
                                                 
{'loss': '0.0003639', 'grad_norm': '0.06806', 'learning_rate': '3.815e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '31.01', 'tokens/total': 9955648, 'tokens/trainable': 150288, 'epoch': '1.285'}
+
 64%|██████▍   | 329/512 [36:56<20:21,  6.67s/it]
 64%|██████▍   | 330/512 [37:03<20:15,  6.68s/it]
                                                 
{'loss': '3.247e-05', 'grad_norm': '0.004817', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.94', 'tokens/total': 9985856, 'tokens/trainable': 150733, 'epoch': '1.289'}
+
 64%|██████▍   | 330/512 [37:03<20:15,  6.68s/it]
 65%|██████▍   | 331/512 [37:10<20:09,  6.68s/it]
                                                 
{'loss': '0.0003107', 'grad_norm': '0.06924', 'learning_rate': '3.761e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 10016208, 'tokens/trainable': 151207, 'epoch': '1.293'}
+
 65%|██████▍   | 331/512 [37:10<20:09,  6.68s/it]
 65%|██████▍   | 332/512 [37:16<20:04,  6.69s/it]
                                                 
{'loss': '0.001514', 'grad_norm': '0.1473', 'learning_rate': '3.734e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.61', 'tokens/total': 10046512, 'tokens/trainable': 151694, 'epoch': '1.297'}
+
 65%|██████▍   | 332/512 [37:16<20:04,  6.69s/it]
 65%|██████▌   | 333/512 [37:23<20:00,  6.71s/it]
                                                 
{'loss': '7.81e-05', 'grad_norm': '0.007326', 'learning_rate': '3.708e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '38.37', 'tokens/total': 10076944, 'tokens/trainable': 152188, 'epoch': '1.301'}
+
 65%|██████▌   | 333/512 [37:23<20:00,  6.71s/it]
 65%|██████▌   | 334/512 [37:30<19:50,  6.69s/it]
                                                 
{'loss': '1.274e-05', 'grad_norm': '0.000476', 'learning_rate': '3.681e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.69', 'tokens/total': 10107296, 'tokens/trainable': 152596, 'epoch': '1.305'}
+
 65%|██████▌   | 334/512 [37:30<19:50,  6.69s/it]
 65%|██████▌   | 335/512 [37:37<19:43,  6.69s/it]
                                                 
{'loss': '0.0004747', 'grad_norm': '0.07003', 'learning_rate': '3.655e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.57', 'tokens/total': 10137568, 'tokens/trainable': 153052, 'epoch': '1.309'}
+
 65%|██████▌   | 335/512 [37:37<19:43,  6.69s/it]
 66%|██████▌   | 336/512 [37:43<19:38,  6.70s/it]
                                                 
{'loss': '0.0004141', 'grad_norm': '0.06498', 'learning_rate': '3.628e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.19', 'tokens/total': 10167952, 'tokens/trainable': 153491, 'epoch': '1.312'}
+
 66%|██████▌   | 336/512 [37:43<19:38,  6.70s/it]
 66%|██████▌   | 337/512 [37:50<19:34,  6.71s/it]
                                                 
{'loss': '7.06e-05', 'grad_norm': '0.009992', 'learning_rate': '3.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '40.43', 'tokens/total': 10198512, 'tokens/trainable': 153983, 'epoch': '1.316'}
+
 66%|██████▌   | 337/512 [37:50<19:34,  6.71s/it]
 66%|██████▌   | 338/512 [37:57<19:27,  6.71s/it]
                                                 
{'loss': '0.0002046', 'grad_norm': '0.03267', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.72', 'tokens/total': 10228752, 'tokens/trainable': 154464, 'epoch': '1.32'}
+
 66%|██████▌   | 338/512 [37:57<19:27,  6.71s/it]
 66%|██████▌   | 339/512 [38:03<19:18,  6.69s/it]
                                                 
{'loss': '8.338e-06', 'grad_norm': '0.0003953', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.87', 'tokens/total': 10259104, 'tokens/trainable': 154924, 'epoch': '1.324'}
+
 66%|██████▌   | 339/512 [38:03<19:18,  6.69s/it]
 66%|██████▋   | 340/512 [38:10<19:09,  6.68s/it]
                                                 
{'loss': '1.682e-05', 'grad_norm': '0.001057', 'learning_rate': '3.523e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.81', 'tokens/total': 10289520, 'tokens/trainable': 155373, 'epoch': '1.328'}
+
 66%|██████▋   | 340/512 [38:10<19:09,  6.68s/it]
 67%|██████▋   | 341/512 [38:17<19:04,  6.69s/it]
                                                 
{'loss': '1.703e-05', 'grad_norm': '0.0008337', 'learning_rate': '3.497e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.47', 'tokens/total': 10320016, 'tokens/trainable': 155873, 'epoch': '1.332'}
+
 67%|██████▋   | 341/512 [38:17<19:04,  6.69s/it]
 67%|██████▋   | 342/512 [38:23<18:55,  6.68s/it]
                                                 
{'loss': '1.627e-05', 'grad_norm': '0.0007872', 'learning_rate': '3.471e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.42', 'tokens/total': 10350368, 'tokens/trainable': 156278, 'epoch': '1.336'}
+
 67%|██████▋   | 342/512 [38:23<18:55,  6.68s/it]
 67%|██████▋   | 343/512 [38:30<18:49,  6.68s/it]
                                                 
{'loss': '0.0009315', 'grad_norm': '0.2122', 'learning_rate': '3.445e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.4', 'tokens/total': 10380624, 'tokens/trainable': 156733, 'epoch': '1.34'}
+
 67%|██████▋   | 343/512 [38:30<18:49,  6.68s/it]
 67%|██████▋   | 344/512 [38:37<18:45,  6.70s/it]
                                                 
{'loss': '7.496e-05', 'grad_norm': '0.01747', 'learning_rate': '3.42e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 10411024, 'tokens/trainable': 157203, 'epoch': '1.344'}
+
 67%|██████▋   | 344/512 [38:37<18:45,  6.70s/it]
 67%|██████▋   | 345/512 [38:44<18:39,  6.70s/it]
                                                 
{'loss': '0.004417', 'grad_norm': '0.2606', 'learning_rate': '3.394e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '37.8', 'tokens/total': 10441248, 'tokens/trainable': 157707, 'epoch': '1.348'}
+
 67%|██████▋   | 345/512 [38:44<18:39,  6.70s/it]
 68%|██████▊   | 346/512 [38:50<18:36,  6.73s/it]
                                                 
{'loss': '7.685e-06', 'grad_norm': '0.0003296', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 10471712, 'tokens/trainable': 158180, 'epoch': '1.352'}
+
 68%|██████▊   | 346/512 [38:50<18:36,  6.73s/it]
 68%|██████▊   | 347/512 [38:57<18:29,  6.72s/it]
                                                 
{'loss': '2.94e-05', 'grad_norm': '0.00212', 'learning_rate': '3.343e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '34.41', 'tokens/total': 10502288, 'tokens/trainable': 158646, 'epoch': '1.355'}
+
 68%|██████▊   | 347/512 [38:57<18:29,  6.72s/it]
 68%|██████▊   | 348/512 [39:04<18:16,  6.68s/it]
                                                 
{'loss': '9.501e-06', 'grad_norm': '0.0004069', 'learning_rate': '3.317e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '35.66', 'tokens/total': 10532304, 'tokens/trainable': 159115, 'epoch': '1.359'}
+
 68%|██████▊   | 348/512 [39:04<18:16,  6.68s/it]
 68%|██████▊   | 349/512 [39:10<18:07,  6.67s/it]
                                                 
{'loss': '9.019e-05', 'grad_norm': '0.006379', 'learning_rate': '3.292e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '31.62', 'tokens/total': 10562608, 'tokens/trainable': 159538, 'epoch': '1.363'}
+
 68%|██████▊   | 349/512 [39:10<18:07,  6.67s/it]
 68%|██████▊   | 350/512 [39:17<18:01,  6.68s/it]
                                                 
{'loss': '5.146e-05', 'grad_norm': '0.005556', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 10592992, 'tokens/trainable': 160016, 'epoch': '1.367'}
+
 68%|██████▊   | 350/512 [39:17<18:01,  6.68s/it]
 69%|██████▊   | 351/512 [39:24<17:53,  6.67s/it]
                                                 
{'loss': '9.038e-05', 'grad_norm': '0.009833', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '34.57', 'tokens/total': 10623328, 'tokens/trainable': 160475, 'epoch': '1.371'}
+
 69%|██████▊   | 351/512 [39:24<17:53,  6.67s/it]
 69%|██████▉   | 352/512 [39:30<17:47,  6.67s/it]
                                                 
{'loss': '7.51e-06', 'grad_norm': '0.0002588', 'learning_rate': '3.217e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 10653632, 'tokens/trainable': 160952, 'epoch': '1.375'}
+
 69%|██████▉   | 352/512 [39:30<17:47,  6.67s/it][2026-08-18 14:44:33,881] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
+
 69%|██████▉   | 353/512 [39:39<19:10,  7.24s/it]
                                                 
{'loss': '0.0002666', 'grad_norm': '0.03195', 'learning_rate': '3.192e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 10684176, 'tokens/trainable': 161420, 'epoch': '1.379'}
+
 69%|██████▉   | 353/512 [39:39<19:10,  7.24s/it]
 69%|██████▉   | 354/512 [39:45<18:34,  7.05s/it]
                                                 
{'loss': '0.0007417', 'grad_norm': '0.07013', 'learning_rate': '3.167e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 10714352, 'tokens/trainable': 161868, 'epoch': '1.383'}
+
 69%|██████▉   | 354/512 [39:45<18:34,  7.05s/it]
 69%|██████▉   | 355/512 [39:52<18:09,  6.94s/it]
                                                 
{'loss': '1.038e-05', 'grad_norm': '0.0005823', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '35.44', 'tokens/total': 10744464, 'tokens/trainable': 162318, 'epoch': '1.387'}
+
 69%|██████▉   | 355/512 [39:52<18:09,  6.94s/it]
 70%|██████▉   | 356/512 [39:59<17:52,  6.87s/it]
                                                 
{'loss': '7.898e-05', 'grad_norm': '0.01244', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.54', 'tokens/total': 10774976, 'tokens/trainable': 162789, 'epoch': '1.391'}
+
 70%|██████▉   | 356/512 [39:59<17:52,  6.87s/it]
 70%|██████▉   | 357/512 [40:06<17:37,  6.82s/it]
                                                 
{'loss': '2.485e-05', 'grad_norm': '0.00456', 'learning_rate': '3.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.43', 'tokens/total': 10805600, 'tokens/trainable': 163254, 'epoch': '1.395'}
+
 70%|██████▉   | 357/512 [40:06<17:37,  6.82s/it]
 70%|██████▉   | 358/512 [40:12<17:22,  6.77s/it]
                                                 
{'loss': '0.002299', 'grad_norm': '0.3044', 'learning_rate': '3.068e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.62', 'tokens/total': 10835920, 'tokens/trainable': 163715, 'epoch': '1.398'}
+
 70%|██████▉   | 358/512 [40:12<17:22,  6.77s/it]
 70%|███████   | 359/512 [40:19<17:12,  6.75s/it]
                                                 
{'loss': '1.963e-05', 'grad_norm': '0.002458', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 10866048, 'tokens/trainable': 164197, 'epoch': '1.402'}
+
 70%|███████   | 359/512 [40:19<17:12,  6.75s/it]
 70%|███████   | 360/512 [40:26<17:02,  6.73s/it]
                                                 
{'loss': '2.745e-05', 'grad_norm': '0.002477', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.74', 'tokens/total': 10896288, 'tokens/trainable': 164661, 'epoch': '1.406'}
+
 70%|███████   | 360/512 [40:26<17:02,  6.73s/it]
 71%|███████   | 361/512 [40:32<16:52,  6.70s/it]
                                                 
{'loss': '6.423e-06', 'grad_norm': '0.0002672', 'learning_rate': '2.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 10926816, 'tokens/trainable': 165111, 'epoch': '1.41'}
+
 71%|███████   | 361/512 [40:32<16:52,  6.70s/it]
 71%|███████   | 362/512 [40:38<16:19,  6.53s/it]
                                                 
{'loss': '1.355e-05', 'grad_norm': '0.0008353', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '30.52', 'tokens/total': 10954928, 'tokens/trainable': 165552, 'epoch': '1.414'}
+
 71%|███████   | 362/512 [40:38<16:19,  6.53s/it]
 71%|███████   | 363/512 [40:45<16:19,  6.57s/it]
                                                 
{'loss': '6.428e-05', 'grad_norm': '0.006071', 'learning_rate': '2.947e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 10985328, 'tokens/trainable': 166026, 'epoch': '1.418'}
+
 71%|███████   | 363/512 [40:45<16:19,  6.57s/it]
 71%|███████   | 364/512 [40:52<16:17,  6.60s/it]
                                                 
{'loss': '0.004244', 'grad_norm': '0.2542', 'learning_rate': '2.924e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '38.11', 'tokens/total': 11015664, 'tokens/trainable': 166486, 'epoch': '1.422'}
+
 71%|███████   | 364/512 [40:52<16:17,  6.60s/it]
 71%|███████▏  | 365/512 [40:58<16:12,  6.62s/it]
                                                 
{'loss': '8.404e-06', 'grad_norm': '0.0005575', 'learning_rate': '2.9e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 11046256, 'tokens/trainable': 166959, 'epoch': '1.426'}
+
 71%|███████▏  | 365/512 [40:58<16:12,  6.62s/it]
 71%|███████▏  | 366/512 [41:05<16:04,  6.61s/it]
                                                 
{'loss': '1.474e-05', 'grad_norm': '0.001434', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 11076288, 'tokens/trainable': 167394, 'epoch': '1.43'}
+
 71%|███████▏  | 366/512 [41:05<16:04,  6.61s/it]
 72%|███████▏  | 367/512 [41:12<15:58,  6.61s/it]
                                                 
{'loss': '1.328e-05', 'grad_norm': '0.001015', 'learning_rate': '2.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '28.58', 'tokens/total': 11106416, 'tokens/trainable': 167832, 'epoch': '1.434'}
+
 72%|███████▏  | 367/512 [41:12<15:58,  6.61s/it]
 72%|███████▏  | 368/512 [41:18<15:53,  6.62s/it]
                                                 
{'loss': '0.0006477', 'grad_norm': '0.1194', 'learning_rate': '2.829e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.02', 'tokens/total': 11136848, 'tokens/trainable': 168242, 'epoch': '1.438'}
+
 72%|███████▏  | 368/512 [41:18<15:53,  6.62s/it]
 72%|███████▏  | 369/512 [41:25<15:48,  6.63s/it]
                                                 
{'loss': '7.449e-05', 'grad_norm': '0.006878', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.74', 'tokens/total': 11167376, 'tokens/trainable': 168715, 'epoch': '1.441'}
+
 72%|███████▏  | 369/512 [41:25<15:48,  6.63s/it]
 72%|███████▏  | 370/512 [41:32<15:44,  6.65s/it]
                                                 
{'loss': '1.144e-05', 'grad_norm': '0.0007151', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.86', 'tokens/total': 11197776, 'tokens/trainable': 169147, 'epoch': '1.445'}
+
 72%|███████▏  | 370/512 [41:32<15:44,  6.65s/it]
 72%|███████▏  | 371/512 [41:38<15:37,  6.65s/it]
                                                 
{'loss': '5.924e-05', 'grad_norm': '0.009094', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.35', 'tokens/total': 11227840, 'tokens/trainable': 169628, 'epoch': '1.449'}
+
 72%|███████▏  | 371/512 [41:38<15:37,  6.65s/it]
 73%|███████▎  | 372/512 [41:45<15:32,  6.66s/it]
                                                 
{'loss': '7.548e-06', 'grad_norm': '0.0002473', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 11258176, 'tokens/trainable': 170085, 'epoch': '1.453'}
+
 73%|███████▎  | 372/512 [41:45<15:32,  6.66s/it]
 73%|███████▎  | 373/512 [41:51<15:24,  6.65s/it]
                                                 
{'loss': '5.401e-06', 'grad_norm': '0.0003029', 'learning_rate': '2.714e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 11288336, 'tokens/trainable': 170584, 'epoch': '1.457'}
+
 73%|███████▎  | 373/512 [41:51<15:24,  6.65s/it]
 73%|███████▎  | 374/512 [41:58<15:18,  6.66s/it]
                                                 
{'loss': '2.381e-05', 'grad_norm': '0.001942', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.68', 'tokens/total': 11318848, 'tokens/trainable': 171058, 'epoch': '1.461'}
+
 73%|███████▎  | 374/512 [41:58<15:18,  6.66s/it]
 73%|███████▎  | 375/512 [42:05<15:13,  6.67s/it]
                                                 
{'loss': '1.523e-05', 'grad_norm': '0.001931', 'learning_rate': '2.668e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.94', 'tokens/total': 11349168, 'tokens/trainable': 171518, 'epoch': '1.465'}
+
 73%|███████▎  | 375/512 [42:05<15:13,  6.67s/it]
 73%|███████▎  | 376/512 [42:11<15:04,  6.65s/it]
                                                 
{'loss': '0.00635', 'grad_norm': '0.6044', 'learning_rate': '2.646e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '40.62', 'tokens/total': 11379328, 'tokens/trainable': 172007, 'epoch': '1.469'}
+
 73%|███████▎  | 376/512 [42:11<15:04,  6.65s/it]
 74%|███████▎  | 377/512 [42:18<14:57,  6.65s/it]
                                                 
{'loss': '0.001872', 'grad_norm': '0.2545', 'learning_rate': '2.624e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 11409568, 'tokens/trainable': 172468, 'epoch': '1.473'}
+
 74%|███████▎  | 377/512 [42:18<14:57,  6.65s/it]
 74%|███████▍  | 378/512 [42:25<14:51,  6.66s/it]
                                                 
{'loss': '3.05e-05', 'grad_norm': '0.001921', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 11439968, 'tokens/trainable': 172923, 'epoch': '1.477'}
+
 74%|███████▍  | 378/512 [42:25<14:51,  6.66s/it]
 74%|███████▍  | 379/512 [42:31<14:44,  6.65s/it]
                                                 
{'loss': '0.02674', 'grad_norm': '0.3349', 'learning_rate': '2.579e-05', 'ppl': '1.027', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 11470496, 'tokens/trainable': 173370, 'epoch': '1.48'}
+
 74%|███████▍  | 379/512 [42:31<14:44,  6.65s/it]
 74%|███████▍  | 380/512 [42:38<14:37,  6.65s/it]
                                                 
{'loss': '5.195e-05', 'grad_norm': '0.006178', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 11500720, 'tokens/trainable': 173837, 'epoch': '1.484'}
+
 74%|███████▍  | 380/512 [42:38<14:37,  6.65s/it]
 74%|███████▍  | 381/512 [42:45<14:32,  6.66s/it]
                                                 
{'loss': '0.0006311', 'grad_norm': '0.06973', 'learning_rate': '2.535e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.54', 'tokens/total': 11531280, 'tokens/trainable': 174294, 'epoch': '1.488'}
+
 74%|███████▍  | 381/512 [42:45<14:32,  6.66s/it]
 75%|███████▍  | 382/512 [42:51<14:23,  6.64s/it]
                                                 
{'loss': '4.85e-05', 'grad_norm': '0.003902', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.84', 'tokens/total': 11561264, 'tokens/trainable': 174701, 'epoch': '1.492'}
+
 75%|███████▍  | 382/512 [42:51<14:23,  6.64s/it]
 75%|███████▍  | 383/512 [42:58<14:18,  6.65s/it]
                                                 
{'loss': '2.349e-05', 'grad_norm': '0.0007975', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 11591616, 'tokens/trainable': 175131, 'epoch': '1.496'}
+
 75%|███████▍  | 383/512 [42:58<14:18,  6.65s/it]
 75%|███████▌  | 384/512 [43:05<14:12,  6.66s/it]
                                                 
{'loss': '0.000115', 'grad_norm': '0.01254', 'learning_rate': '2.47e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 11622000, 'tokens/trainable': 175586, 'epoch': '1.5'}
+
 75%|███████▌  | 384/512 [43:05<14:12,  6.66s/it][2026-08-18 14:48:08,327] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
+
 75%|███████▌  | 385/512 [43:13<15:16,  7.21s/it]
                                                 
{'loss': '6.503e-05', 'grad_norm': '0.004752', 'learning_rate': '2.449e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 11652352, 'tokens/trainable': 176026, 'epoch': '1.504'}
+
 75%|███████▌  | 385/512 [43:13<15:16,  7.21s/it]
 75%|███████▌  | 386/512 [43:20<14:49,  7.06s/it]
                                                 
{'loss': '8.075e-05', 'grad_norm': '0.01419', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 11682736, 'tokens/trainable': 176512, 'epoch': '1.508'}
+
 75%|███████▌  | 386/512 [43:20<14:49,  7.06s/it]
 76%|███████▌  | 387/512 [43:27<14:28,  6.95s/it]
                                                 
{'loss': '8.846e-05', 'grad_norm': '0.007288', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '30.45', 'tokens/total': 11713136, 'tokens/trainable': 176943, 'epoch': '1.512'}
+
 76%|███████▌  | 387/512 [43:27<14:28,  6.95s/it]
 76%|███████▌  | 388/512 [43:33<14:10,  6.86s/it]
                                                 
{'loss': '0.0007253', 'grad_norm': '0.1016', 'learning_rate': '2.385e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.23', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 11743808, 'tokens/trainable': 177370, 'epoch': '1.516'}
+
 76%|███████▌  | 388/512 [43:33<14:10,  6.86s/it]
 76%|███████▌  | 389/512 [43:40<13:57,  6.81s/it]
                                                 
{'loss': '8.653e-05', 'grad_norm': '0.005729', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.58', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 11774144, 'tokens/trainable': 177851, 'epoch': '1.52'}
+
 76%|███████▌  | 389/512 [43:40<13:57,  6.81s/it]
 76%|███████▌  | 390/512 [43:47<13:45,  6.77s/it]
                                                 
{'loss': '0.0007951', 'grad_norm': '0.07154', 'learning_rate': '2.344e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.75', 'tokens/total': 11804272, 'tokens/trainable': 178310, 'epoch': '1.523'}
+
 76%|███████▌  | 390/512 [43:47<13:45,  6.77s/it]
 76%|███████▋  | 391/512 [43:53<13:18,  6.60s/it]
                                                 
{'loss': '0.0001062', 'grad_norm': '0.009775', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 11832432, 'tokens/trainable': 178736, 'epoch': '1.527'}
+
 76%|███████▋  | 391/512 [43:53<13:18,  6.60s/it]
 77%|███████▋  | 392/512 [44:00<13:18,  6.65s/it]
                                                 
{'loss': '2.955e-05', 'grad_norm': '0.001555', 'learning_rate': '2.303e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 11862960, 'tokens/trainable': 179233, 'epoch': '1.531'}
+
 77%|███████▋  | 392/512 [44:00<13:18,  6.65s/it]
 77%|███████▋  | 393/512 [44:06<13:13,  6.67s/it]
                                                 
{'loss': '0.0002111', 'grad_norm': '0.02611', 'learning_rate': '2.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.57', 'tokens/total': 11893232, 'tokens/trainable': 179730, 'epoch': '1.535'}
+
 77%|███████▋  | 393/512 [44:06<13:13,  6.67s/it]
 77%|███████▋  | 394/512 [44:13<13:08,  6.68s/it]
                                                 
{'loss': '0.0002601', 'grad_norm': '0.02035', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.87', 'tokens/total': 11923664, 'tokens/trainable': 180227, 'epoch': '1.539'}
+
 77%|███████▋  | 394/512 [44:13<13:08,  6.68s/it]
 77%|███████▋  | 395/512 [44:20<13:02,  6.69s/it]
                                                 
{'loss': '4.929e-05', 'grad_norm': '0.00542', 'learning_rate': '2.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 11954032, 'tokens/trainable': 180687, 'epoch': '1.543'}
+
 77%|███████▋  | 395/512 [44:20<13:02,  6.69s/it]
 77%|███████▋  | 396/512 [44:26<12:56,  6.69s/it]
                                                 
{'loss': '6.144e-05', 'grad_norm': '0.0031', 'learning_rate': '2.222e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 11984512, 'tokens/trainable': 181141, 'epoch': '1.547'}
+
 77%|███████▋  | 396/512 [44:26<12:56,  6.69s/it]
 78%|███████▊  | 397/512 [44:33<12:51,  6.71s/it]
                                                 
{'loss': '9.172e-05', 'grad_norm': '0.006749', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.02', 'tokens/total': 12014880, 'tokens/trainable': 181648, 'epoch': '1.551'}
+
 78%|███████▊  | 397/512 [44:33<12:51,  6.71s/it]
 78%|███████▊  | 398/512 [44:40<12:42,  6.69s/it]
                                                 
{'loss': '0.001001', 'grad_norm': '0.08928', 'learning_rate': '2.183e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 12044928, 'tokens/trainable': 182109, 'epoch': '1.555'}
+
 78%|███████▊  | 398/512 [44:40<12:42,  6.69s/it]
 78%|███████▊  | 399/512 [44:47<12:35,  6.69s/it]
                                                 
{'loss': '5.344e-05', 'grad_norm': '0.004652', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.58', 'tokens/total': 12075344, 'tokens/trainable': 182559, 'epoch': '1.559'}
+
 78%|███████▊  | 399/512 [44:47<12:35,  6.69s/it]
 78%|███████▊  | 400/512 [44:53<12:27,  6.67s/it]
                                                 
{'loss': '4.57e-05', 'grad_norm': '0.001751', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.17', 'tokens/total': 12105616, 'tokens/trainable': 182985, 'epoch': '1.562'}
+
 78%|███████▊  | 400/512 [44:53<12:27,  6.67s/it]
 78%|███████▊  | 401/512 [45:00<12:20,  6.67s/it]
                                                 
{'loss': '0.0001243', 'grad_norm': '0.01575', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 12136080, 'tokens/trainable': 183440, 'epoch': '1.566'}
+
 78%|███████▊  | 401/512 [45:00<12:20,  6.67s/it]
 79%|███████▊  | 402/512 [45:07<12:14,  6.68s/it]
                                                 
{'loss': '0.0001223', 'grad_norm': '0.007152', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '34.51', 'tokens/total': 12166240, 'tokens/trainable': 183917, 'epoch': '1.57'}
+
 79%|███████▊  | 402/512 [45:07<12:14,  6.68s/it]
 79%|███████▊  | 403/512 [45:13<12:07,  6.67s/it]
                                                 
{'loss': '3.622e-05', 'grad_norm': '0.001534', 'learning_rate': '2.086e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '30.36', 'tokens/total': 12196544, 'tokens/trainable': 184348, 'epoch': '1.574'}
+
 79%|███████▊  | 403/512 [45:13<12:07,  6.67s/it]
 79%|███████▉  | 404/512 [45:20<12:00,  6.67s/it]
                                                 
{'loss': '6.368e-05', 'grad_norm': '0.003521', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 12226784, 'tokens/trainable': 184786, 'epoch': '1.578'}
+
 79%|███████▉  | 404/512 [45:20<12:00,  6.67s/it]
 79%|███████▉  | 405/512 [45:27<11:54,  6.68s/it]
                                                 
{'loss': '2.033e-05', 'grad_norm': '0.0007212', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 12257264, 'tokens/trainable': 185249, 'epoch': '1.582'}
+
 79%|███████▉  | 405/512 [45:27<11:54,  6.68s/it]
 79%|███████▉  | 406/512 [45:33<11:49,  6.69s/it]
                                                 
{'loss': '3.38e-05', 'grad_norm': '0.00148', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.43', 'tokens/total': 12287584, 'tokens/trainable': 185680, 'epoch': '1.586'}
+
 79%|███████▉  | 406/512 [45:33<11:49,  6.69s/it]
 79%|███████▉  | 407/512 [45:40<11:41,  6.68s/it]
                                                 
{'loss': '5.18e-05', 'grad_norm': '0.004398', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 12317984, 'tokens/trainable': 186100, 'epoch': '1.59'}
+
 79%|███████▉  | 407/512 [45:40<11:41,  6.68s/it]
 80%|███████▉  | 408/512 [45:47<11:33,  6.67s/it]
                                                 
{'loss': '3.441e-05', 'grad_norm': '0.002647', 'learning_rate': '1.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.84', 'tokens/total': 12348144, 'tokens/trainable': 186565, 'epoch': '1.594'}
+
 80%|███████▉  | 408/512 [45:47<11:33,  6.67s/it]
 80%|███████▉  | 409/512 [45:53<11:27,  6.67s/it]
                                                 
{'loss': '0.000216', 'grad_norm': '0.02899', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.78', 'tokens/total': 12378544, 'tokens/trainable': 187017, 'epoch': '1.598'}
+
 80%|███████▉  | 409/512 [45:53<11:27,  6.67s/it]
 80%|████████  | 410/512 [46:00<11:21,  6.68s/it]
                                                 
{'loss': '9.646e-05', 'grad_norm': '0.007093', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 12408784, 'tokens/trainable': 187468, 'epoch': '1.602'}
+
 80%|████████  | 410/512 [46:00<11:21,  6.68s/it]
 80%|████████  | 411/512 [46:07<11:14,  6.67s/it]
                                                 
{'loss': '0.0003458', 'grad_norm': '0.02711', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.57', 'tokens/total': 12438832, 'tokens/trainable': 187961, 'epoch': '1.605'}
+
 80%|████████  | 411/512 [46:07<11:14,  6.67s/it]
 80%|████████  | 412/512 [46:13<11:07,  6.68s/it]
                                                 
{'loss': '0.003292', 'grad_norm': '0.2949', 'learning_rate': '1.922e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.93', 'tokens/total': 12469296, 'tokens/trainable': 188447, 'epoch': '1.609'}
+
 80%|████████  | 412/512 [46:13<11:07,  6.68s/it]
 81%|████████  | 413/512 [46:20<11:00,  6.67s/it]
                                                 
{'loss': '0.003581', 'grad_norm': '0.2288', 'learning_rate': '1.904e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '35.97', 'tokens/total': 12499392, 'tokens/trainable': 188903, 'epoch': '1.613'}
+
 81%|████████  | 413/512 [46:20<11:00,  6.67s/it]
 81%|████████  | 414/512 [46:27<10:56,  6.70s/it]
                                                 
{'loss': '6.048e-05', 'grad_norm': '0.002635', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 12529952, 'tokens/trainable': 189398, 'epoch': '1.617'}
+
 81%|████████  | 414/512 [46:27<10:56,  6.70s/it]
 81%|████████  | 415/512 [46:33<10:50,  6.70s/it]
                                                 
{'loss': '4.138e-05', 'grad_norm': '0.001722', 'learning_rate': '1.87e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 12560240, 'tokens/trainable': 189850, 'epoch': '1.621'}
+
 81%|████████  | 415/512 [46:33<10:50,  6.70s/it]
 81%|████████▏ | 416/512 [46:40<10:43,  6.70s/it]
                                                 
{'loss': '0.0001111', 'grad_norm': '0.01184', 'learning_rate': '1.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 12590736, 'tokens/trainable': 190306, 'epoch': '1.625'}
+
 81%|████████▏ | 416/512 [46:40<10:43,  6.70s/it][2026-08-18 14:51:43,726] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
+
 81%|████████▏ | 417/512 [46:49<11:28,  7.25s/it]
                                                 
{'loss': '0.000122', 'grad_norm': '0.01777', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.21', 'tokens/total': 12621168, 'tokens/trainable': 190746, 'epoch': '1.629'}
+
 81%|████████▏ | 417/512 [46:49<11:28,  7.25s/it]
 82%|████████▏ | 418/512 [46:55<11:04,  7.07s/it]
                                                 
{'loss': '5.223e-05', 'grad_norm': '0.003219', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '31.15', 'tokens/total': 12651456, 'tokens/trainable': 191156, 'epoch': '1.633'}
+
 82%|████████▏ | 418/512 [46:55<11:04,  7.07s/it]
 82%|████████▏ | 419/512 [47:02<10:45,  6.94s/it]
                                                 
{'loss': '0.0001365', 'grad_norm': '0.01374', 'learning_rate': '1.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 12681760, 'tokens/trainable': 191573, 'epoch': '1.637'}
+
 82%|████████▏ | 419/512 [47:02<10:45,  6.94s/it]
 82%|████████▏ | 420/512 [47:09<10:31,  6.86s/it]
                                                 
{'loss': '0.002095', 'grad_norm': '0.2501', 'learning_rate': '1.786e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 12712144, 'tokens/trainable': 192014, 'epoch': '1.641'}
+
 82%|████████▏ | 420/512 [47:09<10:31,  6.86s/it]
 82%|████████▏ | 421/512 [47:15<10:19,  6.80s/it]
                                                 
{'loss': '0.001427', 'grad_norm': '0.1429', 'learning_rate': '1.77e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 12742400, 'tokens/trainable': 192453, 'epoch': '1.645'}
+
 82%|████████▏ | 421/512 [47:15<10:19,  6.80s/it]
 82%|████████▏ | 422/512 [47:22<10:07,  6.75s/it]
                                                 
{'loss': '1.823e-05', 'grad_norm': '0.0009391', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 12772288, 'tokens/trainable': 192886, 'epoch': '1.648'}
+
 82%|████████▏ | 422/512 [47:22<10:07,  6.75s/it]
 83%|████████▎ | 423/512 [47:29<10:01,  6.76s/it]
                                                 
{'loss': '6.211e-05', 'grad_norm': '0.004544', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '29.87', 'tokens/total': 12802880, 'tokens/trainable': 193323, 'epoch': '1.652'}
+
 83%|████████▎ | 423/512 [47:29<10:01,  6.76s/it]
 83%|████████▎ | 424/512 [47:35<09:56,  6.78s/it]
                                                 
{'loss': '0.0005114', 'grad_norm': '0.03986', 'learning_rate': '1.722e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 12833456, 'tokens/trainable': 193753, 'epoch': '1.656'}
+
 83%|████████▎ | 424/512 [47:35<09:56,  6.78s/it]
 83%|████████▎ | 425/512 [47:42<09:46,  6.75s/it]
                                                 
{'loss': '7.249e-05', 'grad_norm': '0.004084', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.41', 'tokens/total': 12863952, 'tokens/trainable': 194240, 'epoch': '1.66'}
+
 83%|████████▎ | 425/512 [47:42<09:46,  6.75s/it]
 83%|████████▎ | 426/512 [47:49<09:38,  6.72s/it]
                                                 
{'loss': '0.001506', 'grad_norm': '0.1312', 'learning_rate': '1.69e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '35.54', 'tokens/total': 12894352, 'tokens/trainable': 194680, 'epoch': '1.664'}
+
 83%|████████▎ | 426/512 [47:49<09:38,  6.72s/it]
 83%|████████▎ | 427/512 [47:56<09:31,  6.72s/it]
                                                 
{'loss': '8.445e-05', 'grad_norm': '0.01044', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '38.77', 'tokens/total': 12924688, 'tokens/trainable': 195186, 'epoch': '1.668'}
+
 83%|████████▎ | 427/512 [47:56<09:31,  6.72s/it]
 84%|████████▎ | 428/512 [48:02<09:25,  6.73s/it]
                                                 
{'loss': '9.946e-05', 'grad_norm': '0.007595', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 12955072, 'tokens/trainable': 195693, 'epoch': '1.672'}
+
 84%|████████▎ | 428/512 [48:02<09:25,  6.73s/it]
 84%|████████▍ | 429/512 [48:09<09:16,  6.71s/it]
                                                 
{'loss': '0.0001356', 'grad_norm': '0.006466', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.95', 'tokens/total': 12985328, 'tokens/trainable': 196130, 'epoch': '1.676'}
+
 84%|████████▍ | 429/512 [48:09<09:16,  6.71s/it]
 84%|████████▍ | 430/512 [48:16<09:08,  6.69s/it]
                                                 
{'loss': '3.711e-05', 'grad_norm': '0.002304', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 13015600, 'tokens/trainable': 196530, 'epoch': '1.68'}
+
 84%|████████▍ | 430/512 [48:16<09:08,  6.69s/it]
 84%|████████▍ | 431/512 [48:22<09:00,  6.68s/it]
                                                 
{'loss': '6.117e-05', 'grad_norm': '0.00419', 'learning_rate': '1.615e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.94', 'tokens/total': 13045728, 'tokens/trainable': 196974, 'epoch': '1.684'}
+
 84%|████████▍ | 431/512 [48:22<09:00,  6.68s/it]
 84%|████████▍ | 432/512 [48:29<08:54,  6.68s/it]
                                                 
{'loss': '0.0004236', 'grad_norm': '0.02952', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 13075920, 'tokens/trainable': 197454, 'epoch': '1.688'}
+
 84%|████████▍ | 432/512 [48:29<08:54,  6.68s/it]
 85%|████████▍ | 433/512 [48:36<08:47,  6.68s/it]
                                                 
{'loss': '1.119e-05', 'grad_norm': '0.0003203', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '31.43', 'tokens/total': 13106128, 'tokens/trainable': 197902, 'epoch': '1.691'}
+
 85%|████████▍ | 433/512 [48:36<08:47,  6.68s/it]
 85%|████████▍ | 434/512 [48:42<08:40,  6.67s/it]
                                                 
{'loss': '2.119e-05', 'grad_norm': '0.0006868', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '36.06', 'tokens/total': 13136368, 'tokens/trainable': 198374, 'epoch': '1.695'}
+
 85%|████████▍ | 434/512 [48:42<08:40,  6.67s/it]
 85%|████████▍ | 435/512 [48:49<08:33,  6.67s/it]
                                                 
{'loss': '4.928e-05', 'grad_norm': '0.002541', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.59', 'tokens/total': 13166704, 'tokens/trainable': 198846, 'epoch': '1.699'}
+
 85%|████████▍ | 435/512 [48:49<08:33,  6.67s/it]
 85%|████████▌ | 436/512 [48:56<08:28,  6.69s/it]
                                                 
{'loss': '3.578e-05', 'grad_norm': '0.00139', 'learning_rate': '1.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 13197344, 'tokens/trainable': 199320, 'epoch': '1.703'}
+
 85%|████████▌ | 436/512 [48:56<08:28,  6.69s/it]
 85%|████████▌ | 437/512 [49:02<08:21,  6.69s/it]
                                                 
{'loss': '6.294e-05', 'grad_norm': '0.003646', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.46', 'tokens/total': 13227664, 'tokens/trainable': 199777, 'epoch': '1.707'}
+
 85%|████████▌ | 437/512 [49:02<08:21,  6.69s/it]
 86%|████████▌ | 438/512 [49:09<08:12,  6.66s/it]
                                                 
{'loss': '0.000177', 'grad_norm': '0.03043', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '30.93', 'tokens/total': 13257632, 'tokens/trainable': 200194, 'epoch': '1.711'}
+
 86%|████████▌ | 438/512 [49:09<08:12,  6.66s/it]
 86%|████████▌ | 439/512 [49:16<08:09,  6.70s/it]
                                                 
{'loss': '3.761e-05', 'grad_norm': '0.001823', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 13288240, 'tokens/trainable': 200663, 'epoch': '1.715'}
+
 86%|████████▌ | 439/512 [49:16<08:09,  6.70s/it]
 86%|████████▌ | 440/512 [49:23<08:04,  6.73s/it]
                                                 
{'loss': '4.779e-05', 'grad_norm': '0.002938', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 13318992, 'tokens/trainable': 201121, 'epoch': '1.719'}
+
 86%|████████▌ | 440/512 [49:23<08:04,  6.73s/it]
 86%|████████▌ | 441/512 [49:29<07:56,  6.71s/it]
                                                 
{'loss': '3.15e-05', 'grad_norm': '0.001837', 'learning_rate': '1.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.81', 'tokens/total': 13349392, 'tokens/trainable': 201598, 'epoch': '1.723'}
+
 86%|████████▌ | 441/512 [49:29<07:56,  6.71s/it]
 86%|████████▋ | 442/512 [49:36<07:49,  6.70s/it]
                                                 
{'loss': '2.081e-05', 'grad_norm': '0.001543', 'learning_rate': '1.464e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 13379648, 'tokens/trainable': 202068, 'epoch': '1.727'}
+
 86%|████████▋ | 442/512 [49:36<07:49,  6.70s/it]
 87%|████████▋ | 443/512 [49:43<07:41,  6.69s/it]
                                                 
{'loss': '2.304e-05', 'grad_norm': '0.001077', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.5', 'tokens/total': 13409936, 'tokens/trainable': 202524, 'epoch': '1.73'}
+
 87%|████████▋ | 443/512 [49:43<07:41,  6.69s/it]
 87%|████████▋ | 444/512 [49:49<07:34,  6.69s/it]
                                                 
{'loss': '7.449e-05', 'grad_norm': '0.006372', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 13440288, 'tokens/trainable': 202996, 'epoch': '1.734'}
+
 87%|████████▋ | 444/512 [49:49<07:34,  6.69s/it]
 87%|████████▋ | 445/512 [49:55<07:19,  6.56s/it]
                                                 
{'loss': '4.211e-05', 'grad_norm': '0.007095', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.19', 'tokens/total': 13468880, 'tokens/trainable': 203470, 'epoch': '1.738'}
+
 87%|████████▋ | 445/512 [49:55<07:19,  6.56s/it]
 87%|████████▋ | 446/512 [50:02<07:15,  6.60s/it]
                                                 
{'loss': '8.662e-05', 'grad_norm': '0.006931', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 13499184, 'tokens/trainable': 203919, 'epoch': '1.742'}
+
 87%|████████▋ | 446/512 [50:02<07:15,  6.60s/it]
 87%|████████▋ | 447/512 [50:09<07:10,  6.62s/it]
                                                 
{'loss': '4.787e-05', 'grad_norm': '0.004166', 'learning_rate': '1.402e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.24', 'tokens/total': 13529632, 'tokens/trainable': 204376, 'epoch': '1.746'}
+
 87%|████████▋ | 447/512 [50:09<07:10,  6.62s/it]
 88%|████████▊ | 448/512 [50:16<07:04,  6.64s/it]
                                                 
{'loss': '0.0002709', 'grad_norm': '0.04226', 'learning_rate': '1.39e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 13560080, 'tokens/trainable': 204821, 'epoch': '1.75'}
+
 88%|████████▊ | 448/512 [50:16<07:04,  6.64s/it][2026-08-18 14:55:19,147] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
+
 88%|████████▊ | 449/512 [50:24<07:34,  7.22s/it]
                                                 
{'loss': '1.284e-05', 'grad_norm': '0.0004374', 'learning_rate': '1.378e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 13590496, 'tokens/trainable': 205271, 'epoch': '1.754'}
+
 88%|████████▊ | 449/512 [50:24<07:34,  7.22s/it]
 88%|████████▊ | 450/512 [50:31<07:17,  7.05s/it]
                                                 
{'loss': '2.11e-05', 'grad_norm': '0.001298', 'learning_rate': '1.367e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '30.92', 'tokens/total': 13620960, 'tokens/trainable': 205723, 'epoch': '1.758'}
+
 88%|████████▊ | 450/512 [50:31<07:17,  7.05s/it]
 88%|████████▊ | 451/512 [50:37<07:03,  6.95s/it]
                                                 
{'loss': '7.466e-05', 'grad_norm': '0.005675', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.48', 'tokens/total': 13651344, 'tokens/trainable': 206169, 'epoch': '1.762'}
+
 88%|████████▊ | 451/512 [50:37<07:03,  6.95s/it]
 88%|████████▊ | 452/512 [50:44<06:53,  6.89s/it]
                                                 
{'loss': '2.723e-05', 'grad_norm': '0.0014', 'learning_rate': '1.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 13681792, 'tokens/trainable': 206592, 'epoch': '1.766'}
+
 88%|████████▊ | 452/512 [50:44<06:53,  6.89s/it]
 88%|████████▊ | 453/512 [50:51<06:42,  6.83s/it]
                                                 
{'loss': '9.183e-05', 'grad_norm': '0.01249', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 13712256, 'tokens/trainable': 207055, 'epoch': '1.77'}
+
 88%|████████▊ | 453/512 [50:51<06:42,  6.83s/it]
 89%|████████▊ | 454/512 [50:57<06:31,  6.76s/it]
                                                 
{'loss': '2.162e-05', 'grad_norm': '0.0009246', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.61', 'tokens/total': 13742128, 'tokens/trainable': 207494, 'epoch': '1.773'}
+
 89%|████████▊ | 454/512 [50:57<06:31,  6.76s/it]
 89%|████████▉ | 455/512 [51:04<06:22,  6.72s/it]
                                                 
{'loss': '1.243e-05', 'grad_norm': '0.0004114', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.03', 'tokens/total': 13772160, 'tokens/trainable': 207956, 'epoch': '1.777'}
+
 89%|████████▉ | 455/512 [51:04<06:22,  6.72s/it]
 89%|████████▉ | 456/512 [51:11<06:15,  6.71s/it]
                                                 
{'loss': '3.033e-05', 'grad_norm': '0.001591', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.42', 'tokens/total': 13802448, 'tokens/trainable': 208408, 'epoch': '1.781'}
+
 89%|████████▉ | 456/512 [51:11<06:15,  6.71s/it]
 89%|████████▉ | 457/512 [51:18<06:09,  6.71s/it]
                                                 
{'loss': '3.379e-05', 'grad_norm': '0.003002', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.92', 'tokens/total': 13832880, 'tokens/trainable': 208889, 'epoch': '1.785'}
+
 89%|████████▉ | 457/512 [51:18<06:09,  6.71s/it]
 89%|████████▉ | 458/512 [51:24<06:01,  6.69s/it]
                                                 
{'loss': '3.313e-05', 'grad_norm': '0.001643', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 13862880, 'tokens/trainable': 209336, 'epoch': '1.789'}
+
 89%|████████▉ | 458/512 [51:24<06:01,  6.69s/it]
 90%|████████▉ | 459/512 [51:31<05:54,  6.69s/it]
                                                 
{'loss': '1.784e-05', 'grad_norm': '0.0006427', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 13893440, 'tokens/trainable': 209797, 'epoch': '1.793'}
+
 90%|████████▉ | 459/512 [51:31<05:54,  6.69s/it]
 90%|████████▉ | 460/512 [51:38<05:47,  6.69s/it]
                                                 
{'loss': '3.284e-05', 'grad_norm': '0.001669', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.03', 'tokens/total': 13923824, 'tokens/trainable': 210240, 'epoch': '1.797'}
+
 90%|████████▉ | 460/512 [51:38<05:47,  6.69s/it]
 90%|█████████ | 461/512 [51:44<05:41,  6.69s/it]
                                                 
{'loss': '7.201e-05', 'grad_norm': '0.03455', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '34.9', 'tokens/total': 13954208, 'tokens/trainable': 210702, 'epoch': '1.801'}
+
 90%|█████████ | 461/512 [51:44<05:41,  6.69s/it]
 90%|█████████ | 462/512 [51:51<05:34,  6.70s/it]
                                                 
{'loss': '3.525e-05', 'grad_norm': '0.002149', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '31.44', 'tokens/total': 13984784, 'tokens/trainable': 211140, 'epoch': '1.805'}
+
 90%|█████████ | 462/512 [51:51<05:34,  6.70s/it]
 90%|█████████ | 463/512 [51:58<05:27,  6.69s/it]
                                                 
{'loss': '1.329e-05', 'grad_norm': '0.0007275', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.54', 'tokens/total': 14014896, 'tokens/trainable': 211575, 'epoch': '1.809'}
+
 90%|█████████ | 463/512 [51:58<05:27,  6.69s/it]
 91%|█████████ | 464/512 [52:04<05:21,  6.71s/it]
                                                 
{'loss': '0.0001454', 'grad_norm': '0.03174', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '38.86', 'tokens/total': 14045312, 'tokens/trainable': 212083, 'epoch': '1.812'}
+
 91%|█████████ | 464/512 [52:04<05:21,  6.71s/it]
 91%|█████████ | 465/512 [52:11<05:16,  6.73s/it]
                                                 
{'loss': '1.449e-05', 'grad_norm': '0.0005439', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 14075840, 'tokens/trainable': 212558, 'epoch': '1.816'}
+
 91%|█████████ | 465/512 [52:11<05:16,  6.73s/it]
 91%|█████████ | 466/512 [52:18<05:08,  6.72s/it]
                                                 
{'loss': '2.788e-05', 'grad_norm': '0.001548', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 14106016, 'tokens/trainable': 213014, 'epoch': '1.82'}
+
 91%|█████████ | 466/512 [52:18<05:08,  6.72s/it]
 91%|█████████ | 467/512 [52:25<05:02,  6.72s/it]
                                                 
{'loss': '2.111e-05', 'grad_norm': '0.0007428', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.99', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 14136512, 'tokens/trainable': 213465, 'epoch': '1.824'}
+
 91%|█████████ | 467/512 [52:25<05:02,  6.72s/it]
 91%|█████████▏| 468/512 [52:31<04:55,  6.73s/it]
                                                 
{'loss': '4.059e-05', 'grad_norm': '0.002183', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 14166832, 'tokens/trainable': 213940, 'epoch': '1.828'}
+
 91%|█████████▏| 468/512 [52:31<04:55,  6.73s/it]
 92%|█████████▏| 469/512 [52:38<04:49,  6.74s/it]
                                                 
{'loss': '3.665e-05', 'grad_norm': '0.004772', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.97', 'tokens/total': 14197360, 'tokens/trainable': 214415, 'epoch': '1.832'}
+
 92%|█████████▏| 469/512 [52:38<04:49,  6.74s/it]
 92%|█████████▏| 470/512 [52:45<04:43,  6.74s/it]
                                                 
{'loss': '1.112e-05', 'grad_norm': '0.0003869', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 14227600, 'tokens/trainable': 214880, 'epoch': '1.836'}
+
 92%|█████████▏| 470/512 [52:45<04:43,  6.74s/it]
 92%|█████████▏| 471/512 [52:52<04:35,  6.72s/it]
                                                 
{'loss': '2.091e-05', 'grad_norm': '0.000844', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.34', 'tokens/total': 14257888, 'tokens/trainable': 215334, 'epoch': '1.84'}
+
 92%|█████████▏| 471/512 [52:52<04:35,  6.72s/it]
 92%|█████████▏| 472/512 [52:58<04:29,  6.73s/it]
                                                 
{'loss': '1.179e-05', 'grad_norm': '0.0005927', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 14288208, 'tokens/trainable': 215822, 'epoch': '1.844'}
+
 92%|█████████▏| 472/512 [52:58<04:29,  6.73s/it]
 92%|█████████▏| 473/512 [53:05<04:22,  6.72s/it]
                                                 
{'loss': '5.021e-05', 'grad_norm': '0.003898', 'learning_rate': '1.149e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '30.82', 'tokens/total': 14318448, 'tokens/trainable': 216239, 'epoch': '1.848'}
+
 92%|█████████▏| 473/512 [53:05<04:22,  6.72s/it]
 93%|█████████▎| 474/512 [53:12<04:14,  6.70s/it]
                                                 
{'loss': '2.28e-05', 'grad_norm': '0.001381', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.35', 'tokens/total': 14348912, 'tokens/trainable': 216691, 'epoch': '1.852'}
+
 93%|█████████▎| 474/512 [53:12<04:14,  6.70s/it]
 93%|█████████▎| 475/512 [53:18<04:07,  6.70s/it]
                                                 
{'loss': '1.463e-05', 'grad_norm': '0.0004829', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 14379440, 'tokens/trainable': 217162, 'epoch': '1.855'}
+
 93%|█████████▎| 475/512 [53:18<04:07,  6.70s/it]
 93%|█████████▎| 476/512 [53:25<04:01,  6.71s/it]
                                                 
{'loss': '1.446e-05', 'grad_norm': '0.0004492', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.88', 'tokens/total': 14409840, 'tokens/trainable': 217638, 'epoch': '1.859'}
+
 93%|█████████▎| 476/512 [53:25<04:01,  6.71s/it]
 93%|█████████▎| 477/512 [53:32<03:54,  6.71s/it]
                                                 
{'loss': '4.934e-05', 'grad_norm': '0.002406', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 14440256, 'tokens/trainable': 218053, 'epoch': '1.863'}
+
 93%|█████████▎| 477/512 [53:32<03:54,  6.71s/it]
 93%|█████████▎| 478/512 [53:38<03:47,  6.70s/it]
                                                 
{'loss': '2.053e-05', 'grad_norm': '0.0008485', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 14470704, 'tokens/trainable': 218503, 'epoch': '1.867'}
+
 93%|█████████▎| 478/512 [53:38<03:47,  6.70s/it]
 94%|█████████▎| 479/512 [53:45<03:41,  6.70s/it]
                                                 
{'loss': '8.76e-05', 'grad_norm': '0.01195', 'learning_rate': '1.108e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.19', 'tokens/total': 14501072, 'tokens/trainable': 218964, 'epoch': '1.871'}
+
 94%|█████████▎| 479/512 [53:45<03:41,  6.70s/it]
 94%|█████████▍| 480/512 [53:52<03:33,  6.68s/it]
                                                 
{'loss': '7.079e-05', 'grad_norm': '0.004777', 'learning_rate': '1.102e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 14531200, 'tokens/trainable': 219412, 'epoch': '1.875'}
+
 94%|█████████▍| 480/512 [53:52<03:33,  6.68s/it][2026-08-18 14:58:55,402] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
+
 94%|█████████▍| 481/512 [54:00<03:45,  7.27s/it]
                                                 
{'loss': '1.676e-05', 'grad_norm': '0.0006404', 'learning_rate': '1.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '38.25', 'tokens/total': 14561152, 'tokens/trainable': 219886, 'epoch': '1.879'}
+
 94%|█████████▍| 481/512 [54:00<03:45,  7.27s/it]
 94%|█████████▍| 482/512 [54:07<03:32,  7.09s/it]
                                                 
{'loss': '2.334e-05', 'grad_norm': '0.005523', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '34.52', 'tokens/total': 14591264, 'tokens/trainable': 220343, 'epoch': '1.883'}
+
 94%|█████████▍| 482/512 [54:07<03:32,  7.09s/it]
 94%|█████████▍| 483/512 [54:14<03:22,  6.97s/it]
                                                 
{'loss': '2.361e-05', 'grad_norm': '0.001016', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 14621568, 'tokens/trainable': 220825, 'epoch': '1.887'}
+
 94%|█████████▍| 483/512 [54:14<03:22,  6.97s/it]
 95%|█████████▍| 484/512 [54:20<03:12,  6.89s/it]
                                                 
{'loss': '5.646e-05', 'grad_norm': '0.004034', 'learning_rate': '1.079e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 14652064, 'tokens/trainable': 221282, 'epoch': '1.891'}
+
 95%|█████████▍| 484/512 [54:20<03:12,  6.89s/it]
 95%|█████████▍| 485/512 [54:27<03:04,  6.84s/it]
                                                 
{'loss': '5.676e-05', 'grad_norm': '0.008658', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.89', 'tokens/total': 14682608, 'tokens/trainable': 221780, 'epoch': '1.895'}
+
 95%|█████████▍| 485/512 [54:27<03:04,  6.84s/it]
 95%|█████████▍| 486/512 [54:34<02:57,  6.81s/it]
                                                 
{'loss': '3.921e-05', 'grad_norm': '0.003652', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.56', 'tokens/total': 14713168, 'tokens/trainable': 222210, 'epoch': '1.898'}
+
 95%|█████████▍| 486/512 [54:34<02:57,  6.81s/it]
 95%|█████████▌| 487/512 [54:41<02:49,  6.77s/it]
                                                 
{'loss': '5.354e-05', 'grad_norm': '0.005618', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.37', 'tokens/total': 14743536, 'tokens/trainable': 222646, 'epoch': '1.902'}
+
 95%|█████████▌| 487/512 [54:41<02:49,  6.77s/it]
 95%|█████████▌| 488/512 [54:47<02:41,  6.74s/it]
                                                 
{'loss': '0.0006614', 'grad_norm': '0.07198', 'learning_rate': '1.058e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 14773728, 'tokens/trainable': 223066, 'epoch': '1.906'}
+
 95%|█████████▌| 488/512 [54:47<02:41,  6.74s/it]
 96%|█████████▌| 489/512 [54:54<02:34,  6.72s/it]
                                                 
{'loss': '1.627e-05', 'grad_norm': '0.00117', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 14803744, 'tokens/trainable': 223512, 'epoch': '1.91'}
+
 96%|█████████▌| 489/512 [54:54<02:34,  6.72s/it]
 96%|█████████▌| 490/512 [55:00<02:24,  6.56s/it]
                                                 
{'loss': '3.052e-05', 'grad_norm': '0.001455', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.95', 'tokens/total': 14832112, 'tokens/trainable': 223956, 'epoch': '1.914'}
+
 96%|█████████▌| 490/512 [55:00<02:24,  6.56s/it]
 96%|█████████▌| 491/512 [55:06<02:15,  6.45s/it]
                                                 
{'loss': '1.489e-05', 'grad_norm': '0.0007276', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 14860560, 'tokens/trainable': 224416, 'epoch': '1.918'}
+
 96%|█████████▌| 491/512 [55:06<02:15,  6.45s/it]
 96%|█████████▌| 492/512 [55:13<02:10,  6.52s/it]
                                                 
{'loss': '1.902e-05', 'grad_norm': '0.0007378', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 14890848, 'tokens/trainable': 224883, 'epoch': '1.922'}
+
 96%|█████████▌| 492/512 [55:13<02:10,  6.52s/it]
 96%|█████████▋| 493/512 [55:20<02:04,  6.56s/it]
                                                 
{'loss': '1.498e-05', 'grad_norm': '0.0006582', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.42', 'tokens/total': 14921008, 'tokens/trainable': 225324, 'epoch': '1.926'}
+
 96%|█████████▋| 493/512 [55:20<02:04,  6.56s/it]
 96%|█████████▋| 494/512 [55:26<01:58,  6.60s/it]
                                                 
{'loss': '1.008e-05', 'grad_norm': '0.0008491', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 14951360, 'tokens/trainable': 225783, 'epoch': '1.93'}
+
 96%|█████████▋| 494/512 [55:26<01:58,  6.60s/it]
 97%|█████████▋| 495/512 [55:33<01:52,  6.63s/it]
                                                 
{'loss': '2.822e-05', 'grad_norm': '0.002965', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.48', 'tokens/total': 14981552, 'tokens/trainable': 226281, 'epoch': '1.934'}
+
 97%|█████████▋| 495/512 [55:33<01:52,  6.63s/it]
 97%|█████████▋| 496/512 [55:40<01:45,  6.62s/it]
                                                 
{'loss': '2.592e-05', 'grad_norm': '0.00117', 'learning_rate': '1.027e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 15011584, 'tokens/trainable': 226735, 'epoch': '1.938'}
+
 97%|█████████▋| 496/512 [55:40<01:45,  6.62s/it]
 97%|█████████▋| 497/512 [55:46<01:39,  6.65s/it]
                                                 
{'loss': '9.912e-05', 'grad_norm': '0.006589', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 15042016, 'tokens/trainable': 227198, 'epoch': '1.941'}
+
 97%|█████████▋| 497/512 [55:46<01:39,  6.65s/it]
 97%|█████████▋| 498/512 [55:53<01:32,  6.63s/it]
                                                 
{'loss': '2.851e-05', 'grad_norm': '0.001317', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.07', 'tokens/total': 15071984, 'tokens/trainable': 227623, 'epoch': '1.945'}
+
 97%|█████████▋| 498/512 [55:53<01:32,  6.63s/it]
 97%|█████████▋| 499/512 [56:00<01:26,  6.65s/it]
                                                 
{'loss': '2.48e-05', 'grad_norm': '0.001088', 'learning_rate': '1.018e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '32.17', 'tokens/total': 15102192, 'tokens/trainable': 228074, 'epoch': '1.949'}
+
 97%|█████████▋| 499/512 [56:00<01:26,  6.65s/it]
 98%|█████████▊| 500/512 [56:06<01:19,  6.66s/it]
                                                 
{'loss': '4.626e-05', 'grad_norm': '0.01028', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.59', 'tokens/total': 15132560, 'tokens/trainable': 228541, 'epoch': '1.953'}
+
 98%|█████████▊| 500/512 [56:06<01:19,  6.66s/it]
 98%|█████████▊| 501/512 [56:13<01:13,  6.67s/it]
                                                 
{'loss': '1.94e-05', 'grad_norm': '0.001143', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.78', 'tokens/total': 15163152, 'tokens/trainable': 229031, 'epoch': '1.957'}
+
 98%|█████████▊| 501/512 [56:13<01:13,  6.67s/it]
 98%|█████████▊| 502/512 [56:20<01:06,  6.68s/it]
                                                 
{'loss': '2.073e-05', 'grad_norm': '0.001288', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.76', 'tokens/total': 15193456, 'tokens/trainable': 229512, 'epoch': '1.961'}
+
 98%|█████████▊| 502/512 [56:20<01:06,  6.68s/it]
 98%|█████████▊| 503/512 [56:26<01:00,  6.67s/it]
                                                 
{'loss': '0.0001616', 'grad_norm': '0.02232', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 15223680, 'tokens/trainable': 229989, 'epoch': '1.965'}
+
 98%|█████████▊| 503/512 [56:26<01:00,  6.67s/it]
 98%|█████████▊| 504/512 [56:33<00:53,  6.67s/it]
                                                 
{'loss': '9.707e-06', 'grad_norm': '0.0003185', 'learning_rate': '1.008e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 15254336, 'tokens/trainable': 230447, 'epoch': '1.969'}
+
 98%|█████████▊| 504/512 [56:33<00:53,  6.67s/it]
 99%|█████████▊| 505/512 [56:40<00:46,  6.67s/it]
                                                 
{'loss': '1.712e-05', 'grad_norm': '0.0005881', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.3', 'tokens/total': 15284496, 'tokens/trainable': 230879, 'epoch': '1.973'}
+
 99%|█████████▊| 505/512 [56:40<00:46,  6.67s/it]
 99%|█████████▉| 506/512 [56:46<00:40,  6.69s/it]
                                                 
{'loss': '1.641e-05', 'grad_norm': '0.00115', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.03', 'tokens/total': 15314976, 'tokens/trainable': 231348, 'epoch': '1.977'}
+
 99%|█████████▉| 506/512 [56:46<00:40,  6.69s/it]
 99%|█████████▉| 507/512 [56:53<00:33,  6.69s/it]
                                                 
{'loss': '2.183e-05', 'grad_norm': '0.0007461', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 15345280, 'tokens/trainable': 231819, 'epoch': '1.98'}
+
 99%|█████████▉| 507/512 [56:53<00:33,  6.69s/it]
 99%|█████████▉| 508/512 [57:00<00:26,  6.68s/it]
                                                 
{'loss': '4.144e-05', 'grad_norm': '0.003102', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 15375424, 'tokens/trainable': 232269, 'epoch': '1.984'}
+
 99%|█████████▉| 508/512 [57:00<00:26,  6.68s/it]
 99%|█████████▉| 509/512 [57:06<00:20,  6.67s/it]
                                                 
{'loss': '1.45e-05', 'grad_norm': '0.001036', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 15405872, 'tokens/trainable': 232716, 'epoch': '1.988'}
+
 99%|█████████▉| 509/512 [57:06<00:20,  6.67s/it]
100%|█████████▉| 510/512 [57:13<00:13,  6.68s/it]
                                                 
{'loss': '2.179e-05', 'grad_norm': '0.001914', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.04', 'tokens/total': 15436128, 'tokens/trainable': 233178, 'epoch': '1.992'}
+
100%|█████████▉| 510/512 [57:13<00:13,  6.68s/it]
100%|█████████▉| 511/512 [57:20<00:06,  6.68s/it]
                                                 
{'loss': '2.199e-05', 'grad_norm': '0.001745', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.93', 'tokens/total': 15466464, 'tokens/trainable': 233647, 'epoch': '1.996'}
+
100%|█████████▉| 511/512 [57:20<00:06,  6.68s/it]
100%|██████████| 512/512 [57:27<00:00,  6.70s/it]
                                                 
{'loss': '0.0005933', 'grad_norm': '0.06926', 'learning_rate': '1e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.4', 'tokens/total': 15496368, 'tokens/trainable': 234060, 'epoch': '2'}
+
100%|██████████| 512/512 [57:27<00:00,  6.70s/it][2026-08-18 15:02:30,225] [INFO] [axolotl.core.trainers.base._save:828] [PID:9922] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
+
                                                 
{'train_runtime': '3449', 'train_samples_per_second': '4.75', 'train_steps_per_second': '0.148', 'train_loss': '0.007172', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '38.44', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
+
100%|██████████| 512/512 [57:29<00:00,  6.70s/it]
100%|██████████| 512/512 [57:29<00:00,  6.74s/it]
+[2026-08-18 15:02:32,241] [INFO] [axolotl.train.save_trained_model:267] [PID:9922] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
+[2026-08-18 15:02:32,700] [INFO] [axolotl.train.save_trained_model:388] [PID:9922] Model successfully saved to /workspace/wave/training/checkpoints
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/health/training_started.json b/aft_wave_v2/coin_real_4x__agreement/training/health/training_started.json
index 096709d13618550fd48fe908f4713e06dfcbd327..b4003f19a544d2e0525e00d09b0394546d4959f8 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/health/training_started.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/health/training_started.json
@@ -1,6 +1,6 @@
 {
   "loss": 0.1025,
   "observed": "first_optimizer_loss",
-  "observed_at": "2026-08-18T12:43:58+00:00",
+  "observed_at": "2026-08-18T14:05:11+00:00",
   "status": "training_started"
 }
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/run.json b/aft_wave_v2/coin_real_4x__agreement/training/run.json
index cdc6dc51c272b3debd2d8bf7ba5543632443527d..08b0ab9941a799981d6f845f2c1ea555c219e841 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/run.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/run.json
@@ -1,9 +1,9 @@
 {
   "run_name": "coin_real_4x__agreement",
-  "git_commit": "dc47d38dd7732e27f42285faab76fb29f6840b74",
+  "git_commit": "e0c479b41f5718f428d5a199ba67c513d7fb9574",
   "git_dirty": false,
   "host": "3aa2e56fa12b",
-  "started_at": "2026-08-18T12:41:50+00:00",
+  "started_at": "2026-08-18T14:03:47+00:00",
   "configs": {
     "axolotl": "/workspace/wave/training/config/axolotl.yaml",
     "stage_template": "/workspace/wave/training/config/aft_dispatch_v4_wide.yaml"
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/train.log b/aft_wave_v2/coin_real_4x__agreement/training/train.log
index 490c9fd97c80344a7ee016d13b1f73a502b77c68..b9fb5b38df4d471b74f45e51699ade245ab03b2b 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/train.log
+++ b/aft_wave_v2/coin_real_4x__agreement/training/train.log
@@ -1,8 +1,8 @@
-[2026-08-18 12:41:53,037] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:03:49,938] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-W0818 12:41:55.060000 2842 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:41:55.084000 2842 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:03:51.792000 9658 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:03:51.814000 9658 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
 
      #@@ #@@      @@# @@#
     @@  @@          @@  @@           =@@#                               @@                 #@    =@@#.
@@ -22,17 +22,17 @@ The following values were not passed to `accelerate launch` and had defaults use
 	`--mixed_precision` was set to a value of `'no'`
 	`--dynamo_backend` was set to a value of `'no'`
 To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
-[2026-08-18 12:42:07,166] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:01,811] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-W0818 12:42:10.116000 3104 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:42:10.137000 3104 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-[2026-08-18 12:42:13,081] [INFO] [axolotl.integrations.base] Attempting to load plugin: axolotl.integrations.liger.LigerPlugin
-[2026-08-18 12:42:13,086] [INFO] [axolotl.integrations.base] Plugin loaded successfully: axolotl.integrations.liger.LigerPlugin
-[2026-08-18 12:42:13,145] [WARNING] [axolotl.utils.schemas.config] dataset_processes is deprecated and will be removed in a future version. Please use dataset_num_proc instead.
-[2026-08-18 12:42:13,145] [WARNING] [axolotl.utils.schemas.config] Auto-enabling LoRA kernel optimizations for faster training. Please explicitly set `lora_*_kernel` config values to `false` to disable. See https://docs.axolotl.ai/docs/lora_optims.html for more info.
-[2026-08-18 12:42:13,146] [WARNING] [axolotl.utils.schemas.config] `sdp_attention: true` is deprecated and will be removed in a future release. Use `attn_implementation: sdpa` instead.
-[2026-08-18 12:42:13,482] [INFO] [axolotl.cli.config] config:
+W0818 14:04:04.894000 9922 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:04.927000 9922 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+[2026-08-18 14:04:07,687] [INFO] [axolotl.integrations.base] Attempting to load plugin: axolotl.integrations.liger.LigerPlugin
+[2026-08-18 14:04:07,690] [INFO] [axolotl.integrations.base] Plugin loaded successfully: axolotl.integrations.liger.LigerPlugin
+[2026-08-18 14:04:07,743] [WARNING] [axolotl.utils.schemas.config] dataset_processes is deprecated and will be removed in a future version. Please use dataset_num_proc instead.
+[2026-08-18 14:04:07,743] [WARNING] [axolotl.utils.schemas.config] Auto-enabling LoRA kernel optimizations for faster training. Please explicitly set `lora_*_kernel` config values to `false` to disable. See https://docs.axolotl.ai/docs/lora_optims.html for more info.
+[2026-08-18 14:04:07,743] [WARNING] [axolotl.utils.schemas.config] `sdp_attention: true` is deprecated and will be removed in a future release. Use `attn_implementation: sdpa` instead.
+[2026-08-18 14:04:08,056] [INFO] [axolotl.cli.config] config:
 {
   "activation_offloading": false,
   "adapter": "lora",
@@ -232,12 +232,11 @@ W0818 12:42:10.137000 3104 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMo
   "weight_decay": 0.01,
   "world_size": 1
 }
-[2026-08-18 12:42:20,135] [INFO] [axolotl.utils.data.shared] Unable to find prepared dataset in /workspace/wave/training/prepared/6307088ba5f1ed42c0c9294b0bcc8eed
-[2026-08-18 12:42:20,135] [INFO] [axolotl.utils.data.sft] Loading raw datasets...
-[2026-08-18 12:42:20,135] [WARNING] [axolotl.utils.data.sft] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
-
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 8192 examples [00:00, 124620.94 examples/s]
-[2026-08-18 12:42:21,184] [INFO] [axolotl.utils.data.wrappers] Loading dataset: /workspace/wave/data/datasets/aft_agreement.jsonl with base_type: chat_template and prompt_style: None
-[2026-08-18 12:42:21,200] [INFO] [axolotl.prompt_strategies.chat_template] Using chat template:
+[2026-08-18 14:04:11,623] [INFO] [axolotl.utils.data.shared] Unable to find prepared dataset in /workspace/wave/training/prepared/6307088ba5f1ed42c0c9294b0bcc8eed
+[2026-08-18 14:04:11,623] [INFO] [axolotl.utils.data.sft] Loading raw datasets...
+[2026-08-18 14:04:11,623] [WARNING] [axolotl.utils.data.sft] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
+[2026-08-18 14:04:12,435] [INFO] [axolotl.utils.data.wrappers] Loading dataset: /workspace/wave/data/datasets/aft_agreement.jsonl with base_type: chat_template and prompt_style: None
+[2026-08-18 14:04:12,438] [INFO] [axolotl.prompt_strategies.chat_template] Using chat template:
 ---
 {{ bos_token }}
 {%- if messages[0]['role'] == 'system' -%}
@@ -288,598 +287,596 @@ W0818 12:42:10.137000 3104 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMo
 {%- endif -%}
 
 ---
-
Tokenizing Prompts (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▏        | 1000/8192 [00:07<00:55, 130.41 examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▎        | 1024/8192 [00:07<00:54, 132.24 examples/s]
Tokenizing Prompts (num_proc=8):  25%|██▍       | 2024/8192 [00:11<00:31, 195.21 examples/s]
Tokenizing Prompts (num_proc=8):  25%|██▌       | 2048/8192 [00:11<00:31, 195.91 examples/s]
Tokenizing Prompts (num_proc=8):  37%|███▋      | 3048/8192 [00:15<00:21, 237.52 examples/s]
Tokenizing Prompts (num_proc=8):  50%|████▉     | 4072/8192 [00:18<00:16, 257.13 examples/s]
Tokenizing Prompts (num_proc=8):  62%|██████▏   | 5096/8192 [00:22<00:11, 267.72 examples/s]
Tokenizing Prompts (num_proc=8):  75%|███████▍  | 6120/8192 [00:26<00:07, 259.63 examples/s]
Tokenizing Prompts (num_proc=8):  87%|████████▋ | 7144/8192 [00:29<00:03, 274.56 examples/s]
Tokenizing Prompts (num_proc=8): 100%|█████████▉| 8168/8192 [00:33<00:00, 277.05 examples/s]
Tokenizing Prompts (num_proc=8): 100%|██████████| 8192/8192 [00:33<00:00, 241.43 examples/s]
-[2026-08-18 12:43:10,544] [INFO] [axolotl.utils.data.utils] min_input_len: 446
-[2026-08-18 12:43:10,545] [INFO] [axolotl.utils.data.utils] max_input_len: 967
-
Dropping Invalid Sequences (<None or >1280) (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8):  12%|█▎        | 1024/8192 [00:00<00:01, 5433.57 examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8): 100%|██████████| 8192/8192 [00:00<00:00, 22279.27 examples/s]
-
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s][2026-08-18 12:43:13,349] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:27,935] [INFO] [axolotl.utils.data.utils] min_input_len: 446
+[2026-08-18 14:04:27,936] [INFO] [axolotl.utils.data.utils] max_input_len: 967
+
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s][2026-08-18 14:04:30,357] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,349] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,357] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,349] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,416] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,349] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,416] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,357] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,416] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,363] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,416] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,365] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,416] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,368] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,417] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:43:13,372] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:30,421] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-W0818 12:43:16.322000 3294 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.328000 3291 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.336000 3297 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.343000 3294 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.349000 3291 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.351000 3295 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.353000 3301 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.352000 3292 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.356000 3297 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.371000 3295 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.373000 3301 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.384000 3292 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.420000 3290 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.442000 3290 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.457000 3293 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.458000 3296 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.479000 3296 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:43:16.480000 3293 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:07<00:55, 129.38 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:07<00:55, 129.38 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:07<00:47, 129.38 examples/s]
Saving the dataset (3/8 shards):  38%|███▊      | 3072/8192 [00:07<00:39, 129.38 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:07<00:31, 129.38 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:07<00:23, 129.38 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:07<00:15, 129.38 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:07<00:07, 129.38 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:07<00:00, 129.38 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 907.33 examples/s]
-[2026-08-18 12:43:20,221] [INFO] [axolotl.utils.data.sft] Maximum number of steps set at 512
-[2026-08-18 12:43:31,493] [WARNING] [axolotl.loaders.patch_manager] Cannot patch self-attention - requires no dropout
-[2026-08-18 12:43:33,399] [INFO] [axolotl.integrations.liger.plugin] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
-
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  49%|████▊     | 518/1066 [00:00<00:00, 5152.10it/s]
Loading weights:  97%|█████████▋| 1034/1066 [00:00<00:00, 4754.96it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 4797.53it/s]
-[2026-08-18 12:43:37,048] [INFO] [axolotl.loaders.model] Converting modules to torch.bfloat16
+W0818 14:04:33.240000 10002 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.244000 10000 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.248000 9999 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.261000 10002 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.265000 10000 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.269000 9999 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.437000 10009 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.471000 10009 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.501000 10005 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.513000 10004 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.523000 10005 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.535000 10004 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.586000 10001 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.595000 9998 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.609000 10003 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.615000 10001 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.625000 9998 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:33.633000 10003 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:08<00:57, 124.55 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:08<00:57, 124.55 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:08<00:49, 124.55 examples/s]
Saving the dataset (3/8 shards):  38%|███▊      | 3072/8192 [00:08<00:41, 124.55 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:08<00:32, 124.55 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:08<00:24, 124.55 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:08<00:16, 124.55 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:08<00:08, 124.55 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:08<00:00, 124.55 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 871.58 examples/s]
+[2026-08-18 14:04:37,581] [INFO] [axolotl.utils.data.sft] Maximum number of steps set at 512
+[2026-08-18 14:04:48,430] [WARNING] [axolotl.loaders.patch_manager] Cannot patch self-attention - requires no dropout
+[2026-08-18 14:04:49,649] [INFO] [axolotl.integrations.liger.plugin] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
+
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  53%|█████▎    | 564/1066 [00:00<00:00, 5633.46it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 5757.61it/s]
+[2026-08-18 14:04:53,143] [INFO] [axolotl.loaders.model] Converting modules to torch.bfloat16
 trainable params: 136,912,896 || all params: 12,324,237,936 || trainable%: 1.1109
-[2026-08-18 12:43:41,482] [INFO] [axolotl.monkeypatch.lora_kernels] LoRA kernels: dropout=0.05 enabled
-[2026-08-18 12:43:46,529] [INFO] [axolotl.train] Pre-saving adapter config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:46,529] [INFO] [axolotl.train] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:46,958] [INFO] [axolotl.train] Pre-saving model config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:46,962] [INFO] [axolotl.train] Pre-saving processor to /workspace/wave/training/checkpoints...
-[2026-08-18 12:43:47,360] [INFO] [axolotl.train] Starting trainer...
+[2026-08-18 14:04:57,312] [INFO] [axolotl.monkeypatch.lora_kernels] LoRA kernels: dropout=0.05 enabled
+[2026-08-18 14:05:01,950] [INFO] [axolotl.train] Pre-saving adapter config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:01,950] [INFO] [axolotl.train] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:02,286] [INFO] [axolotl.train] Pre-saving model config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:02,290] [INFO] [axolotl.train] Pre-saving processor to /workspace/wave/training/checkpoints...
+[2026-08-18 14:05:02,604] [INFO] [axolotl.train] Starting trainer...
 
  0%|          | 0/512 [00:00<?, ?it/s][transformers] `use_return_dict` is deprecated! Use `return_dict` instead!
-
  0%|          | 1/512 [00:10<1:26:45, 10.19s/it]
                                                 
{'loss': '0.1025', 'grad_norm': '0.8707', 'learning_rate': '0', 'ppl': '1.108', 'memory/max_active (GiB)': '32.98', 'memory/max_allocated (GiB)': '32.98', 'memory/device_reserved (GiB)': '34.94', 'tokens/train_per_sec_per_gpu': '23.98', 'tokens/total': 30464, 'tokens/trainable': 470, 'epoch': '0.003906'}
-
  0%|          | 1/512 [00:10<1:26:45, 10.19s/it]
  0%|          | 2/512 [00:16<1:08:51,  8.10s/it]
                                                 
{'loss': '0.09678', 'grad_norm': '0.8108', 'learning_rate': '4e-06', 'ppl': '1.102', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.83', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 60800, 'tokens/trainable': 922, 'epoch': '0.007812'}
-
  0%|          | 2/512 [00:16<1:08:51,  8.10s/it]
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
                                                 
{'loss': '0.1095', 'grad_norm': '1.803', 'learning_rate': '8e-06', 'ppl': '1.116', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '34.48', 'tokens/total': 91136, 'tokens/trainable': 1396, 'epoch': '0.01172'}
-
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
  1%|          | 4/512 [00:30<1:00:40,  7.17s/it]
                                                 
{'loss': '0.103', 'grad_norm': '1.035', 'learning_rate': '1.2e-05', 'ppl': '1.109', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '38.01', 'tokens/total': 121552, 'tokens/trainable': 1850, 'epoch': '0.01562'}
-
  1%|          | 4/512 [00:30<1:00:40,  7.17s/it]
  1%|          | 5/512 [00:37<1:00:12,  7.12s/it]
                                                 
{'loss': '0.1095', 'grad_norm': '1.078', 'learning_rate': '1.6e-05', 'ppl': '1.116', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 152304, 'tokens/trainable': 2302, 'epoch': '0.01953'}
-
  1%|          | 5/512 [00:37<1:00:12,  7.12s/it]
  1%|          | 6/512 [00:43<58:53,  6.98s/it]  
                                               
{'loss': '0.08589', 'grad_norm': '0.893', 'learning_rate': '2e-05', 'ppl': '1.09', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '31.47', 'tokens/total': 182448, 'tokens/trainable': 2742, 'epoch': '0.02344'}
-
  1%|          | 6/512 [00:43<58:53,  6.98s/it]
  1%|▏         | 7/512 [00:50<58:03,  6.90s/it]
                                               
{'loss': '0.07353', 'grad_norm': '1.641', 'learning_rate': '2.4e-05', 'ppl': '1.076', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 212736, 'tokens/trainable': 3208, 'epoch': '0.02734'}
-
  1%|▏         | 7/512 [00:50<58:03,  6.90s/it]
  2%|▏         | 8/512 [00:57<57:23,  6.83s/it]
                                               
{'loss': '0.07798', 'grad_norm': '1.584', 'learning_rate': '2.8e-05', 'ppl': '1.081', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 243024, 'tokens/trainable': 3655, 'epoch': '0.03125'}
-
  2%|▏         | 8/512 [00:57<57:23,  6.83s/it]
  2%|▏         | 9/512 [01:03<55:38,  6.64s/it]
                                               
{'loss': '0.04634', 'grad_norm': '1.373', 'learning_rate': '3.2e-05', 'ppl': '1.047', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '35.62', 'tokens/total': 271264, 'tokens/trainable': 4091, 'epoch': '0.03516'}
-
  2%|▏         | 9/512 [01:03<55:38,  6.64s/it]
  2%|▏         | 10/512 [01:10<55:46,  6.67s/it]
                                                
{'loss': '0.08678', 'grad_norm': '2.545', 'learning_rate': '3.6e-05', 'ppl': '1.091', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 301520, 'tokens/trainable': 4553, 'epoch': '0.03906'}
-
  2%|▏         | 10/512 [01:10<55:46,  6.67s/it]
  2%|▏         | 11/512 [01:17<55:44,  6.68s/it]
                                                
{'loss': '0.07754', 'grad_norm': '1.636', 'learning_rate': '4e-05', 'ppl': '1.081', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 331808, 'tokens/trainable': 4992, 'epoch': '0.04297'}
-
  2%|▏         | 11/512 [01:17<55:44,  6.68s/it]
  2%|▏         | 12/512 [01:23<55:54,  6.71s/it]
                                                
{'loss': '0.1177', 'grad_norm': '2.167', 'learning_rate': '4.4e-05', 'ppl': '1.125', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '38.18', 'tokens/total': 362224, 'tokens/trainable': 5494, 'epoch': '0.04688'}
-
  2%|▏         | 12/512 [01:23<55:54,  6.71s/it]
  3%|▎         | 13/512 [01:30<56:11,  6.76s/it]
                                                
{'loss': '0.03511', 'grad_norm': '3.197', 'learning_rate': '4.8e-05', 'ppl': '1.036', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 392432, 'tokens/trainable': 5933, 'epoch': '0.05078'}
-
  3%|▎         | 13/512 [01:30<56:11,  6.76s/it]
  3%|▎         | 14/512 [01:37<56:02,  6.75s/it]
                                                
{'loss': '0.0611', 'grad_norm': '1.965', 'learning_rate': '5.2e-05', 'ppl': '1.063', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 422784, 'tokens/trainable': 6369, 'epoch': '0.05469'}
-
  3%|▎         | 14/512 [01:37<56:02,  6.75s/it]
  3%|▎         | 15/512 [01:44<55:44,  6.73s/it]
                                                
{'loss': '0.05385', 'grad_norm': '1.934', 'learning_rate': '5.6e-05', 'ppl': '1.055', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 453376, 'tokens/trainable': 6820, 'epoch': '0.05859'}
-
  3%|▎         | 15/512 [01:44<55:44,  6.73s/it]
  3%|▎         | 16/512 [01:50<55:30,  6.72s/it]
                                                
{'loss': '0.05215', 'grad_norm': '1.466', 'learning_rate': '6e-05', 'ppl': '1.054', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.68', 'tokens/total': 483504, 'tokens/trainable': 7258, 'epoch': '0.0625'}
-
  3%|▎         | 16/512 [01:50<55:30,  6.72s/it]
  3%|▎         | 17/512 [01:57<55:34,  6.74s/it]
                                                
{'loss': '0.05092', 'grad_norm': '1.965', 'learning_rate': '6.4e-05', 'ppl': '1.052', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '35.09', 'tokens/total': 514032, 'tokens/trainable': 7710, 'epoch': '0.06641'}
-
  3%|▎         | 17/512 [01:57<55:34,  6.74s/it]
  4%|▎         | 18/512 [02:04<55:28,  6.74s/it]
                                                
{'loss': '0.03116', 'grad_norm': '0.6892', 'learning_rate': '6.8e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '31.61', 'tokens/total': 544432, 'tokens/trainable': 8174, 'epoch': '0.07031'}
-
  4%|▎         | 18/512 [02:04<55:28,  6.74s/it]
  4%|▎         | 19/512 [02:11<55:15,  6.72s/it]
                                                
{'loss': '0.03036', 'grad_norm': '0.8662', 'learning_rate': '7.2e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '34.37', 'tokens/total': 574880, 'tokens/trainable': 8617, 'epoch': '0.07422'}
-
  4%|▎         | 19/512 [02:11<55:15,  6.72s/it]
  4%|▍         | 20/512 [02:17<54:59,  6.71s/it]
                                                
{'loss': '0.03459', 'grad_norm': '0.7999', 'learning_rate': '7.6e-05', 'ppl': '1.035', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.32', 'tokens/total': 605408, 'tokens/trainable': 9037, 'epoch': '0.07812'}
-
  4%|▍         | 20/512 [02:17<54:59,  6.71s/it]
  4%|▍         | 21/512 [02:24<54:51,  6.70s/it]
                                                
{'loss': '0.02401', 'grad_norm': '0.5817', 'learning_rate': '8e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 635584, 'tokens/trainable': 9503, 'epoch': '0.08203'}
-
  4%|▍         | 21/512 [02:24<54:51,  6.70s/it]
  4%|▍         | 22/512 [02:31<54:48,  6.71s/it]
                                                
{'loss': '0.01873', 'grad_norm': '0.6762', 'learning_rate': '8.4e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.71', 'tokens/total': 665952, 'tokens/trainable': 9972, 'epoch': '0.08594'}
-
  4%|▍         | 22/512 [02:31<54:48,  6.71s/it]
  4%|▍         | 23/512 [02:37<54:50,  6.73s/it]
                                                
{'loss': '0.01749', 'grad_norm': '0.9078', 'learning_rate': '8.8e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '37.39', 'tokens/total': 696240, 'tokens/trainable': 10447, 'epoch': '0.08984'}
-
  4%|▍         | 23/512 [02:37<54:50,  6.73s/it]
  5%|▍         | 24/512 [02:44<54:32,  6.71s/it]
                                                
{'loss': '0.02842', 'grad_norm': '1.323', 'learning_rate': '9.2e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '37.12', 'tokens/total': 726464, 'tokens/trainable': 10899, 'epoch': '0.09375'}
-
  5%|▍         | 24/512 [02:44<54:32,  6.71s/it]
  5%|▍         | 25/512 [02:51<54:30,  6.72s/it]
                                                
{'loss': '0.02597', 'grad_norm': '0.9712', 'learning_rate': '9.6e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 756704, 'tokens/trainable': 11360, 'epoch': '0.09766'}
-
  5%|▍         | 25/512 [02:51<54:30,  6.72s/it]
  5%|▌         | 26/512 [02:57<54:22,  6.71s/it]
                                                
{'loss': '0.02243', 'grad_norm': '0.8639', 'learning_rate': '0.0001', 'ppl': '1.023', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '29.23', 'tokens/total': 786912, 'tokens/trainable': 11775, 'epoch': '0.1016'}
-
  5%|▌         | 26/512 [02:57<54:22,  6.71s/it]
  5%|▌         | 27/512 [03:04<53:01,  6.56s/it]
                                                
{'loss': '0.01454', 'grad_norm': '0.6629', 'learning_rate': '0.0001', 'ppl': '1.015', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.82', 'tokens/total': 815424, 'tokens/trainable': 12242, 'epoch': '0.1055'}
-
  5%|▌         | 27/512 [03:04<53:01,  6.56s/it]
  5%|▌         | 28/512 [03:10<53:09,  6.59s/it]
                                                
{'loss': '0.01195', 'grad_norm': '0.3078', 'learning_rate': '0.0001', 'ppl': '1.012', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.52', 'tokens/total': 845584, 'tokens/trainable': 12696, 'epoch': '0.1094'}
-
  5%|▌         | 28/512 [03:10<53:09,  6.59s/it]
  6%|▌         | 29/512 [03:17<53:17,  6.62s/it]
                                                
{'loss': '0.02254', 'grad_norm': '1.13', 'learning_rate': '9.999e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 875808, 'tokens/trainable': 13153, 'epoch': '0.1133'}
-
  6%|▌         | 29/512 [03:17<53:17,  6.62s/it]
  6%|▌         | 30/512 [03:23<52:17,  6.51s/it]
                                                
{'loss': '0.01187', 'grad_norm': '0.4109', 'learning_rate': '9.999e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.83', 'tokens/total': 904096, 'tokens/trainable': 13624, 'epoch': '0.1172'}
-
  6%|▌         | 30/512 [03:23<52:17,  6.51s/it]
  6%|▌         | 31/512 [03:30<52:42,  6.57s/it]
                                                
{'loss': '0.0137', 'grad_norm': '0.6773', 'learning_rate': '9.998e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.07', 'tokens/total': 934400, 'tokens/trainable': 14064, 'epoch': '0.1211'}
-
  6%|▌         | 31/512 [03:30<52:42,  6.57s/it]
  6%|▋         | 32/512 [03:37<53:00,  6.63s/it]
                                                
{'loss': '0.01186', 'grad_norm': '1.208', 'learning_rate': '9.997e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.9', 'tokens/total': 964832, 'tokens/trainable': 14565, 'epoch': '0.125'}
-
  6%|▋         | 32/512 [03:37<53:00,  6.63s/it][2026-08-18 12:47:25,280] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
-
  6%|▋         | 33/512 [03:46<58:17,  7.30s/it]
                                                
{'loss': '0.01248', 'grad_norm': '0.4651', 'learning_rate': '9.995e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 995040, 'tokens/trainable': 15005, 'epoch': '0.1289'}
-
  6%|▋         | 33/512 [03:46<58:17,  7.30s/it]
  7%|▋         | 34/512 [03:52<56:31,  7.10s/it]
                                                
{'loss': '0.03285', 'grad_norm': '1.767', 'learning_rate': '9.994e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 1024896, 'tokens/trainable': 15459, 'epoch': '0.1328'}
-
  7%|▋         | 34/512 [03:52<56:31,  7.10s/it]
  7%|▋         | 35/512 [03:59<55:26,  6.97s/it]
                                                
{'loss': '0.02875', 'grad_norm': '0.7816', 'learning_rate': '9.992e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '38.15', 'tokens/total': 1054992, 'tokens/trainable': 15960, 'epoch': '0.1367'}
-
  7%|▋         | 35/512 [03:59<55:26,  6.97s/it]
  7%|▋         | 36/512 [04:06<54:34,  6.88s/it]
                                                
{'loss': '0.02006', 'grad_norm': '0.684', 'learning_rate': '9.991e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 1085280, 'tokens/trainable': 16428, 'epoch': '0.1406'}
-
  7%|▋         | 36/512 [04:06<54:34,  6.88s/it]
  7%|▋         | 37/512 [04:12<53:54,  6.81s/it]
                                                
{'loss': '0.01045', 'grad_norm': '0.6798', 'learning_rate': '9.989e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '31.77', 'tokens/total': 1115504, 'tokens/trainable': 16884, 'epoch': '0.1445'}
-
  7%|▋         | 37/512 [04:12<53:54,  6.81s/it]
  7%|▋         | 38/512 [04:19<53:34,  6.78s/it]
                                                
{'loss': '0.01739', 'grad_norm': '0.8899', 'learning_rate': '9.987e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 1145712, 'tokens/trainable': 17341, 'epoch': '0.1484'}
-
  7%|▋         | 38/512 [04:19<53:34,  6.78s/it]
  8%|▊         | 39/512 [04:26<53:12,  6.75s/it]
                                                
{'loss': '0.02252', 'grad_norm': '0.8133', 'learning_rate': '9.984e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 1176128, 'tokens/trainable': 17786, 'epoch': '0.1523'}
-
  8%|▊         | 39/512 [04:26<53:12,  6.75s/it]
  8%|▊         | 40/512 [04:32<53:05,  6.75s/it]
                                                
{'loss': '0.01197', 'grad_norm': '0.4431', 'learning_rate': '9.982e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.99', 'tokens/total': 1206576, 'tokens/trainable': 18262, 'epoch': '0.1562'}
-
  8%|▊         | 40/512 [04:32<53:05,  6.75s/it]
  8%|▊         | 41/512 [04:39<52:56,  6.74s/it]
                                                
{'loss': '0.01488', 'grad_norm': '0.5075', 'learning_rate': '9.979e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 1236848, 'tokens/trainable': 18716, 'epoch': '0.1602'}
-
  8%|▊         | 41/512 [04:39<52:56,  6.74s/it]
  8%|▊         | 42/512 [04:46<52:42,  6.73s/it]
                                                
{'loss': '0.01751', 'grad_norm': '0.5221', 'learning_rate': '9.976e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 1267088, 'tokens/trainable': 19157, 'epoch': '0.1641'}
-
  8%|▊         | 42/512 [04:46<52:42,  6.73s/it]
  8%|▊         | 43/512 [04:53<52:37,  6.73s/it]
                                                
{'loss': '0.03043', 'grad_norm': '0.6889', 'learning_rate': '9.973e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '36.5', 'tokens/total': 1297360, 'tokens/trainable': 19647, 'epoch': '0.168'}
-
  8%|▊         | 43/512 [04:53<52:37,  6.73s/it]
  9%|▊         | 44/512 [04:59<52:27,  6.72s/it]
                                                
{'loss': '0.01372', 'grad_norm': '0.56', 'learning_rate': '9.97e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '34.9', 'tokens/total': 1327744, 'tokens/trainable': 20119, 'epoch': '0.1719'}
-
  9%|▊         | 44/512 [04:59<52:27,  6.72s/it]
  9%|▉         | 45/512 [05:06<52:14,  6.71s/it]
                                                
{'loss': '0.008726', 'grad_norm': '0.4292', 'learning_rate': '9.966e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 1358096, 'tokens/trainable': 20560, 'epoch': '0.1758'}
-
  9%|▉         | 45/512 [05:06<52:14,  6.71s/it]
  9%|▉         | 46/512 [05:13<52:00,  6.70s/it]
                                                
{'loss': '0.02525', 'grad_norm': '0.9443', 'learning_rate': '9.963e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '32.59', 'tokens/total': 1388240, 'tokens/trainable': 20992, 'epoch': '0.1797'}
-
  9%|▉         | 46/512 [05:13<52:00,  6.70s/it]
  9%|▉         | 47/512 [05:19<50:39,  6.54s/it]
                                                
{'loss': '0.005067', 'grad_norm': '0.2425', 'learning_rate': '9.959e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.26', 'memory/max_allocated (GiB)': '33.26', 'memory/device_reserved (GiB)': '35.87', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 1416240, 'tokens/trainable': 21441, 'epoch': '0.1836'}
-
  9%|▉         | 47/512 [05:19<50:39,  6.54s/it]
  9%|▉         | 48/512 [05:25<50:52,  6.58s/it]
                                                
{'loss': '0.033', 'grad_norm': '1.236', 'learning_rate': '9.955e-05', 'ppl': '1.034', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.7', 'tokens/total': 1446880, 'tokens/trainable': 21901, 'epoch': '0.1875'}
-
  9%|▉         | 48/512 [05:25<50:52,  6.58s/it]
 10%|▉         | 49/512 [05:32<50:59,  6.61s/it]
                                                
{'loss': '0.02598', 'grad_norm': '0.7278', 'learning_rate': '9.951e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.27', 'tokens/total': 1477168, 'tokens/trainable': 22341, 'epoch': '0.1914'}
-
 10%|▉         | 49/512 [05:32<50:59,  6.61s/it]
 10%|▉         | 50/512 [05:39<51:07,  6.64s/it]
                                                
{'loss': '0.01032', 'grad_norm': '0.3724', 'learning_rate': '9.946e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 1507712, 'tokens/trainable': 22833, 'epoch': '0.1953'}
-
 10%|▉         | 50/512 [05:39<51:07,  6.64s/it]
 10%|▉         | 51/512 [05:46<51:08,  6.66s/it]
                                                
{'loss': '0.004857', 'grad_norm': '0.2526', 'learning_rate': '9.942e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.64', 'tokens/total': 1537936, 'tokens/trainable': 23277, 'epoch': '0.1992'}
-
 10%|▉         | 51/512 [05:46<51:08,  6.66s/it]
 10%|█         | 52/512 [05:52<51:10,  6.68s/it]
                                                
{'loss': '0.01646', 'grad_norm': '0.7386', 'learning_rate': '9.937e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 1568352, 'tokens/trainable': 23759, 'epoch': '0.2031'}
-
 10%|█         | 52/512 [05:52<51:10,  6.68s/it]
 10%|█         | 53/512 [05:59<51:11,  6.69s/it]
                                                
{'loss': '0.0192', 'grad_norm': '1.273', 'learning_rate': '9.932e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.98', 'tokens/total': 1598992, 'tokens/trainable': 24193, 'epoch': '0.207'}
-
 10%|█         | 53/512 [05:59<51:11,  6.69s/it]
 11%|█         | 54/512 [06:06<51:01,  6.68s/it]
                                                
{'loss': '0.006891', 'grad_norm': '0.5316', 'learning_rate': '9.927e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.47', 'tokens/total': 1629488, 'tokens/trainable': 24619, 'epoch': '0.2109'}
-
 11%|█         | 54/512 [06:06<51:01,  6.68s/it]
 11%|█         | 55/512 [06:12<50:58,  6.69s/it]
                                                
{'loss': '0.02118', 'grad_norm': '0.6924', 'learning_rate': '9.921e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 1659696, 'tokens/trainable': 25075, 'epoch': '0.2148'}
-
 11%|█         | 55/512 [06:12<50:58,  6.69s/it]
 11%|█         | 56/512 [06:19<50:49,  6.69s/it]
                                                
{'loss': '0.01759', 'grad_norm': '0.6836', 'learning_rate': '9.916e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 1689872, 'tokens/trainable': 25519, 'epoch': '0.2188'}
-
 11%|█         | 56/512 [06:19<50:49,  6.69s/it]
 11%|█         | 57/512 [06:26<50:46,  6.69s/it]
                                                
{'loss': '0.02556', 'grad_norm': '0.8593', 'learning_rate': '9.91e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 1720144, 'tokens/trainable': 25966, 'epoch': '0.2227'}
-
 11%|█         | 57/512 [06:26<50:46,  6.69s/it]
 11%|█▏        | 58/512 [06:32<50:40,  6.70s/it]
                                                
{'loss': '0.00579', 'grad_norm': '0.2926', 'learning_rate': '9.904e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 1750448, 'tokens/trainable': 26423, 'epoch': '0.2266'}
-
 11%|█▏        | 58/512 [06:32<50:40,  6.70s/it]
 12%|█▏        | 59/512 [06:39<50:43,  6.72s/it]
                                                
{'loss': '0.007335', 'grad_norm': '0.6059', 'learning_rate': '9.898e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.42', 'tokens/total': 1781088, 'tokens/trainable': 26934, 'epoch': '0.2305'}
-
 12%|█▏        | 59/512 [06:39<50:43,  6.72s/it]
 12%|█▏        | 60/512 [06:46<50:34,  6.71s/it]
                                                
{'loss': '0.00462', 'grad_norm': '0.2943', 'learning_rate': '9.892e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 1811344, 'tokens/trainable': 27393, 'epoch': '0.2344'}
-
 12%|█▏        | 60/512 [06:46<50:34,  6.71s/it]
 12%|█▏        | 61/512 [06:53<50:25,  6.71s/it]
                                                
{'loss': '0.004282', 'grad_norm': '0.2587', 'learning_rate': '9.886e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.84', 'tokens/total': 1841600, 'tokens/trainable': 27852, 'epoch': '0.2383'}
-
 12%|█▏        | 61/512 [06:53<50:25,  6.71s/it]
 12%|█▏        | 62/512 [06:59<50:15,  6.70s/it]
                                                
{'loss': '0.01351', 'grad_norm': '0.9455', 'learning_rate': '9.879e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 1872048, 'tokens/trainable': 28292, 'epoch': '0.2422'}
-
 12%|█▏        | 62/512 [06:59<50:15,  6.70s/it]
 12%|█▏        | 63/512 [07:06<50:17,  6.72s/it]
                                                
{'loss': '0.02086', 'grad_norm': '1.149', 'learning_rate': '9.872e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '37.27', 'tokens/total': 1902592, 'tokens/trainable': 28798, 'epoch': '0.2461'}
-
 12%|█▏        | 63/512 [07:06<50:17,  6.72s/it]
 12%|█▎        | 64/512 [07:13<50:09,  6.72s/it]
                                                
{'loss': '0.01604', 'grad_norm': '0.7422', 'learning_rate': '9.865e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.76', 'tokens/total': 1933088, 'tokens/trainable': 29283, 'epoch': '0.25'}
-
 12%|█▎        | 64/512 [07:13<50:09,  6.72s/it][2026-08-18 12:51:01,299] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
-
 13%|█▎        | 65/512 [07:21<54:06,  7.26s/it]
                                                
{'loss': '0.005373', 'grad_norm': '0.3611', 'learning_rate': '9.858e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.92', 'tokens/total': 1963296, 'tokens/trainable': 29745, 'epoch': '0.2539'}
-
 13%|█▎        | 65/512 [07:21<54:06,  7.26s/it]
 13%|█▎        | 66/512 [07:28<52:53,  7.11s/it]
                                                
{'loss': '0.01849', 'grad_norm': '0.7074', 'learning_rate': '9.851e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '34.88', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 1993760, 'tokens/trainable': 30186, 'epoch': '0.2578'}
-
 13%|█▎        | 66/512 [07:28<52:53,  7.11s/it]
 13%|█▎        | 67/512 [07:35<51:48,  6.99s/it]
                                                
{'loss': '0.006208', 'grad_norm': '0.4311', 'learning_rate': '9.844e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 2024144, 'tokens/trainable': 30622, 'epoch': '0.2617'}
-
 13%|█▎        | 67/512 [07:35<51:48,  6.99s/it]
 13%|█▎        | 68/512 [07:41<51:07,  6.91s/it]
                                                
{'loss': '0.00543', 'grad_norm': '0.3996', 'learning_rate': '9.836e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 2054608, 'tokens/trainable': 31097, 'epoch': '0.2656'}
-
 13%|█▎        | 68/512 [07:41<51:07,  6.91s/it]
 13%|█▎        | 69/512 [07:48<50:34,  6.85s/it]
                                                
{'loss': '0.01158', 'grad_norm': '0.4444', 'learning_rate': '9.828e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 2084912, 'tokens/trainable': 31595, 'epoch': '0.2695'}
-
 13%|█▎        | 69/512 [07:48<50:34,  6.85s/it]
 14%|█▎        | 70/512 [07:55<50:09,  6.81s/it]
                                                
{'loss': '0.002159', 'grad_norm': '0.1698', 'learning_rate': '9.82e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '30.67', 'tokens/total': 2115216, 'tokens/trainable': 32036, 'epoch': '0.2734'}
-
 14%|█▎        | 70/512 [07:55<50:09,  6.81s/it]
 14%|█▍        | 71/512 [08:02<49:44,  6.77s/it]
                                                
{'loss': '0.002899', 'grad_norm': '0.1297', 'learning_rate': '9.812e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.06', 'tokens/total': 2145424, 'tokens/trainable': 32428, 'epoch': '0.2773'}
-
 14%|█▍        | 71/512 [08:02<49:44,  6.77s/it]
 14%|█▍        | 72/512 [08:08<49:29,  6.75s/it]
                                                
{'loss': '0.0008802', 'grad_norm': '0.0649', 'learning_rate': '9.803e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.08', 'tokens/total': 2175648, 'tokens/trainable': 32880, 'epoch': '0.2812'}
-
 14%|█▍        | 72/512 [08:08<49:29,  6.75s/it]
 14%|█▍        | 73/512 [08:15<49:14,  6.73s/it]
                                                
{'loss': '0.00195', 'grad_norm': '0.2068', 'learning_rate': '9.795e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.48', 'tokens/total': 2205856, 'tokens/trainable': 33323, 'epoch': '0.2852'}
-
 14%|█▍        | 73/512 [08:15<49:14,  6.73s/it]
 14%|█▍        | 74/512 [08:22<49:00,  6.71s/it]
                                                
{'loss': '0.02436', 'grad_norm': '1.684', 'learning_rate': '9.786e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 2236352, 'tokens/trainable': 33792, 'epoch': '0.2891'}
-
 14%|█▍        | 74/512 [08:22<49:00,  6.71s/it]
 15%|█▍        | 75/512 [08:28<48:48,  6.70s/it]
                                                
{'loss': '0.01827', 'grad_norm': '0.6665', 'learning_rate': '9.777e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 2266480, 'tokens/trainable': 34223, 'epoch': '0.293'}
-
 15%|█▍        | 75/512 [08:28<48:48,  6.70s/it]
 15%|█▍        | 76/512 [08:35<48:36,  6.69s/it]
                                                
{'loss': '0.02372', 'grad_norm': '0.6469', 'learning_rate': '9.768e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.59', 'tokens/total': 2296496, 'tokens/trainable': 34649, 'epoch': '0.2969'}
-
 15%|█▍        | 76/512 [08:35<48:36,  6.69s/it]
 15%|█▌        | 77/512 [08:42<48:26,  6.68s/it]
                                                
{'loss': '0.0152', 'grad_norm': '1.392', 'learning_rate': '9.759e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 2327040, 'tokens/trainable': 35094, 'epoch': '0.3008'}
-
 15%|█▌        | 77/512 [08:42<48:26,  6.68s/it]
 15%|█▌        | 78/512 [08:48<48:18,  6.68s/it]
                                                
{'loss': '0.02025', 'grad_norm': '0.5753', 'learning_rate': '9.749e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 2357328, 'tokens/trainable': 35534, 'epoch': '0.3047'}
-
 15%|█▌        | 78/512 [08:48<48:18,  6.68s/it]
 15%|█▌        | 79/512 [08:55<48:16,  6.69s/it]
                                                
{'loss': '0.01047', 'grad_norm': '0.276', 'learning_rate': '9.74e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.76', 'tokens/total': 2387664, 'tokens/trainable': 36029, 'epoch': '0.3086'}
-
 15%|█▌        | 79/512 [08:55<48:16,  6.69s/it]
 16%|█▌        | 80/512 [09:02<48:03,  6.67s/it]
                                                
{'loss': '0.01077', 'grad_norm': '0.4178', 'learning_rate': '9.73e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 2418000, 'tokens/trainable': 36469, 'epoch': '0.3125'}
-
 16%|█▌        | 80/512 [09:02<48:03,  6.67s/it]
 16%|█▌        | 81/512 [09:08<48:05,  6.70s/it]
                                                
{'loss': '0.01542', 'grad_norm': '0.4906', 'learning_rate': '9.72e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.69', 'tokens/total': 2448720, 'tokens/trainable': 36906, 'epoch': '0.3164'}
-
 16%|█▌        | 81/512 [09:08<48:05,  6.70s/it]
 16%|█▌        | 82/512 [09:15<47:58,  6.69s/it]
                                                
{'loss': '0.009146', 'grad_norm': '0.2814', 'learning_rate': '9.71e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.89', 'tokens/total': 2479248, 'tokens/trainable': 37390, 'epoch': '0.3203'}
-
 16%|█▌        | 82/512 [09:15<47:58,  6.69s/it]
 16%|█▌        | 83/512 [09:22<47:51,  6.69s/it]
                                                
{'loss': '0.008894', 'grad_norm': '0.395', 'learning_rate': '9.699e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.78', 'tokens/total': 2509408, 'tokens/trainable': 37838, 'epoch': '0.3242'}
-
 16%|█▌        | 83/512 [09:22<47:51,  6.69s/it]
 16%|█▋        | 84/512 [09:28<47:44,  6.69s/it]
                                                
{'loss': '0.005293', 'grad_norm': '0.4946', 'learning_rate': '9.689e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.98', 'tokens/total': 2539776, 'tokens/trainable': 38310, 'epoch': '0.3281'}
-
 16%|█▋        | 84/512 [09:28<47:44,  6.69s/it]
 17%|█▋        | 85/512 [09:35<47:35,  6.69s/it]
                                                
{'loss': '0.04036', 'grad_norm': '1.329', 'learning_rate': '9.678e-05', 'ppl': '1.041', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 2569840, 'tokens/trainable': 38789, 'epoch': '0.332'}
-
 17%|█▋        | 85/512 [09:35<47:35,  6.69s/it]
 17%|█▋        | 86/512 [09:42<47:32,  6.70s/it]
                                                
{'loss': '0.01013', 'grad_norm': '0.5834', 'learning_rate': '9.667e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 2600192, 'tokens/trainable': 39277, 'epoch': '0.3359'}
-
 17%|█▋        | 86/512 [09:42<47:32,  6.70s/it]
 17%|█▋        | 87/512 [09:49<47:28,  6.70s/it]
                                                
{'loss': '0.02043', 'grad_norm': '0.7192', 'learning_rate': '9.656e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 2630640, 'tokens/trainable': 39737, 'epoch': '0.3398'}
-
 17%|█▋        | 87/512 [09:49<47:28,  6.70s/it]
 17%|█▋        | 88/512 [09:55<47:18,  6.69s/it]
                                                
{'loss': '0.0078', 'grad_norm': '0.3419', 'learning_rate': '9.645e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 2660832, 'tokens/trainable': 40179, 'epoch': '0.3438'}
-
 17%|█▋        | 88/512 [09:55<47:18,  6.69s/it]
 17%|█▋        | 89/512 [10:02<47:15,  6.70s/it]
                                                
{'loss': '0.00566', 'grad_norm': '0.2328', 'learning_rate': '9.633e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.59', 'tokens/total': 2691328, 'tokens/trainable': 40649, 'epoch': '0.3477'}
-
 17%|█▋        | 89/512 [10:02<47:15,  6.70s/it]
 18%|█▊        | 90/512 [10:08<46:03,  6.55s/it]
                                                
{'loss': '0.02008', 'grad_norm': '0.6988', 'learning_rate': '9.622e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 2719696, 'tokens/trainable': 41080, 'epoch': '0.3516'}
-
 18%|█▊        | 90/512 [10:08<46:03,  6.55s/it]
 18%|█▊        | 91/512 [10:15<46:24,  6.61s/it]
                                                
{'loss': '0.002235', 'grad_norm': '0.1133', 'learning_rate': '9.61e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.99', 'tokens/total': 2750272, 'tokens/trainable': 41599, 'epoch': '0.3555'}
-
 18%|█▊        | 91/512 [10:15<46:24,  6.61s/it]
 18%|█▊        | 92/512 [10:22<46:27,  6.64s/it]
                                                
{'loss': '0.004549', 'grad_norm': '0.4348', 'learning_rate': '9.598e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.77', 'tokens/total': 2780656, 'tokens/trainable': 42076, 'epoch': '0.3594'}
-
 18%|█▊        | 92/512 [10:22<46:27,  6.64s/it]
 18%|█▊        | 93/512 [10:28<46:28,  6.66s/it]
                                                
{'loss': '0.01227', 'grad_norm': '0.3487', 'learning_rate': '9.586e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.6', 'tokens/total': 2811088, 'tokens/trainable': 42522, 'epoch': '0.3633'}
-
 18%|█▊        | 93/512 [10:28<46:28,  6.66s/it]
 18%|█▊        | 94/512 [10:35<46:28,  6.67s/it]
                                                
{'loss': '0.01553', 'grad_norm': '0.5157', 'learning_rate': '9.574e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 2841520, 'tokens/trainable': 42974, 'epoch': '0.3672'}
-
 18%|█▊        | 94/512 [10:35<46:28,  6.67s/it]
 19%|█▊        | 95/512 [10:42<46:10,  6.64s/it]
                                                
{'loss': '0.01171', 'grad_norm': '0.3765', 'learning_rate': '9.562e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '30.71', 'tokens/total': 2871600, 'tokens/trainable': 43380, 'epoch': '0.3711'}
-
 19%|█▊        | 95/512 [10:42<46:10,  6.64s/it]
 19%|█▉        | 96/512 [10:48<46:15,  6.67s/it]
                                                
{'loss': '0.008411', 'grad_norm': '0.3837', 'learning_rate': '9.549e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 2902144, 'tokens/trainable': 43865, 'epoch': '0.375'}
-
 19%|█▉        | 96/512 [10:48<46:15,  6.67s/it][2026-08-18 12:54:36,892] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
-
 19%|█▉        | 97/512 [10:57<50:07,  7.25s/it]
                                                
{'loss': '0.007853', 'grad_norm': '0.3131', 'learning_rate': '9.536e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '31.89', 'tokens/total': 2932272, 'tokens/trainable': 44328, 'epoch': '0.3789'}
-
 19%|█▉        | 97/512 [10:57<50:07,  7.25s/it]
 19%|█▉        | 98/512 [11:04<48:38,  7.05s/it]
                                                
{'loss': '0.002575', 'grad_norm': '0.08544', 'learning_rate': '9.523e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '30.98', 'tokens/total': 2962032, 'tokens/trainable': 44778, 'epoch': '0.3828'}
-
 19%|█▉        | 98/512 [11:04<48:38,  7.05s/it]
 19%|█▉        | 99/512 [11:10<47:42,  6.93s/it]
                                                
{'loss': '0.02547', 'grad_norm': '0.6496', 'learning_rate': '9.51e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.11', 'tokens/total': 2992256, 'tokens/trainable': 45225, 'epoch': '0.3867'}
-
 19%|█▉        | 99/512 [11:10<47:42,  6.93s/it]
 20%|█▉        | 100/512 [11:17<47:04,  6.86s/it]
                                                 
{'loss': '0.003553', 'grad_norm': '0.1533', 'learning_rate': '9.497e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.8', 'tokens/total': 3022608, 'tokens/trainable': 45678, 'epoch': '0.3906'}
-
 20%|█▉        | 100/512 [11:17<47:04,  6.86s/it]
 20%|█▉        | 101/512 [11:24<46:33,  6.80s/it]
                                                 
{'loss': '0.004096', 'grad_norm': '0.2531', 'learning_rate': '9.484e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 3052992, 'tokens/trainable': 46125, 'epoch': '0.3945'}
-
 20%|█▉        | 101/512 [11:24<46:33,  6.80s/it]
 20%|█▉        | 102/512 [11:30<46:21,  6.78s/it]
                                                 
{'loss': '0.001389', 'grad_norm': '0.1501', 'learning_rate': '9.47e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '39.71', 'tokens/total': 3083392, 'tokens/trainable': 46612, 'epoch': '0.3984'}
-
 20%|█▉        | 102/512 [11:30<46:21,  6.78s/it]
 20%|██        | 103/512 [11:37<46:07,  6.77s/it]
                                                 
{'loss': '0.02253', 'grad_norm': '0.577', 'learning_rate': '9.456e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '32.57', 'tokens/total': 3113744, 'tokens/trainable': 47059, 'epoch': '0.4023'}
-
 20%|██        | 103/512 [11:37<46:07,  6.77s/it]
 20%|██        | 104/512 [11:44<46:10,  6.79s/it]
                                                 
{'loss': '0.01689', 'grad_norm': '0.5657', 'learning_rate': '9.442e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 3144272, 'tokens/trainable': 47536, 'epoch': '0.4062'}
-
 20%|██        | 104/512 [11:44<46:10,  6.79s/it]
 21%|██        | 105/512 [11:51<45:54,  6.77s/it]
                                                 
{'loss': '0.01797', 'grad_norm': '0.2961', 'learning_rate': '9.428e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 3174512, 'tokens/trainable': 48037, 'epoch': '0.4102'}
-
 21%|██        | 105/512 [11:51<45:54,  6.77s/it]
 21%|██        | 106/512 [11:57<45:36,  6.74s/it]
                                                 
{'loss': '0.01619', 'grad_norm': '0.5242', 'learning_rate': '9.414e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 3204896, 'tokens/trainable': 48491, 'epoch': '0.4141'}
-
 21%|██        | 106/512 [11:57<45:36,  6.74s/it]
 21%|██        | 107/512 [12:04<45:32,  6.75s/it]
                                                 
{'loss': '0.01091', 'grad_norm': '0.3886', 'learning_rate': '9.4e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 3235360, 'tokens/trainable': 48940, 'epoch': '0.418'}
-
 21%|██        | 107/512 [12:04<45:32,  6.75s/it]
 21%|██        | 108/512 [12:11<45:16,  6.72s/it]
                                                 
{'loss': '0.008071', 'grad_norm': '0.2443', 'learning_rate': '9.385e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 3265552, 'tokens/trainable': 49389, 'epoch': '0.4219'}
-
 21%|██        | 108/512 [12:11<45:16,  6.72s/it]
 21%|██▏       | 109/512 [12:17<43:58,  6.55s/it]
                                                 
{'loss': '0.003786', 'grad_norm': '0.1457', 'learning_rate': '9.37e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '36.01', 'tokens/total': 3293856, 'tokens/trainable': 49838, 'epoch': '0.4258'}
-
 21%|██▏       | 109/512 [12:17<43:58,  6.55s/it]
 21%|██▏       | 110/512 [12:24<44:10,  6.59s/it]
                                                 
{'loss': '0.006992', 'grad_norm': '0.3956', 'learning_rate': '9.355e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.61', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 3323984, 'tokens/trainable': 50310, 'epoch': '0.4297'}
-
 21%|██▏       | 110/512 [12:24<44:10,  6.59s/it]
 22%|██▏       | 111/512 [12:30<44:11,  6.61s/it]
                                                 
{'loss': '0.01485', 'grad_norm': '0.6225', 'learning_rate': '9.34e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 3354320, 'tokens/trainable': 50755, 'epoch': '0.4336'}
-
 22%|██▏       | 111/512 [12:30<44:11,  6.61s/it]
 22%|██▏       | 112/512 [12:37<44:15,  6.64s/it]
                                                 
{'loss': '0.004368', 'grad_norm': '0.3701', 'learning_rate': '9.325e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.65', 'tokens/total': 3384880, 'tokens/trainable': 51213, 'epoch': '0.4375'}
-
 22%|██▏       | 112/512 [12:37<44:15,  6.64s/it]
 22%|██▏       | 113/512 [12:44<44:06,  6.63s/it]
                                                 
{'loss': '0.002598', 'grad_norm': '0.1353', 'learning_rate': '9.31e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.83', 'tokens/total': 3415104, 'tokens/trainable': 51660, 'epoch': '0.4414'}
-
 22%|██▏       | 113/512 [12:44<44:06,  6.63s/it]
 22%|██▏       | 114/512 [12:50<44:07,  6.65s/it]
                                                 
{'loss': '0.002497', 'grad_norm': '0.1898', 'learning_rate': '9.294e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.43', 'tokens/total': 3445504, 'tokens/trainable': 52133, 'epoch': '0.4453'}
-
 22%|██▏       | 114/512 [12:50<44:07,  6.65s/it]
 22%|██▏       | 115/512 [12:57<44:06,  6.67s/it]
                                                 
{'loss': '0.02949', 'grad_norm': '1.182', 'learning_rate': '9.278e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.47', 'tokens/total': 3475744, 'tokens/trainable': 52580, 'epoch': '0.4492'}
-
 22%|██▏       | 115/512 [12:57<44:06,  6.67s/it]
 23%|██▎       | 116/512 [13:04<44:06,  6.68s/it]
                                                 
{'loss': '0.009586', 'grad_norm': '0.4436', 'learning_rate': '9.263e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.83', 'tokens/total': 3506016, 'tokens/trainable': 53033, 'epoch': '0.4531'}
-
 23%|██▎       | 116/512 [13:04<44:06,  6.68s/it]
 23%|██▎       | 117/512 [13:10<44:04,  6.69s/it]
                                                 
{'loss': '0.0202', 'grad_norm': '0.5075', 'learning_rate': '9.247e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 3536256, 'tokens/trainable': 53517, 'epoch': '0.457'}
-
 23%|██▎       | 117/512 [13:10<44:04,  6.69s/it]
 23%|██▎       | 118/512 [13:17<43:49,  6.67s/it]
                                                 
{'loss': '0.005689', 'grad_norm': '0.3565', 'learning_rate': '9.23e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 3566480, 'tokens/trainable': 53967, 'epoch': '0.4609'}
-
 23%|██▎       | 118/512 [13:17<43:49,  6.67s/it]
 23%|██▎       | 119/512 [13:24<43:45,  6.68s/it]
                                                 
{'loss': '0.003189', 'grad_norm': '0.1655', 'learning_rate': '9.214e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 3596624, 'tokens/trainable': 54430, 'epoch': '0.4648'}
-
 23%|██▎       | 119/512 [13:24<43:45,  6.68s/it]
 23%|██▎       | 120/512 [13:30<43:41,  6.69s/it]
                                                 
{'loss': '0.003235', 'grad_norm': '0.2072', 'learning_rate': '9.198e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 3627168, 'tokens/trainable': 54896, 'epoch': '0.4688'}
-
 23%|██▎       | 120/512 [13:30<43:41,  6.69s/it]
 24%|██▎       | 121/512 [13:37<43:39,  6.70s/it]
                                                 
{'loss': '0.005671', 'grad_norm': '0.4755', 'learning_rate': '9.181e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.77', 'tokens/total': 3657632, 'tokens/trainable': 55379, 'epoch': '0.4727'}
-
 24%|██▎       | 121/512 [13:37<43:39,  6.70s/it]
 24%|██▍       | 122/512 [13:44<43:24,  6.68s/it]
                                                 
{'loss': '0.0009903', 'grad_norm': '0.08305', 'learning_rate': '9.164e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.7', 'tokens/total': 3687824, 'tokens/trainable': 55803, 'epoch': '0.4766'}
-
 24%|██▍       | 122/512 [13:44<43:24,  6.68s/it]
 24%|██▍       | 123/512 [13:50<43:16,  6.68s/it]
                                                 
{'loss': '0.007413', 'grad_norm': '0.7253', 'learning_rate': '9.147e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '36.45', 'tokens/total': 3718320, 'tokens/trainable': 56249, 'epoch': '0.4805'}
-
 24%|██▍       | 123/512 [13:50<43:16,  6.68s/it]
 24%|██▍       | 124/512 [13:57<43:16,  6.69s/it]
                                                 
{'loss': '0.00297', 'grad_norm': '0.3141', 'learning_rate': '9.13e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.41', 'tokens/total': 3748672, 'tokens/trainable': 56713, 'epoch': '0.4844'}
-
 24%|██▍       | 124/512 [13:57<43:16,  6.69s/it]
 24%|██▍       | 125/512 [14:04<43:14,  6.70s/it]
                                                 
{'loss': '0.003796', 'grad_norm': '0.6083', 'learning_rate': '9.113e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '39.08', 'tokens/total': 3778976, 'tokens/trainable': 57196, 'epoch': '0.4883'}
-
 24%|██▍       | 125/512 [14:04<43:14,  6.70s/it]
 25%|██▍       | 126/512 [14:11<43:08,  6.71s/it]
                                                 
{'loss': '0.001475', 'grad_norm': '0.0604', 'learning_rate': '9.096e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '36.49', 'tokens/total': 3809296, 'tokens/trainable': 57680, 'epoch': '0.4922'}
-
 25%|██▍       | 126/512 [14:11<43:08,  6.71s/it]
 25%|██▍       | 127/512 [14:17<42:54,  6.69s/it]
                                                 
{'loss': '0.004137', 'grad_norm': '0.2111', 'learning_rate': '9.078e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.8', 'tokens/total': 3839328, 'tokens/trainable': 58110, 'epoch': '0.4961'}
-
 25%|██▍       | 127/512 [14:17<42:54,  6.69s/it]
 25%|██▌       | 128/512 [14:24<42:49,  6.69s/it]
                                                 
{'loss': '0.007372', 'grad_norm': '0.3234', 'learning_rate': '9.06e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 3869824, 'tokens/trainable': 58534, 'epoch': '0.5'}
-
 25%|██▌       | 128/512 [14:24<42:49,  6.69s/it][2026-08-18 12:58:12,429] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
-
 25%|██▌       | 129/512 [14:33<46:35,  7.30s/it]
                                                 
{'loss': '0.00798', 'grad_norm': '0.5826', 'learning_rate': '9.043e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 3900080, 'tokens/trainable': 58998, 'epoch': '0.5039'}
-
 25%|██▌       | 129/512 [14:33<46:35,  7.30s/it]
 25%|██▌       | 130/512 [14:39<45:19,  7.12s/it]
                                                 
{'loss': '0.005638', 'grad_norm': '0.3385', 'learning_rate': '9.025e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.81', 'tokens/total': 3930560, 'tokens/trainable': 59448, 'epoch': '0.5078'}
-
 25%|██▌       | 130/512 [14:39<45:19,  7.12s/it]
 26%|██▌       | 131/512 [14:46<44:16,  6.97s/it]
                                                 
{'loss': '0.003793', 'grad_norm': '0.2839', 'learning_rate': '9.007e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.97', 'tokens/total': 3960496, 'tokens/trainable': 59906, 'epoch': '0.5117'}
-
 26%|██▌       | 131/512 [14:46<44:16,  6.97s/it]
 26%|██▌       | 132/512 [14:53<43:37,  6.89s/it]
                                                 
{'loss': '0.01023', 'grad_norm': '0.4598', 'learning_rate': '8.988e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 3990928, 'tokens/trainable': 60350, 'epoch': '0.5156'}
-
 26%|██▌       | 132/512 [14:53<43:37,  6.89s/it]
 26%|██▌       | 133/512 [14:59<43:08,  6.83s/it]
                                                 
{'loss': '0.009468', 'grad_norm': '0.7073', 'learning_rate': '8.97e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 4021264, 'tokens/trainable': 60831, 'epoch': '0.5195'}
-
 26%|██▌       | 133/512 [14:59<43:08,  6.83s/it]
 26%|██▌       | 134/512 [15:06<42:46,  6.79s/it]
                                                 
{'loss': '0.006774', 'grad_norm': '0.3642', 'learning_rate': '8.951e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 4051728, 'tokens/trainable': 61305, 'epoch': '0.5234'}
-
 26%|██▌       | 134/512 [15:06<42:46,  6.79s/it]
 26%|██▋       | 135/512 [15:13<42:33,  6.77s/it]
                                                 
{'loss': '0.002285', 'grad_norm': '0.1071', 'learning_rate': '8.933e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.57', 'tokens/total': 4081920, 'tokens/trainable': 61792, 'epoch': '0.5273'}
-
 26%|██▋       | 135/512 [15:13<42:33,  6.77s/it]
 27%|██▋       | 136/512 [15:20<42:22,  6.76s/it]
                                                 
{'loss': '0.001668', 'grad_norm': '0.2247', 'learning_rate': '8.914e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.33', 'tokens/total': 4112304, 'tokens/trainable': 62253, 'epoch': '0.5312'}
-
 27%|██▋       | 136/512 [15:20<42:22,  6.76s/it]
 27%|██▋       | 137/512 [15:26<42:10,  6.75s/it]
                                                 
{'loss': '0.01744', 'grad_norm': '0.5423', 'learning_rate': '8.895e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 4142512, 'tokens/trainable': 62707, 'epoch': '0.5352'}
-
 27%|██▋       | 137/512 [15:26<42:10,  6.75s/it]
 27%|██▋       | 138/512 [15:33<41:53,  6.72s/it]
                                                 
{'loss': '0.01526', 'grad_norm': '0.4749', 'learning_rate': '8.876e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '29.32', 'tokens/total': 4172992, 'tokens/trainable': 63145, 'epoch': '0.5391'}
-
 27%|██▋       | 138/512 [15:33<41:53,  6.72s/it]
 27%|██▋       | 139/512 [15:40<41:44,  6.72s/it]
                                                 
{'loss': '0.00106', 'grad_norm': '0.05535', 'learning_rate': '8.856e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 4203136, 'tokens/trainable': 63587, 'epoch': '0.543'}
-
 27%|██▋       | 139/512 [15:40<41:44,  6.72s/it]
 27%|██▋       | 140/512 [15:46<41:35,  6.71s/it]
                                                 
{'loss': '0.002773', 'grad_norm': '0.1527', 'learning_rate': '8.837e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '36.75', 'tokens/total': 4233456, 'tokens/trainable': 64052, 'epoch': '0.5469'}
-
 27%|██▋       | 140/512 [15:46<41:35,  6.71s/it]
 28%|██▊       | 141/512 [15:53<41:27,  6.71s/it]
                                                 
{'loss': '0.00517', 'grad_norm': '0.2169', 'learning_rate': '8.817e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '34.36', 'tokens/total': 4263728, 'tokens/trainable': 64526, 'epoch': '0.5508'}
-
 28%|██▊       | 141/512 [15:53<41:27,  6.71s/it]
 28%|██▊       | 142/512 [16:00<41:32,  6.74s/it]
                                                 
{'loss': '0.003176', 'grad_norm': '0.1491', 'learning_rate': '8.798e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '34.66', 'tokens/total': 4294432, 'tokens/trainable': 64983, 'epoch': '0.5547'}
-
 28%|██▊       | 142/512 [16:00<41:32,  6.74s/it]
 28%|██▊       | 143/512 [16:07<41:31,  6.75s/it]
                                                 
{'loss': '0.003301', 'grad_norm': '0.1937', 'learning_rate': '8.778e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '37.6', 'tokens/total': 4324960, 'tokens/trainable': 65492, 'epoch': '0.5586'}
-
 28%|██▊       | 143/512 [16:07<41:31,  6.75s/it]
 28%|██▊       | 144/512 [16:13<41:11,  6.72s/it]
                                                 
{'loss': '0.004676', 'grad_norm': '0.4305', 'learning_rate': '8.758e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '35.71', 'tokens/total': 4355424, 'tokens/trainable': 65925, 'epoch': '0.5625'}
-
 28%|██▊       | 144/512 [16:13<41:11,  6.72s/it]
 28%|██▊       | 145/512 [16:20<41:06,  6.72s/it]
                                                 
{'loss': '0.005307', 'grad_norm': '1.153', 'learning_rate': '8.738e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.24', 'tokens/total': 4385712, 'tokens/trainable': 66383, 'epoch': '0.5664'}
-
 28%|██▊       | 145/512 [16:20<41:06,  6.72s/it]
 29%|██▊       | 146/512 [16:27<40:56,  6.71s/it]
                                                 
{'loss': '0.01282', 'grad_norm': '0.774', 'learning_rate': '8.718e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '31.37', 'tokens/total': 4416016, 'tokens/trainable': 66840, 'epoch': '0.5703'}
-
 29%|██▊       | 146/512 [16:27<40:56,  6.71s/it]
 29%|██▊       | 147/512 [16:33<40:47,  6.71s/it]
                                                 
{'loss': '0.02721', 'grad_norm': '0.6979', 'learning_rate': '8.697e-05', 'ppl': '1.028', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 4446368, 'tokens/trainable': 67297, 'epoch': '0.5742'}
-
 29%|██▊       | 147/512 [16:33<40:47,  6.71s/it]
 29%|██▉       | 148/512 [16:40<40:42,  6.71s/it]
                                                 
{'loss': '0.008016', 'grad_norm': '0.1734', 'learning_rate': '8.677e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '37.36', 'tokens/total': 4476880, 'tokens/trainable': 67769, 'epoch': '0.5781'}
-
 29%|██▉       | 148/512 [16:40<40:42,  6.71s/it]
 29%|██▉       | 149/512 [16:47<40:35,  6.71s/it]
                                                 
{'loss': '0.01083', 'grad_norm': '0.6356', 'learning_rate': '8.656e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 4507232, 'tokens/trainable': 68257, 'epoch': '0.582'}
-
 29%|██▉       | 149/512 [16:47<40:35,  6.71s/it]
 29%|██▉       | 150/512 [16:53<40:22,  6.69s/it]
                                                 
{'loss': '0.005296', 'grad_norm': '0.2509', 'learning_rate': '8.635e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 4537376, 'tokens/trainable': 68706, 'epoch': '0.5859'}
-
 29%|██▉       | 150/512 [16:53<40:22,  6.69s/it]
 29%|██▉       | 151/512 [17:00<40:14,  6.69s/it]
                                                 
{'loss': '0.02201', 'grad_norm': '0.5293', 'learning_rate': '8.615e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.25', 'tokens/total': 4567584, 'tokens/trainable': 69131, 'epoch': '0.5898'}
-
 29%|██▉       | 151/512 [17:00<40:14,  6.69s/it]
 30%|██▉       | 152/512 [17:07<40:02,  6.67s/it]
                                                 
{'loss': '0.02459', 'grad_norm': '0.3471', 'learning_rate': '8.594e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.92', 'tokens/total': 4597792, 'tokens/trainable': 69586, 'epoch': '0.5938'}
-
 30%|██▉       | 152/512 [17:07<40:02,  6.67s/it]
 30%|██▉       | 153/512 [17:13<39:59,  6.68s/it]
                                                 
{'loss': '0.002612', 'grad_norm': '0.08057', 'learning_rate': '8.572e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 4627936, 'tokens/trainable': 70061, 'epoch': '0.5977'}
-
 30%|██▉       | 153/512 [17:13<39:59,  6.68s/it]
 30%|███       | 154/512 [17:20<39:50,  6.68s/it]
                                                 
{'loss': '0.003361', 'grad_norm': '0.1623', 'learning_rate': '8.551e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '27.93', 'tokens/total': 4658352, 'tokens/trainable': 70467, 'epoch': '0.6016'}
-
 30%|███       | 154/512 [17:20<39:50,  6.68s/it]
 30%|███       | 155/512 [17:27<39:47,  6.69s/it]
                                                 
{'loss': '0.00435', 'grad_norm': '0.1761', 'learning_rate': '8.53e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 4688896, 'tokens/trainable': 70933, 'epoch': '0.6055'}
-
 30%|███       | 155/512 [17:27<39:47,  6.69s/it]
 30%|███       | 156/512 [17:34<39:44,  6.70s/it]
                                                 
{'loss': '0.01062', 'grad_norm': '0.2591', 'learning_rate': '8.508e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 4719168, 'tokens/trainable': 71382, 'epoch': '0.6094'}
-
 30%|███       | 156/512 [17:34<39:44,  6.70s/it]
 31%|███       | 157/512 [17:40<39:39,  6.70s/it]
                                                 
{'loss': '0.002628', 'grad_norm': '0.1564', 'learning_rate': '8.487e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '30.22', 'tokens/total': 4749136, 'tokens/trainable': 71808, 'epoch': '0.6133'}
-
 31%|███       | 157/512 [17:40<39:39,  6.70s/it]
 31%|███       | 158/512 [17:47<39:32,  6.70s/it]
                                                 
{'loss': '0.01087', 'grad_norm': '0.3443', 'learning_rate': '8.465e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 4779472, 'tokens/trainable': 72249, 'epoch': '0.6172'}
-
 31%|███       | 158/512 [17:47<39:32,  6.70s/it]
 31%|███       | 159/512 [17:54<39:23,  6.69s/it]
                                                 
{'loss': '0.02005', 'grad_norm': '0.3859', 'learning_rate': '8.443e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 4809632, 'tokens/trainable': 72726, 'epoch': '0.6211'}
-
 31%|███       | 159/512 [17:54<39:23,  6.69s/it]
 31%|███▏      | 160/512 [18:00<39:13,  6.68s/it]
                                                 
{'loss': '0.009756', 'grad_norm': '0.3002', 'learning_rate': '8.421e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 4840112, 'tokens/trainable': 73148, 'epoch': '0.625'}
-
 31%|███▏      | 160/512 [18:00<39:13,  6.68s/it][2026-08-18 13:01:48,787] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
-
 31%|███▏      | 161/512 [18:09<42:32,  7.27s/it]
                                                 
{'loss': '0.009085', 'grad_norm': '0.2029', 'learning_rate': '8.399e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 4870656, 'tokens/trainable': 73600, 'epoch': '0.6289'}
-
 31%|███▏      | 161/512 [18:09<42:32,  7.27s/it]
 32%|███▏      | 162/512 [18:16<41:15,  7.07s/it]
                                                 
{'loss': '0.01259', 'grad_norm': '0.2961', 'learning_rate': '8.376e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.71', 'tokens/total': 4900608, 'tokens/trainable': 74068, 'epoch': '0.6328'}
-
 32%|███▏      | 162/512 [18:16<41:15,  7.07s/it]
 32%|███▏      | 163/512 [18:22<40:25,  6.95s/it]
                                                 
{'loss': '0.003114', 'grad_norm': '0.1145', 'learning_rate': '8.354e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.71', 'tokens/total': 4930752, 'tokens/trainable': 74527, 'epoch': '0.6367'}
-
 32%|███▏      | 163/512 [18:22<40:25,  6.95s/it]
 32%|███▏      | 164/512 [18:29<39:49,  6.87s/it]
                                                 
{'loss': '0.005665', 'grad_norm': '0.1261', 'learning_rate': '8.332e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '31.81', 'tokens/total': 4961008, 'tokens/trainable': 74992, 'epoch': '0.6406'}
-
 32%|███▏      | 164/512 [18:29<39:49,  6.87s/it]
 32%|███▏      | 165/512 [18:36<39:24,  6.81s/it]
                                                 
{'loss': '0.01254', 'grad_norm': '0.4959', 'learning_rate': '8.309e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 4991200, 'tokens/trainable': 75442, 'epoch': '0.6445'}
-
 32%|███▏      | 165/512 [18:36<39:24,  6.81s/it]
 32%|███▏      | 166/512 [18:42<39:09,  6.79s/it]
                                                 
{'loss': '0.007678', 'grad_norm': '0.1712', 'learning_rate': '8.286e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 5021600, 'tokens/trainable': 75887, 'epoch': '0.6484'}
-
 32%|███▏      | 166/512 [18:42<39:09,  6.79s/it]
 33%|███▎      | 167/512 [18:49<38:51,  6.76s/it]
                                                 
{'loss': '0.007361', 'grad_norm': '0.1903', 'learning_rate': '8.263e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 5052032, 'tokens/trainable': 76331, 'epoch': '0.6523'}
-
 33%|███▎      | 167/512 [18:49<38:51,  6.76s/it]
 33%|███▎      | 168/512 [18:55<37:45,  6.59s/it]
                                                 
{'loss': '0.001689', 'grad_norm': '0.08689', 'learning_rate': '8.24e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.9', 'tokens/total': 5080256, 'tokens/trainable': 76745, 'epoch': '0.6562'}
-
 33%|███▎      | 168/512 [18:55<37:45,  6.59s/it]
 33%|███▎      | 169/512 [19:02<37:45,  6.61s/it]
                                                 
{'loss': '0.02567', 'grad_norm': '1.222', 'learning_rate': '8.217e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.3', 'tokens/total': 5110784, 'tokens/trainable': 77201, 'epoch': '0.6602'}
-
 33%|███▎      | 169/512 [19:02<37:45,  6.61s/it]
 33%|███▎      | 170/512 [19:08<37:47,  6.63s/it]
                                                 
{'loss': '0.005606', 'grad_norm': '0.2243', 'learning_rate': '8.194e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.38', 'tokens/total': 5141200, 'tokens/trainable': 77655, 'epoch': '0.6641'}
-
 33%|███▎      | 170/512 [19:08<37:47,  6.63s/it]
 33%|███▎      | 171/512 [19:15<37:45,  6.64s/it]
                                                 
{'loss': '0.001366', 'grad_norm': '0.06314', 'learning_rate': '8.171e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '39.25', 'tokens/total': 5171760, 'tokens/trainable': 78122, 'epoch': '0.668'}
-
 33%|███▎      | 171/512 [19:15<37:45,  6.64s/it]
 34%|███▎      | 172/512 [19:22<37:44,  6.66s/it]
                                                 
{'loss': '0.008099', 'grad_norm': '0.3466', 'learning_rate': '8.147e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.59', 'tokens/total': 5202272, 'tokens/trainable': 78576, 'epoch': '0.6719'}
-
 34%|███▎      | 172/512 [19:22<37:44,  6.66s/it]
 34%|███▍      | 173/512 [19:29<37:40,  6.67s/it]
                                                 
{'loss': '0.02078', 'grad_norm': '0.4534', 'learning_rate': '8.124e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '36.49', 'tokens/total': 5232800, 'tokens/trainable': 79059, 'epoch': '0.6758'}
-
 34%|███▍      | 173/512 [19:29<37:40,  6.67s/it]
 34%|███▍      | 174/512 [19:35<37:26,  6.65s/it]
                                                 
{'loss': '0.01096', 'grad_norm': '0.3241', 'learning_rate': '8.1e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '29.73', 'tokens/total': 5262672, 'tokens/trainable': 79498, 'epoch': '0.6797'}
-
 34%|███▍      | 174/512 [19:35<37:26,  6.65s/it]
 34%|███▍      | 175/512 [19:42<37:18,  6.64s/it]
                                                 
{'loss': '0.009046', 'grad_norm': '0.5539', 'learning_rate': '8.076e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '30.45', 'tokens/total': 5293072, 'tokens/trainable': 79922, 'epoch': '0.6836'}
-
 34%|███▍      | 175/512 [19:42<37:18,  6.64s/it]
 34%|███▍      | 176/512 [19:48<36:24,  6.50s/it]
                                                 
{'loss': '0.01483', 'grad_norm': '0.2632', 'learning_rate': '8.053e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.67', 'tokens/total': 5321520, 'tokens/trainable': 80383, 'epoch': '0.6875'}
-
 34%|███▍      | 176/512 [19:48<36:24,  6.50s/it]
 35%|███▍      | 177/512 [19:55<36:35,  6.55s/it]
                                                 
{'loss': '0.006698', 'grad_norm': '0.2084', 'learning_rate': '8.029e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.09', 'tokens/total': 5351904, 'tokens/trainable': 80824, 'epoch': '0.6914'}
-
 35%|███▍      | 177/512 [19:55<36:35,  6.55s/it]
 35%|███▍      | 178/512 [20:01<36:37,  6.58s/it]
                                                 
{'loss': '0.00976', 'grad_norm': '0.3557', 'learning_rate': '8.005e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '32.27', 'tokens/total': 5382432, 'tokens/trainable': 81243, 'epoch': '0.6953'}
-
 35%|███▍      | 178/512 [20:01<36:37,  6.58s/it]
 35%|███▍      | 179/512 [20:08<36:39,  6.60s/it]
                                                 
{'loss': '0.009939', 'grad_norm': '0.2144', 'learning_rate': '7.98e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.36', 'tokens/total': 5412960, 'tokens/trainable': 81716, 'epoch': '0.6992'}
-
 35%|███▍      | 179/512 [20:08<36:39,  6.60s/it]
 35%|███▌      | 180/512 [20:15<36:43,  6.64s/it]
                                                 
{'loss': '0.005616', 'grad_norm': '0.1301', 'learning_rate': '7.956e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.61', 'tokens/total': 5443232, 'tokens/trainable': 82181, 'epoch': '0.7031'}
-
 35%|███▌      | 180/512 [20:15<36:43,  6.64s/it]
 35%|███▌      | 181/512 [20:21<35:51,  6.50s/it]
                                                 
{'loss': '0.007541', 'grad_norm': '0.2382', 'learning_rate': '7.932e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.29', 'memory/max_allocated (GiB)': '33.29', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.63', 'tokens/total': 5471440, 'tokens/trainable': 82626, 'epoch': '0.707'}
-
 35%|███▌      | 181/512 [20:21<35:51,  6.50s/it]
 36%|███▌      | 182/512 [20:28<36:03,  6.56s/it]
                                                 
{'loss': '0.003529', 'grad_norm': '0.1383', 'learning_rate': '7.907e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.17', 'tokens/total': 5501744, 'tokens/trainable': 83089, 'epoch': '0.7109'}
-
 36%|███▌      | 182/512 [20:28<36:03,  6.56s/it]
 36%|███▌      | 183/512 [20:34<36:11,  6.60s/it]
                                                 
{'loss': '0.002233', 'grad_norm': '0.08244', 'learning_rate': '7.883e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 5532272, 'tokens/trainable': 83590, 'epoch': '0.7148'}
-
 36%|███▌      | 183/512 [20:34<36:11,  6.60s/it]
 36%|███▌      | 184/512 [20:41<36:16,  6.64s/it]
                                                 
{'loss': '0.0112', 'grad_norm': '0.2647', 'learning_rate': '7.858e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 5562784, 'tokens/trainable': 84046, 'epoch': '0.7188'}
-
 36%|███▌      | 184/512 [20:41<36:16,  6.64s/it]
 36%|███▌      | 185/512 [20:48<36:10,  6.64s/it]
                                                 
{'loss': '0.0033', 'grad_norm': '0.1043', 'learning_rate': '7.833e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 5592800, 'tokens/trainable': 84517, 'epoch': '0.7227'}
-
 36%|███▌      | 185/512 [20:48<36:10,  6.64s/it]
 36%|███▋      | 186/512 [20:54<36:13,  6.67s/it]
                                                 
{'loss': '0.03358', 'grad_norm': '0.6546', 'learning_rate': '7.808e-05', 'ppl': '1.034', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 5623088, 'tokens/trainable': 84974, 'epoch': '0.7266'}
-
 36%|███▋      | 186/512 [20:54<36:13,  6.67s/it]
 37%|███▋      | 187/512 [21:01<36:13,  6.69s/it]
                                                 
{'loss': '0.0004142', 'grad_norm': '0.0167', 'learning_rate': '7.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 5653344, 'tokens/trainable': 85460, 'epoch': '0.7305'}
-
 37%|███▋      | 187/512 [21:01<36:13,  6.69s/it]
 37%|███▋      | 188/512 [21:08<36:09,  6.70s/it]
                                                 
{'loss': '0.002929', 'grad_norm': '0.1661', 'learning_rate': '7.758e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '35.9', 'tokens/total': 5683600, 'tokens/trainable': 85939, 'epoch': '0.7344'}
-
 37%|███▋      | 188/512 [21:08<36:09,  6.70s/it]
 37%|███▋      | 189/512 [21:14<35:11,  6.54s/it]
                                                 
{'loss': '0.00125', 'grad_norm': '0.3206', 'learning_rate': '7.733e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '38.26', 'tokens/total': 5711952, 'tokens/trainable': 86377, 'epoch': '0.7383'}
-
 37%|███▋      | 189/512 [21:14<35:11,  6.54s/it]
 37%|███▋      | 190/512 [21:21<35:21,  6.59s/it]
                                                 
{'loss': '0.009377', 'grad_norm': '0.5194', 'learning_rate': '7.708e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 5742336, 'tokens/trainable': 86834, 'epoch': '0.7422'}
-
 37%|███▋      | 190/512 [21:21<35:21,  6.59s/it]
 37%|███▋      | 191/512 [21:27<35:20,  6.60s/it]
                                                 
{'loss': '0.009413', 'grad_norm': '0.6359', 'learning_rate': '7.683e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '32.08', 'tokens/total': 5772560, 'tokens/trainable': 87265, 'epoch': '0.7461'}
-
 37%|███▋      | 191/512 [21:27<35:20,  6.60s/it]
 38%|███▊      | 192/512 [21:34<35:28,  6.65s/it]
                                                 
{'loss': '0.001757', 'grad_norm': '0.1094', 'learning_rate': '7.657e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 5803136, 'tokens/trainable': 87747, 'epoch': '0.75'}
-
 38%|███▊      | 192/512 [21:34<35:28,  6.65s/it][2026-08-18 13:05:22,559] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
-
 38%|███▊      | 193/512 [21:43<38:42,  7.28s/it]
                                                 
{'loss': '0.001762', 'grad_norm': '0.08651', 'learning_rate': '7.632e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.57', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 5833440, 'tokens/trainable': 88239, 'epoch': '0.7539'}
-
 38%|███▊      | 193/512 [21:43<38:42,  7.28s/it]
 38%|███▊      | 194/512 [21:49<37:40,  7.11s/it]
                                                 
{'loss': '0.0002287', 'grad_norm': '0.01772', 'learning_rate': '7.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 5863552, 'tokens/trainable': 88718, 'epoch': '0.7578'}
-
 38%|███▊      | 194/512 [21:49<37:40,  7.11s/it]
 38%|███▊      | 195/512 [21:56<36:49,  6.97s/it]
                                                 
{'loss': '0.003605', 'grad_norm': '0.2394', 'learning_rate': '7.58e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '31.27', 'tokens/total': 5893680, 'tokens/trainable': 89129, 'epoch': '0.7617'}
-
 38%|███▊      | 195/512 [21:56<36:49,  6.97s/it]
 38%|███▊      | 196/512 [22:03<36:14,  6.88s/it]
                                                 
{'loss': '0.0001226', 'grad_norm': '0.006238', 'learning_rate': '7.555e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '32.25', 'tokens/total': 5923744, 'tokens/trainable': 89567, 'epoch': '0.7656'}
-
 38%|███▊      | 196/512 [22:03<36:14,  6.88s/it]
 38%|███▊      | 197/512 [22:09<35:48,  6.82s/it]
                                                 
{'loss': '0.03544', 'grad_norm': '0.9131', 'learning_rate': '7.529e-05', 'ppl': '1.036', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '30.25', 'tokens/total': 5954128, 'tokens/trainable': 89988, 'epoch': '0.7695'}
-
 38%|███▊      | 197/512 [22:09<35:48,  6.82s/it]
 39%|███▊      | 198/512 [22:16<35:30,  6.79s/it]
                                                 
{'loss': '0.001015', 'grad_norm': '0.1169', 'learning_rate': '7.503e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 5984352, 'tokens/trainable': 90434, 'epoch': '0.7734'}
-
 39%|███▊      | 198/512 [22:16<35:30,  6.79s/it]
 39%|███▉      | 199/512 [22:23<35:16,  6.76s/it]
                                                 
{'loss': '0.009798', 'grad_norm': '0.5136', 'learning_rate': '7.477e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.61', 'tokens/train_per_sec_per_gpu': '34.17', 'tokens/total': 6014704, 'tokens/trainable': 90913, 'epoch': '0.7773'}
-
 39%|███▉      | 199/512 [22:23<35:16,  6.76s/it]
 39%|███▉      | 200/512 [22:30<34:58,  6.73s/it]
                                                 
{'loss': '0.009519', 'grad_norm': '0.267', 'learning_rate': '7.451e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.17', 'tokens/total': 6045056, 'tokens/trainable': 91355, 'epoch': '0.7812'}
-
 39%|███▉      | 200/512 [22:30<34:58,  6.73s/it]
 39%|███▉      | 201/512 [22:36<34:58,  6.75s/it]
                                                 
{'loss': '0.006804', 'grad_norm': '0.315', 'learning_rate': '7.424e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '37.38', 'tokens/total': 6075504, 'tokens/trainable': 91856, 'epoch': '0.7852'}
-
 39%|███▉      | 201/512 [22:36<34:58,  6.75s/it]
 39%|███▉      | 202/512 [22:43<34:43,  6.72s/it]
                                                 
{'loss': '0.01742', 'grad_norm': '0.5194', 'learning_rate': '7.398e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 6105904, 'tokens/trainable': 92311, 'epoch': '0.7891'}
-
 39%|███▉      | 202/512 [22:43<34:43,  6.72s/it]
 40%|███▉      | 203/512 [22:50<34:38,  6.73s/it]
                                                 
{'loss': '0.002557', 'grad_norm': '0.1296', 'learning_rate': '7.372e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.55', 'tokens/total': 6136272, 'tokens/trainable': 92764, 'epoch': '0.793'}
-
 40%|███▉      | 203/512 [22:50<34:38,  6.73s/it]
 40%|███▉      | 204/512 [22:56<34:24,  6.70s/it]
                                                 
{'loss': '0.00419', 'grad_norm': '0.1787', 'learning_rate': '7.345e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 6166336, 'tokens/trainable': 93227, 'epoch': '0.7969'}
-
 40%|███▉      | 204/512 [22:56<34:24,  6.70s/it]
 40%|████      | 205/512 [23:03<34:12,  6.69s/it]
                                                 
{'loss': '0.0102', 'grad_norm': '0.3807', 'learning_rate': '7.319e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 6196720, 'tokens/trainable': 93675, 'epoch': '0.8008'}
-
 40%|████      | 205/512 [23:03<34:12,  6.69s/it]
 40%|████      | 206/512 [23:10<34:11,  6.70s/it]
                                                 
{'loss': '0.008401', 'grad_norm': '0.3957', 'learning_rate': '7.292e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.33', 'tokens/total': 6227424, 'tokens/trainable': 94120, 'epoch': '0.8047'}
-
 40%|████      | 206/512 [23:10<34:11,  6.70s/it]
 40%|████      | 207/512 [23:16<34:02,  6.70s/it]
                                                 
{'loss': '0.002036', 'grad_norm': '0.1237', 'learning_rate': '7.266e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 6257616, 'tokens/trainable': 94557, 'epoch': '0.8086'}
-
 40%|████      | 207/512 [23:16<34:02,  6.70s/it]
 41%|████      | 208/512 [23:23<33:59,  6.71s/it]
                                                 
{'loss': '0.01048', 'grad_norm': '0.2043', 'learning_rate': '7.239e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '40.52', 'tokens/total': 6288000, 'tokens/trainable': 95035, 'epoch': '0.8125'}
-
 41%|████      | 208/512 [23:23<33:59,  6.71s/it]
 41%|████      | 209/512 [23:30<33:52,  6.71s/it]
                                                 
{'loss': '0.001463', 'grad_norm': '0.06323', 'learning_rate': '7.212e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '37.41', 'tokens/total': 6318336, 'tokens/trainable': 95517, 'epoch': '0.8164'}
-
 41%|████      | 209/512 [23:30<33:52,  6.71s/it]
 41%|████      | 210/512 [23:37<33:47,  6.71s/it]
                                                 
{'loss': '0.007305', 'grad_norm': '0.1743', 'learning_rate': '7.185e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.31', 'tokens/total': 6348656, 'tokens/trainable': 96026, 'epoch': '0.8203'}
-
 41%|████      | 210/512 [23:37<33:47,  6.71s/it]
 41%|████      | 211/512 [23:43<33:38,  6.71s/it]
                                                 
{'loss': '0.002277', 'grad_norm': '0.08656', 'learning_rate': '7.158e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.44', 'tokens/total': 6379040, 'tokens/trainable': 96477, 'epoch': '0.8242'}
-
 41%|████      | 211/512 [23:43<33:38,  6.71s/it]
 41%|████▏     | 212/512 [23:50<33:30,  6.70s/it]
                                                 
{'loss': '0.003339', 'grad_norm': '0.1084', 'learning_rate': '7.131e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 6409312, 'tokens/trainable': 96951, 'epoch': '0.8281'}
-
 41%|████▏     | 212/512 [23:50<33:30,  6.70s/it]
 42%|████▏     | 213/512 [23:57<33:22,  6.70s/it]
                                                 
{'loss': '0.00445', 'grad_norm': '0.2011', 'learning_rate': '7.104e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '30.68', 'tokens/total': 6439520, 'tokens/trainable': 97350, 'epoch': '0.832'}
-
 42%|████▏     | 213/512 [23:57<33:22,  6.70s/it]
 42%|████▏     | 214/512 [24:03<33:14,  6.69s/it]
                                                 
{'loss': '0.001114', 'grad_norm': '0.08952', 'learning_rate': '7.077e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '29.79', 'tokens/total': 6469888, 'tokens/trainable': 97792, 'epoch': '0.8359'}
-
 42%|████▏     | 214/512 [24:03<33:14,  6.69s/it]
 42%|████▏     | 215/512 [24:10<33:10,  6.70s/it]
                                                 
{'loss': '0.002481', 'grad_norm': '0.2406', 'learning_rate': '7.05e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 6500128, 'tokens/trainable': 98257, 'epoch': '0.8398'}
-
 42%|████▏     | 215/512 [24:10<33:10,  6.70s/it]
 42%|████▏     | 216/512 [24:17<33:13,  6.73s/it]
                                                 
{'loss': '0.005798', 'grad_norm': '0.2198', 'learning_rate': '7.022e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.46', 'tokens/total': 6530832, 'tokens/trainable': 98731, 'epoch': '0.8438'}
-
 42%|████▏     | 216/512 [24:17<33:13,  6.73s/it]
 42%|████▏     | 217/512 [24:24<33:04,  6.73s/it]
                                                 
{'loss': '0.0002352', 'grad_norm': '0.02031', 'learning_rate': '6.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.37', 'tokens/total': 6561248, 'tokens/trainable': 99212, 'epoch': '0.8477'}
-
 42%|████▏     | 217/512 [24:24<33:04,  6.73s/it]
 43%|████▎     | 218/512 [24:30<32:50,  6.70s/it]
                                                 
{'loss': '0.0007598', 'grad_norm': '0.04731', 'learning_rate': '6.968e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 6591648, 'tokens/trainable': 99654, 'epoch': '0.8516'}
-
 43%|████▎     | 218/512 [24:30<32:50,  6.70s/it]
 43%|████▎     | 219/512 [24:37<32:48,  6.72s/it]
                                                 
{'loss': '0.006445', 'grad_norm': '0.3725', 'learning_rate': '6.94e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 6622368, 'tokens/trainable': 100086, 'epoch': '0.8555'}
-
 43%|████▎     | 219/512 [24:37<32:48,  6.72s/it]
 43%|████▎     | 220/512 [24:44<32:39,  6.71s/it]
                                                 
{'loss': '0.00061', 'grad_norm': '0.04439', 'learning_rate': '6.913e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 6652848, 'tokens/trainable': 100524, 'epoch': '0.8594'}
-
 43%|████▎     | 220/512 [24:44<32:39,  6.71s/it]
 43%|████▎     | 221/512 [24:51<32:39,  6.73s/it]
                                                 
{'loss': '0.01045', 'grad_norm': '0.4204', 'learning_rate': '6.885e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.27', 'tokens/total': 6683392, 'tokens/trainable': 100996, 'epoch': '0.8633'}
-
 43%|████▎     | 221/512 [24:51<32:39,  6.73s/it]
 43%|████▎     | 222/512 [24:57<32:28,  6.72s/it]
                                                 
{'loss': '0.0005217', 'grad_norm': '0.03999', 'learning_rate': '6.857e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '31.9', 'tokens/total': 6713712, 'tokens/trainable': 101449, 'epoch': '0.8672'}
-
 43%|████▎     | 222/512 [24:57<32:28,  6.72s/it]
 44%|████▎     | 223/512 [25:04<32:17,  6.70s/it]
                                                 
{'loss': '0.01121', 'grad_norm': '0.4914', 'learning_rate': '6.83e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 6744176, 'tokens/trainable': 101905, 'epoch': '0.8711'}
-
 44%|████▎     | 223/512 [25:04<32:17,  6.70s/it]
 44%|████▍     | 224/512 [25:11<32:13,  6.71s/it]
                                                 
{'loss': '0.001121', 'grad_norm': '0.1168', 'learning_rate': '6.802e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.36', 'tokens/total': 6774416, 'tokens/trainable': 102411, 'epoch': '0.875'}
-
 44%|████▍     | 224/512 [25:11<32:13,  6.71s/it][2026-08-18 13:08:59,122] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
-
 44%|████▍     | 225/512 [25:19<34:54,  7.30s/it]
                                                 
{'loss': '0.001277', 'grad_norm': '0.08471', 'learning_rate': '6.774e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 6804592, 'tokens/trainable': 102882, 'epoch': '0.8789'}
-
 44%|████▍     | 225/512 [25:19<34:54,  7.30s/it]
 44%|████▍     | 226/512 [25:26<33:54,  7.11s/it]
                                                 
{'loss': '0.0003806', 'grad_norm': '0.03987', 'learning_rate': '6.746e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '34.57', 'tokens/total': 6834976, 'tokens/trainable': 103332, 'epoch': '0.8828'}
-
 44%|████▍     | 226/512 [25:26<33:54,  7.11s/it]
 44%|████▍     | 227/512 [25:33<33:14,  7.00s/it]
                                                 
{'loss': '0.008083', 'grad_norm': '0.4415', 'learning_rate': '6.718e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 6865408, 'tokens/trainable': 103790, 'epoch': '0.8867'}
-
 44%|████▍     | 227/512 [25:33<33:14,  7.00s/it]
 45%|████▍     | 228/512 [25:39<32:39,  6.90s/it]
                                                 
{'loss': '0.004411', 'grad_norm': '0.2599', 'learning_rate': '6.69e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.35', 'tokens/total': 6895664, 'tokens/trainable': 104246, 'epoch': '0.8906'}
-
 45%|████▍     | 228/512 [25:39<32:39,  6.90s/it]
 45%|████▍     | 229/512 [25:46<32:15,  6.84s/it]
                                                 
{'loss': '0.01891', 'grad_norm': '0.5188', 'learning_rate': '6.662e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.66', 'tokens/total': 6926016, 'tokens/trainable': 104707, 'epoch': '0.8945'}
-
 45%|████▍     | 229/512 [25:46<32:15,  6.84s/it]
 45%|████▍     | 230/512 [25:53<31:57,  6.80s/it]
                                                 
{'loss': '0.006382', 'grad_norm': '0.2526', 'learning_rate': '6.634e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.83', 'tokens/total': 6956384, 'tokens/trainable': 105167, 'epoch': '0.8984'}
-
 45%|████▍     | 230/512 [25:53<31:57,  6.80s/it]
 45%|████▌     | 231/512 [25:59<31:37,  6.75s/it]
                                                 
{'loss': '0.0007957', 'grad_norm': '0.04988', 'learning_rate': '6.606e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 6986768, 'tokens/trainable': 105576, 'epoch': '0.9023'}
-
 45%|████▌     | 231/512 [25:59<31:37,  6.75s/it]
 45%|████▌     | 232/512 [26:06<31:23,  6.73s/it]
                                                 
{'loss': '0.0148', 'grad_norm': '0.4156', 'learning_rate': '6.578e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.76', 'tokens/total': 7017232, 'tokens/trainable': 106043, 'epoch': '0.9062'}
-
 45%|████▌     | 232/512 [26:06<31:23,  6.73s/it]
 46%|████▌     | 233/512 [26:13<31:14,  6.72s/it]
                                                 
{'loss': '0.003381', 'grad_norm': '0.16', 'learning_rate': '6.55e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 7047568, 'tokens/trainable': 106506, 'epoch': '0.9102'}
-
 46%|████▌     | 233/512 [26:13<31:14,  6.72s/it]
 46%|████▌     | 234/512 [26:19<31:00,  6.69s/it]
                                                 
{'loss': '0.0007064', 'grad_norm': '0.04383', 'learning_rate': '6.522e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 7077760, 'tokens/trainable': 106951, 'epoch': '0.9141'}
-
 46%|████▌     | 234/512 [26:19<31:00,  6.69s/it]
 46%|████▌     | 235/512 [26:26<30:45,  6.66s/it]
                                                 
{'loss': '0.003099', 'grad_norm': '0.133', 'learning_rate': '6.493e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.78', 'tokens/total': 7107760, 'tokens/trainable': 107382, 'epoch': '0.918'}
-
 46%|████▌     | 235/512 [26:26<30:45,  6.66s/it]
 46%|████▌     | 236/512 [26:33<30:40,  6.67s/it]
                                                 
{'loss': '0.00322', 'grad_norm': '0.3202', 'learning_rate': '6.465e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 7138144, 'tokens/trainable': 107814, 'epoch': '0.9219'}
-
 46%|████▌     | 236/512 [26:33<30:40,  6.67s/it]
 46%|████▋     | 237/512 [26:39<30:34,  6.67s/it]
                                                 
{'loss': '0.002884', 'grad_norm': '0.1548', 'learning_rate': '6.437e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 7168352, 'tokens/trainable': 108274, 'epoch': '0.9258'}
-
 46%|████▋     | 237/512 [26:39<30:34,  6.67s/it]
 46%|████▋     | 238/512 [26:46<30:28,  6.67s/it]
                                                 
{'loss': '0.002136', 'grad_norm': '0.09865', 'learning_rate': '6.408e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.22', 'tokens/total': 7198624, 'tokens/trainable': 108714, 'epoch': '0.9297'}
-
 46%|████▋     | 238/512 [26:46<30:28,  6.67s/it]
 47%|████▋     | 239/512 [26:53<30:22,  6.68s/it]
                                                 
{'loss': '0.001417', 'grad_norm': '0.06801', 'learning_rate': '6.38e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.79', 'tokens/total': 7229216, 'tokens/trainable': 109220, 'epoch': '0.9336'}
-
 47%|████▋     | 239/512 [26:53<30:22,  6.68s/it]
 47%|████▋     | 240/512 [26:59<30:15,  6.67s/it]
                                                 
{'loss': '0.002807', 'grad_norm': '0.217', 'learning_rate': '6.351e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.12', 'tokens/total': 7259648, 'tokens/trainable': 109672, 'epoch': '0.9375'}
-
 47%|████▋     | 240/512 [26:59<30:15,  6.67s/it]
 47%|████▋     | 241/512 [27:06<30:09,  6.68s/it]
                                                 
{'loss': '0.01761', 'grad_norm': '0.4365', 'learning_rate': '6.323e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 7289824, 'tokens/trainable': 110135, 'epoch': '0.9414'}
-
 47%|████▋     | 241/512 [27:06<30:09,  6.68s/it]
 47%|████▋     | 242/512 [27:13<30:05,  6.69s/it]
                                                 
{'loss': '0.0006151', 'grad_norm': '0.06252', 'learning_rate': '6.294e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 7320288, 'tokens/trainable': 110563, 'epoch': '0.9453'}
-
 47%|████▋     | 242/512 [27:13<30:05,  6.69s/it]
 47%|████▋     | 243/512 [27:19<29:58,  6.69s/it]
                                                 
{'loss': '0.003188', 'grad_norm': '0.1603', 'learning_rate': '6.266e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.12', 'tokens/total': 7350736, 'tokens/trainable': 111049, 'epoch': '0.9492'}
-
 47%|████▋     | 243/512 [27:19<29:58,  6.69s/it]
 48%|████▊     | 244/512 [27:26<30:00,  6.72s/it]
                                                 
{'loss': '0.0007044', 'grad_norm': '0.1788', 'learning_rate': '6.237e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 7381280, 'tokens/trainable': 111495, 'epoch': '0.9531'}
-
 48%|████▊     | 244/512 [27:26<30:00,  6.72s/it]
 48%|████▊     | 245/512 [27:33<29:51,  6.71s/it]
                                                 
{'loss': '0.0006958', 'grad_norm': '0.05235', 'learning_rate': '6.208e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.82', 'tokens/total': 7411632, 'tokens/trainable': 111968, 'epoch': '0.957'}
-
 48%|████▊     | 245/512 [27:33<29:51,  6.71s/it]
 48%|████▊     | 246/512 [27:40<29:38,  6.68s/it]
                                                 
{'loss': '0.008788', 'grad_norm': '0.1738', 'learning_rate': '6.18e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 7441904, 'tokens/trainable': 112416, 'epoch': '0.9609'}
-
 48%|████▊     | 246/512 [27:40<29:38,  6.68s/it]
 48%|████▊     | 247/512 [27:46<29:36,  6.71s/it]
                                                 
{'loss': '0.001352', 'grad_norm': '0.1043', 'learning_rate': '6.151e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 7472368, 'tokens/trainable': 112882, 'epoch': '0.9648'}
-
 48%|████▊     | 247/512 [27:46<29:36,  6.71s/it]
 48%|████▊     | 248/512 [27:53<29:30,  6.71s/it]
                                                 
{'loss': '0.001037', 'grad_norm': '0.0733', 'learning_rate': '6.122e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '40.07', 'tokens/total': 7502656, 'tokens/trainable': 113390, 'epoch': '0.9688'}
-
 48%|████▊     | 248/512 [27:53<29:30,  6.71s/it]
 49%|████▊     | 249/512 [28:00<29:22,  6.70s/it]
                                                 
{'loss': '0.01024', 'grad_norm': '0.7874', 'learning_rate': '6.093e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '37.84', 'tokens/total': 7532800, 'tokens/trainable': 113904, 'epoch': '0.9727'}
-
 49%|████▊     | 249/512 [28:00<29:22,  6.70s/it]
 49%|████▉     | 250/512 [28:06<29:17,  6.71s/it]
                                                 
{'loss': '0.0006508', 'grad_norm': '0.041', 'learning_rate': '6.065e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 7563264, 'tokens/trainable': 114351, 'epoch': '0.9766'}
-
 49%|████▉     | 250/512 [28:06<29:17,  6.71s/it]
 49%|████▉     | 251/512 [28:13<29:12,  6.71s/it]
                                                 
{'loss': '0.002258', 'grad_norm': '0.147', 'learning_rate': '6.036e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 7593840, 'tokens/trainable': 114842, 'epoch': '0.9805'}
-
 49%|████▉     | 251/512 [28:13<29:12,  6.71s/it]
 49%|████▉     | 252/512 [28:20<29:02,  6.70s/it]
                                                 
{'loss': '0.00276', 'grad_norm': '0.209', 'learning_rate': '6.007e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '28.92', 'tokens/total': 7624416, 'tokens/trainable': 115263, 'epoch': '0.9844'}
-
 49%|████▉     | 252/512 [28:20<29:02,  6.70s/it]
 49%|████▉     | 253/512 [28:27<28:54,  6.70s/it]
                                                 
{'loss': '0.001512', 'grad_norm': '0.1135', 'learning_rate': '5.978e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 7654688, 'tokens/trainable': 115703, 'epoch': '0.9883'}
-
 49%|████▉     | 253/512 [28:27<28:54,  6.70s/it]
 50%|████▉     | 254/512 [28:33<28:43,  6.68s/it]
                                                 
{'loss': '0.00226', 'grad_norm': '0.2045', 'learning_rate': '5.949e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 7685088, 'tokens/trainable': 116157, 'epoch': '0.9922'}
-
 50%|████▉     | 254/512 [28:33<28:43,  6.68s/it]
 50%|████▉     | 255/512 [28:40<28:33,  6.67s/it]
                                                 
{'loss': '0.01213', 'grad_norm': '0.662', 'learning_rate': '5.92e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 7715552, 'tokens/trainable': 116567, 'epoch': '0.9961'}
-
 50%|████▉     | 255/512 [28:40<28:33,  6.67s/it]
 50%|█████     | 256/512 [28:47<28:36,  6.71s/it]
                                                 
{'loss': '0.0002805', 'grad_norm': '0.01708', 'learning_rate': '5.891e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '31.26', 'tokens/total': 7745872, 'tokens/trainable': 117030, 'epoch': '1'}
-
 50%|█████     | 256/512 [28:47<28:36,  6.71s/it][2026-08-18 13:12:35,117] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
-
 50%|█████     | 257/512 [28:56<31:19,  7.37s/it]
                                                 
{'loss': '0.0001653', 'grad_norm': '0.007685', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 7776208, 'tokens/trainable': 117517, 'epoch': '1.004'}
-
 50%|█████     | 257/512 [28:56<31:19,  7.37s/it]
 50%|█████     | 258/512 [29:02<30:20,  7.17s/it]
                                                 
{'loss': '0.0004825', 'grad_norm': '0.04673', 'learning_rate': '5.834e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '38.44', 'tokens/total': 7806480, 'tokens/trainable': 118015, 'epoch': '1.008'}
-
 50%|█████     | 258/512 [29:02<30:20,  7.17s/it]
 51%|█████     | 259/512 [29:09<29:37,  7.03s/it]
                                                 
{'loss': '0.001111', 'grad_norm': '0.09396', 'learning_rate': '5.805e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 7836800, 'tokens/trainable': 118520, 'epoch': '1.012'}
-
 51%|█████     | 259/512 [29:09<29:37,  7.03s/it]
 51%|█████     | 260/512 [29:16<29:06,  6.93s/it]
                                                 
{'loss': '0.0003673', 'grad_norm': '0.02427', 'learning_rate': '5.776e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 7867040, 'tokens/trainable': 118992, 'epoch': '1.016'}
-
 51%|█████     | 260/512 [29:16<29:06,  6.93s/it]
 51%|█████     | 261/512 [29:22<28:41,  6.86s/it]
                                                 
{'loss': '0.002127', 'grad_norm': '0.1085', 'learning_rate': '5.747e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 7897408, 'tokens/trainable': 119438, 'epoch': '1.02'}
-
 51%|█████     | 261/512 [29:22<28:41,  6.86s/it]
 51%|█████     | 262/512 [29:29<28:18,  6.79s/it]
                                                 
{'loss': '0.001213', 'grad_norm': '0.1838', 'learning_rate': '5.718e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.26', 'tokens/total': 7927648, 'tokens/trainable': 119881, 'epoch': '1.023'}
-
 51%|█████     | 262/512 [29:29<28:18,  6.79s/it]
 51%|█████▏    | 263/512 [29:36<27:59,  6.74s/it]
                                                 
{'loss': '0.002099', 'grad_norm': '0.2827', 'learning_rate': '5.689e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '27.91', 'tokens/total': 7957968, 'tokens/trainable': 120285, 'epoch': '1.027'}
-
 51%|█████▏    | 263/512 [29:36<27:59,  6.74s/it]
 52%|█████▏    | 264/512 [29:42<27:43,  6.71s/it]
                                                 
{'loss': '0.0004398', 'grad_norm': '0.07407', 'learning_rate': '5.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.2', 'tokens/total': 7987888, 'tokens/trainable': 120755, 'epoch': '1.031'}
-
 52%|█████▏    | 264/512 [29:42<27:43,  6.71s/it]
 52%|█████▏    | 265/512 [29:49<27:32,  6.69s/it]
                                                 
{'loss': '0.0001488', 'grad_norm': '0.01879', 'learning_rate': '5.631e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.69', 'tokens/total': 8018048, 'tokens/trainable': 121194, 'epoch': '1.035'}
-
 52%|█████▏    | 265/512 [29:49<27:32,  6.69s/it]
 52%|█████▏    | 266/512 [29:56<27:24,  6.69s/it]
                                                 
{'loss': '0.0001396', 'grad_norm': '0.01209', 'learning_rate': '5.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 8048448, 'tokens/trainable': 121670, 'epoch': '1.039'}
-
 52%|█████▏    | 266/512 [29:56<27:24,  6.69s/it]
 52%|█████▏    | 267/512 [30:02<27:21,  6.70s/it]
                                                 
{'loss': '0.0001756', 'grad_norm': '0.03192', 'learning_rate': '5.573e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 8078848, 'tokens/trainable': 122142, 'epoch': '1.043'}
-
 52%|█████▏    | 267/512 [30:02<27:21,  6.70s/it]
 52%|█████▏    | 268/512 [30:09<27:19,  6.72s/it]
                                                 
{'loss': '0.0001036', 'grad_norm': '0.008872', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.41', 'tokens/total': 8109360, 'tokens/trainable': 122585, 'epoch': '1.047'}
-
 52%|█████▏    | 268/512 [30:09<27:19,  6.72s/it]
 53%|█████▎    | 269/512 [30:16<27:12,  6.72s/it]
                                                 
{'loss': '0.002882', 'grad_norm': '0.3732', 'learning_rate': '5.515e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '38.85', 'tokens/total': 8139888, 'tokens/trainable': 123091, 'epoch': '1.051'}
-
 53%|█████▎    | 269/512 [30:16<27:12,  6.72s/it]
 53%|█████▎    | 270/512 [30:23<27:14,  6.75s/it]
                                                 
{'loss': '0.0003205', 'grad_norm': '0.01947', 'learning_rate': '5.485e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '30.9', 'tokens/total': 8170480, 'tokens/trainable': 123505, 'epoch': '1.055'}
-
 53%|█████▎    | 270/512 [30:23<27:14,  6.75s/it]
 53%|█████▎    | 271/512 [30:29<26:28,  6.59s/it]
                                                 
{'loss': '0.0003346', 'grad_norm': '0.0526', 'learning_rate': '5.456e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 8198848, 'tokens/trainable': 123953, 'epoch': '1.059'}
-
 53%|█████▎    | 271/512 [30:29<26:28,  6.59s/it]
 53%|█████▎    | 272/512 [30:36<26:32,  6.64s/it]
                                                 
{'loss': '0.0004355', 'grad_norm': '0.09935', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '28.66', 'tokens/total': 8229200, 'tokens/trainable': 124400, 'epoch': '1.062'}
-
 53%|█████▎    | 272/512 [30:36<26:32,  6.64s/it]
 53%|█████▎    | 273/512 [30:42<26:30,  6.66s/it]
                                                 
{'loss': '0.00702', 'grad_norm': '0.3927', 'learning_rate': '5.398e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.05', 'tokens/total': 8259536, 'tokens/trainable': 124870, 'epoch': '1.066'}
-
 53%|█████▎    | 273/512 [30:42<26:30,  6.66s/it]
 54%|█████▎    | 274/512 [30:49<26:25,  6.66s/it]
                                                 
{'loss': '0.0003887', 'grad_norm': '0.0385', 'learning_rate': '5.369e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.07', 'tokens/total': 8289904, 'tokens/trainable': 125333, 'epoch': '1.07'}
-
 54%|█████▎    | 274/512 [30:49<26:25,  6.66s/it]
 54%|█████▎    | 275/512 [30:56<26:23,  6.68s/it]
                                                 
{'loss': '8.032e-05', 'grad_norm': '0.01037', 'learning_rate': '5.34e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 8320176, 'tokens/trainable': 125834, 'epoch': '1.074'}
-
 54%|█████▎    | 275/512 [30:56<26:23,  6.68s/it]
 54%|█████▍    | 276/512 [31:02<26:19,  6.69s/it]
                                                 
{'loss': '0.0002651', 'grad_norm': '0.05539', 'learning_rate': '5.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '37.97', 'tokens/total': 8350832, 'tokens/trainable': 126316, 'epoch': '1.078'}
-
 54%|█████▍    | 276/512 [31:02<26:19,  6.69s/it]
 54%|█████▍    | 277/512 [31:09<26:14,  6.70s/it]
                                                 
{'loss': '0.0007178', 'grad_norm': '0.2011', 'learning_rate': '5.282e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 8380976, 'tokens/trainable': 126798, 'epoch': '1.082'}
-
 54%|█████▍    | 277/512 [31:09<26:14,  6.70s/it]
 54%|█████▍    | 278/512 [31:16<26:00,  6.67s/it]
                                                 
{'loss': '4.906e-05', 'grad_norm': '0.005951', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.57', 'tokens/total': 8411072, 'tokens/trainable': 127223, 'epoch': '1.086'}
-
 54%|█████▍    | 278/512 [31:16<26:00,  6.67s/it]
 54%|█████▍    | 279/512 [31:22<25:45,  6.63s/it]
                                                 
{'loss': '8.996e-05', 'grad_norm': '0.009279', 'learning_rate': '5.224e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.01', 'tokens/total': 8440736, 'tokens/trainable': 127672, 'epoch': '1.09'}
-
 54%|█████▍    | 279/512 [31:22<25:45,  6.63s/it]
 55%|█████▍    | 280/512 [31:29<25:39,  6.64s/it]
                                                 
{'loss': '0.0006871', 'grad_norm': '0.1146', 'learning_rate': '5.195e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.07', 'tokens/total': 8470944, 'tokens/trainable': 128116, 'epoch': '1.094'}
-
 55%|█████▍    | 280/512 [31:29<25:39,  6.64s/it]
 55%|█████▍    | 281/512 [31:36<25:39,  6.66s/it]
                                                 
{'loss': '4.705e-05', 'grad_norm': '0.004336', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 8501104, 'tokens/trainable': 128589, 'epoch': '1.098'}
-
 55%|█████▍    | 281/512 [31:36<25:39,  6.66s/it]
 55%|█████▌    | 282/512 [31:42<25:34,  6.67s/it]
                                                 
{'loss': '0.01315', 'grad_norm': '0.2142', 'learning_rate': '5.138e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 8531696, 'tokens/trainable': 129036, 'epoch': '1.102'}
-
 55%|█████▌    | 282/512 [31:42<25:34,  6.67s/it]
 55%|█████▌    | 283/512 [31:49<25:31,  6.69s/it]
                                                 
{'loss': '1.559e-05', 'grad_norm': '0.0006189', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.16', 'tokens/total': 8561936, 'tokens/trainable': 129485, 'epoch': '1.105'}
-
 55%|█████▌    | 283/512 [31:49<25:31,  6.69s/it]
 55%|█████▌    | 284/512 [31:56<25:26,  6.69s/it]
                                                 
{'loss': '8.038e-05', 'grad_norm': '0.005335', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.49', 'tokens/total': 8592192, 'tokens/trainable': 129920, 'epoch': '1.109'}
-
 55%|█████▌    | 284/512 [31:56<25:26,  6.69s/it]
 56%|█████▌    | 285/512 [32:02<25:22,  6.71s/it]
                                                 
{'loss': '4.425e-05', 'grad_norm': '0.003026', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.06', 'tokens/total': 8622432, 'tokens/trainable': 130417, 'epoch': '1.113'}
-
 56%|█████▌    | 285/512 [32:02<25:22,  6.71s/it]
 56%|█████▌    | 286/512 [32:09<25:16,  6.71s/it]
                                                 
{'loss': '5.365e-05', 'grad_norm': '0.003212', 'learning_rate': '5.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.89', 'tokens/total': 8652768, 'tokens/trainable': 130903, 'epoch': '1.117'}
-
 56%|█████▌    | 286/512 [32:09<25:16,  6.71s/it]
 56%|█████▌    | 287/512 [32:16<25:08,  6.70s/it]
                                                 
{'loss': '9.199e-05', 'grad_norm': '0.01538', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 8683296, 'tokens/trainable': 131343, 'epoch': '1.121'}
-
 56%|█████▌    | 287/512 [32:16<25:08,  6.70s/it]
 56%|█████▋    | 288/512 [32:23<25:00,  6.70s/it]
                                                 
{'loss': '0.006935', 'grad_norm': '0.8839', 'learning_rate': '4.964e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.19', 'tokens/total': 8713504, 'tokens/trainable': 131763, 'epoch': '1.125'}
-
 56%|█████▋    | 288/512 [32:23<25:00,  6.70s/it][2026-08-18 13:16:11,062] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
-
 56%|█████▋    | 289/512 [32:31<27:23,  7.37s/it]
                                                 
{'loss': '2.326e-05', 'grad_norm': '0.001999', 'learning_rate': '4.935e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 8743888, 'tokens/trainable': 132229, 'epoch': '1.129'}
-
 56%|█████▋    | 289/512 [32:31<27:23,  7.37s/it]
 57%|█████▋    | 290/512 [32:38<26:30,  7.16s/it]
                                                 
{'loss': '1.431e-05', 'grad_norm': '0.0004976', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.54', 'tokens/train_per_sec_per_gpu': '35.46', 'tokens/total': 8774336, 'tokens/trainable': 132678, 'epoch': '1.133'}
-
 57%|█████▋    | 290/512 [32:38<26:30,  7.16s/it]
 57%|█████▋    | 291/512 [32:45<25:52,  7.03s/it]
                                                 
{'loss': '0.0001439', 'grad_norm': '0.01294', 'learning_rate': '4.878e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.39', 'tokens/total': 8804816, 'tokens/trainable': 133136, 'epoch': '1.137'}
-
 57%|█████▋    | 291/512 [32:45<25:52,  7.03s/it]
 57%|█████▋    | 292/512 [32:52<25:25,  6.93s/it]
                                                 
{'loss': '0.003074', 'grad_norm': '0.3451', 'learning_rate': '4.849e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '35.31', 'tokens/total': 8835024, 'tokens/trainable': 133608, 'epoch': '1.141'}
-
 57%|█████▋    | 292/512 [32:52<25:25,  6.93s/it]
 57%|█████▋    | 293/512 [32:58<25:01,  6.86s/it]
                                                 
{'loss': '0.0001658', 'grad_norm': '0.02419', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '40.44', 'tokens/total': 8865408, 'tokens/trainable': 134108, 'epoch': '1.145'}
-
 57%|█████▋    | 293/512 [32:58<25:01,  6.86s/it]
 57%|█████▋    | 294/512 [33:05<24:44,  6.81s/it]
                                                 
{'loss': '3.24e-05', 'grad_norm': '0.001724', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.72', 'tokens/total': 8895824, 'tokens/trainable': 134541, 'epoch': '1.148'}
-
 57%|█████▋    | 294/512 [33:05<24:44,  6.81s/it]
 58%|█████▊    | 295/512 [33:12<24:29,  6.77s/it]
                                                 
{'loss': '0.000106', 'grad_norm': '0.01028', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.51', 'tokens/total': 8926336, 'tokens/trainable': 134966, 'epoch': '1.152'}
-
 58%|█████▊    | 295/512 [33:12<24:29,  6.77s/it]
 58%|█████▊    | 296/512 [33:18<24:14,  6.73s/it]
                                                 
{'loss': '0.0002625', 'grad_norm': '0.02922', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '33.56', 'tokens/total': 8956656, 'tokens/trainable': 135402, 'epoch': '1.156'}
-
 58%|█████▊    | 296/512 [33:18<24:14,  6.73s/it]
 58%|█████▊    | 297/512 [33:25<23:58,  6.69s/it]
                                                 
{'loss': '0.004879', 'grad_norm': '0.4516', 'learning_rate': '4.706e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.05', 'tokens/total': 8986752, 'tokens/trainable': 135804, 'epoch': '1.16'}
-
 58%|█████▊    | 297/512 [33:25<23:58,  6.69s/it]
 58%|█████▊    | 298/512 [33:31<23:46,  6.67s/it]
                                                 
{'loss': '0.0001912', 'grad_norm': '0.0781', 'learning_rate': '4.677e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 9017024, 'tokens/trainable': 136231, 'epoch': '1.164'}
-
 58%|█████▊    | 298/512 [33:31<23:46,  6.67s/it]
 58%|█████▊    | 299/512 [33:38<23:41,  6.67s/it]
                                                 
{'loss': '6.207e-05', 'grad_norm': '0.008365', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '34.27', 'tokens/total': 9047424, 'tokens/trainable': 136705, 'epoch': '1.168'}
-
 58%|█████▊    | 299/512 [33:38<23:41,  6.67s/it]
 59%|█████▊    | 300/512 [33:45<23:31,  6.66s/it]
                                                 
{'loss': '0.008261', 'grad_norm': '0.3563', 'learning_rate': '4.62e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.86', 'tokens/total': 9077648, 'tokens/trainable': 137178, 'epoch': '1.172'}
-
 59%|█████▊    | 300/512 [33:45<23:31,  6.66s/it]
 59%|█████▉    | 301/512 [33:51<23:26,  6.67s/it]
                                                 
{'loss': '0.0006729', 'grad_norm': '0.1326', 'learning_rate': '4.592e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.42', 'tokens/total': 9108032, 'tokens/trainable': 137643, 'epoch': '1.176'}
-
 59%|█████▉    | 301/512 [33:51<23:26,  6.67s/it]
 59%|█████▉    | 302/512 [33:58<23:20,  6.67s/it]
                                                 
{'loss': '0.0008891', 'grad_norm': '0.1117', 'learning_rate': '4.563e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 9138240, 'tokens/trainable': 138078, 'epoch': '1.18'}
-
 59%|█████▉    | 302/512 [33:58<23:20,  6.67s/it]
 59%|█████▉    | 303/512 [34:05<23:11,  6.66s/it]
                                                 
{'loss': '0.004502', 'grad_norm': '0.6568', 'learning_rate': '4.535e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 9168352, 'tokens/trainable': 138515, 'epoch': '1.184'}
-
 59%|█████▉    | 303/512 [34:05<23:11,  6.66s/it]
 59%|█████▉    | 304/512 [34:12<23:10,  6.69s/it]
                                                 
{'loss': '9.651e-05', 'grad_norm': '0.007431', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.99', 'tokens/total': 9198864, 'tokens/trainable': 138948, 'epoch': '1.188'}
-
 59%|█████▉    | 304/512 [34:12<23:10,  6.69s/it]
 60%|█████▉    | 305/512 [34:18<23:03,  6.68s/it]
                                                 
{'loss': '5.174e-05', 'grad_norm': '0.003385', 'learning_rate': '4.478e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '38.17', 'tokens/total': 9229104, 'tokens/trainable': 139408, 'epoch': '1.191'}
-
 60%|█████▉    | 305/512 [34:18<23:03,  6.68s/it]
 60%|█████▉    | 306/512 [34:24<22:24,  6.53s/it]
                                                 
{'loss': '3.198e-05', 'grad_norm': '0.001723', 'learning_rate': '4.45e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '33.48', 'tokens/total': 9257344, 'tokens/trainable': 139840, 'epoch': '1.195'}
-
 60%|█████▉    | 306/512 [34:24<22:24,  6.53s/it]
 60%|█████▉    | 307/512 [34:31<22:27,  6.57s/it]
                                                 
{'loss': '0.0004305', 'grad_norm': '0.06396', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 9287520, 'tokens/trainable': 140281, 'epoch': '1.199'}
-
 60%|█████▉    | 307/512 [34:31<22:27,  6.57s/it]
 60%|██████    | 308/512 [34:38<22:27,  6.61s/it]
                                                 
{'loss': '0.000683', 'grad_norm': '0.1434', 'learning_rate': '4.394e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '37.44', 'tokens/total': 9317888, 'tokens/trainable': 140776, 'epoch': '1.203'}
-
 60%|██████    | 308/512 [34:38<22:27,  6.61s/it]
 60%|██████    | 309/512 [34:44<22:24,  6.63s/it]
                                                 
{'loss': '0.005062', 'grad_norm': '0.392', 'learning_rate': '4.366e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '36.89', 'tokens/total': 9348256, 'tokens/trainable': 141228, 'epoch': '1.207'}
-
 60%|██████    | 309/512 [34:44<22:24,  6.63s/it]
 61%|██████    | 310/512 [34:51<22:18,  6.63s/it]
                                                 
{'loss': '0.0001797', 'grad_norm': '0.01664', 'learning_rate': '4.338e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '37.42', 'tokens/total': 9378560, 'tokens/trainable': 141679, 'epoch': '1.211'}
-
 61%|██████    | 310/512 [34:51<22:18,  6.63s/it]
 61%|██████    | 311/512 [34:58<22:12,  6.63s/it]
                                                 
{'loss': '1.943e-05', 'grad_norm': '0.0008493', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '32.27', 'tokens/total': 9408848, 'tokens/trainable': 142122, 'epoch': '1.215'}
-
 61%|██████    | 311/512 [34:58<22:12,  6.63s/it]
 61%|██████    | 312/512 [35:04<22:06,  6.63s/it]
                                                 
{'loss': '0.0005122', 'grad_norm': '0.04301', 'learning_rate': '4.282e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.11', 'tokens/total': 9439344, 'tokens/trainable': 142561, 'epoch': '1.219'}
-
 61%|██████    | 312/512 [35:04<22:06,  6.63s/it]
 61%|██████    | 313/512 [35:11<22:04,  6.65s/it]
                                                 
{'loss': '0.0001515', 'grad_norm': '0.02346', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.66', 'tokens/total': 9469792, 'tokens/trainable': 143046, 'epoch': '1.223'}
-
 61%|██████    | 313/512 [35:11<22:04,  6.65s/it]
 61%|██████▏   | 314/512 [35:18<22:00,  6.67s/it]
                                                 
{'loss': '0.0006228', 'grad_norm': '0.06129', 'learning_rate': '4.226e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 9500128, 'tokens/trainable': 143493, 'epoch': '1.227'}
-
 61%|██████▏   | 314/512 [35:18<22:00,  6.67s/it]
 62%|██████▏   | 315/512 [35:24<21:51,  6.66s/it]
                                                 
{'loss': '7.468e-05', 'grad_norm': '0.007639', 'learning_rate': '4.198e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 9530480, 'tokens/trainable': 143956, 'epoch': '1.23'}
-
 62%|██████▏   | 315/512 [35:24<21:51,  6.66s/it]
 62%|██████▏   | 316/512 [35:31<21:47,  6.67s/it]
                                                 
{'loss': '0.00437', 'grad_norm': '0.3535', 'learning_rate': '4.17e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 9560912, 'tokens/trainable': 144411, 'epoch': '1.234'}
-
 62%|██████▏   | 316/512 [35:31<21:47,  6.67s/it]
 62%|██████▏   | 317/512 [35:38<21:44,  6.69s/it]
                                                 
{'loss': '0.01097', 'grad_norm': '0.502', 'learning_rate': '4.143e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '38.48', 'tokens/total': 9591344, 'tokens/trainable': 144892, 'epoch': '1.238'}
-
 62%|██████▏   | 317/512 [35:38<21:44,  6.69s/it]
 62%|██████▏   | 318/512 [35:44<21:35,  6.68s/it]
                                                 
{'loss': '4.459e-05', 'grad_norm': '0.002735', 'learning_rate': '4.115e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '32.6', 'tokens/total': 9621392, 'tokens/trainable': 145339, 'epoch': '1.242'}
-
 62%|██████▏   | 318/512 [35:44<21:35,  6.68s/it]
 62%|██████▏   | 319/512 [35:51<21:31,  6.69s/it]
                                                 
{'loss': '8.017e-05', 'grad_norm': '0.01343', 'learning_rate': '4.087e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 9651728, 'tokens/trainable': 145811, 'epoch': '1.246'}
-
 62%|██████▏   | 319/512 [35:51<21:31,  6.69s/it]
 62%|██████▎   | 320/512 [35:58<21:27,  6.71s/it]
                                                 
{'loss': '0.0001716', 'grad_norm': '0.02255', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '38.76', 'tokens/total': 9682224, 'tokens/trainable': 146314, 'epoch': '1.25'}
-
 62%|██████▎   | 320/512 [35:58<21:27,  6.71s/it][2026-08-18 13:19:46,437] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
-
 63%|██████▎   | 321/512 [36:06<23:06,  7.26s/it]
                                                 
{'loss': '0.0005239', 'grad_norm': '0.0436', 'learning_rate': '4.032e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '31.85', 'tokens/total': 9712656, 'tokens/trainable': 146781, 'epoch': '1.254'}
-
 63%|██████▎   | 321/512 [36:06<23:06,  7.26s/it]
 63%|██████▎   | 322/512 [36:13<22:23,  7.07s/it]
                                                 
{'loss': '0.0001891', 'grad_norm': '0.02378', 'learning_rate': '4.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '34.89', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 9743088, 'tokens/trainable': 147225, 'epoch': '1.258'}
-
 63%|██████▎   | 322/512 [36:13<22:23,  7.07s/it]
 63%|██████▎   | 323/512 [36:20<21:58,  6.98s/it]
                                                 
{'loss': '8.73e-05', 'grad_norm': '0.01257', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.17', 'tokens/train_per_sec_per_gpu': '34.82', 'tokens/total': 9773808, 'tokens/trainable': 147673, 'epoch': '1.262'}
-
 63%|██████▎   | 323/512 [36:20<21:58,  6.98s/it]
 63%|██████▎   | 324/512 [36:26<21:29,  6.86s/it]
                                                 
{'loss': '0.002842', 'grad_norm': '0.2107', 'learning_rate': '3.95e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 9804016, 'tokens/trainable': 148103, 'epoch': '1.266'}
-
 63%|██████▎   | 324/512 [36:26<21:29,  6.86s/it]
 63%|██████▎   | 325/512 [36:33<21:11,  6.80s/it]
                                                 
{'loss': '0.0001613', 'grad_norm': '0.02561', 'learning_rate': '3.923e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '31.73', 'tokens/total': 9834592, 'tokens/trainable': 148535, 'epoch': '1.27'}
-
 63%|██████▎   | 325/512 [36:33<21:11,  6.80s/it]
 64%|██████▎   | 326/512 [36:40<20:58,  6.77s/it]
                                                 
{'loss': '0.0003376', 'grad_norm': '0.03319', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '34.07', 'tokens/total': 9864848, 'tokens/trainable': 148999, 'epoch': '1.273'}
-
 64%|██████▎   | 326/512 [36:40<20:58,  6.77s/it]
 64%|██████▍   | 327/512 [36:46<20:46,  6.74s/it]
                                                 
{'loss': '0.0007673', 'grad_norm': '0.07407', 'learning_rate': '3.869e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 9895168, 'tokens/trainable': 149431, 'epoch': '1.277'}
-
 64%|██████▍   | 327/512 [36:46<20:46,  6.74s/it]
 64%|██████▍   | 328/512 [36:53<20:35,  6.71s/it]
                                                 
{'loss': '0.0007135', 'grad_norm': '0.05904', 'learning_rate': '3.842e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '34.11', 'tokens/total': 9925696, 'tokens/trainable': 149886, 'epoch': '1.281'}
-
 64%|██████▍   | 328/512 [36:53<20:35,  6.71s/it]
 64%|██████▍   | 329/512 [37:00<20:21,  6.68s/it]
                                                 
{'loss': '7.994e-05', 'grad_norm': '0.00763', 'learning_rate': '3.815e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.98', 'tokens/total': 9955648, 'tokens/trainable': 150288, 'epoch': '1.285'}
-
 64%|██████▍   | 329/512 [37:00<20:21,  6.68s/it]
 64%|██████▍   | 330/512 [37:06<20:16,  6.69s/it]
                                                 
{'loss': '0.0001955', 'grad_norm': '0.02474', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 9985856, 'tokens/trainable': 150733, 'epoch': '1.289'}
-
 64%|██████▍   | 330/512 [37:06<20:16,  6.69s/it]
 65%|██████▍   | 331/512 [37:13<20:11,  6.69s/it]
                                                 
{'loss': '0.0001031', 'grad_norm': '0.006353', 'learning_rate': '3.761e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 10016208, 'tokens/trainable': 151207, 'epoch': '1.293'}
-
 65%|██████▍   | 331/512 [37:13<20:11,  6.69s/it]
 65%|██████▍   | 332/512 [37:20<20:06,  6.70s/it]
                                                 
{'loss': '0.001474', 'grad_norm': '0.162', 'learning_rate': '3.734e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.5', 'tokens/total': 10046512, 'tokens/trainable': 151694, 'epoch': '1.297'}
-
 65%|██████▍   | 332/512 [37:20<20:06,  6.70s/it]
 65%|██████▌   | 333/512 [37:27<20:03,  6.72s/it]
                                                 
{'loss': '0.000419', 'grad_norm': '0.04182', 'learning_rate': '3.708e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '38.27', 'tokens/total': 10076944, 'tokens/trainable': 152188, 'epoch': '1.301'}
-
 65%|██████▌   | 333/512 [37:27<20:03,  6.72s/it]
 65%|██████▌   | 334/512 [37:33<19:53,  6.71s/it]
                                                 
{'loss': '3.35e-05', 'grad_norm': '0.001125', 'learning_rate': '3.681e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.6', 'tokens/total': 10107296, 'tokens/trainable': 152596, 'epoch': '1.305'}
-
 65%|██████▌   | 334/512 [37:33<19:53,  6.71s/it]
 65%|██████▌   | 335/512 [37:40<19:46,  6.70s/it]
                                                 
{'loss': '4.685e-05', 'grad_norm': '0.003708', 'learning_rate': '3.655e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 10137568, 'tokens/trainable': 153052, 'epoch': '1.309'}
-
 65%|██████▌   | 335/512 [37:40<19:46,  6.70s/it]
 66%|██████▌   | 336/512 [37:47<19:40,  6.71s/it]
                                                 
{'loss': '2.363e-05', 'grad_norm': '0.001753', 'learning_rate': '3.628e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.19', 'tokens/total': 10167952, 'tokens/trainable': 153491, 'epoch': '1.312'}
-
 66%|██████▌   | 336/512 [37:47<19:40,  6.71s/it]
 66%|██████▌   | 337/512 [37:53<19:36,  6.72s/it]
                                                 
{'loss': '0.0006341', 'grad_norm': '0.06262', 'learning_rate': '3.602e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '40.28', 'tokens/total': 10198512, 'tokens/trainable': 153983, 'epoch': '1.316'}
-
 66%|██████▌   | 337/512 [37:53<19:36,  6.72s/it]
 66%|██████▌   | 338/512 [38:00<19:29,  6.72s/it]
                                                 
{'loss': '0.0001541', 'grad_norm': '0.01662', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 10228752, 'tokens/trainable': 154464, 'epoch': '1.32'}
-
 66%|██████▌   | 338/512 [38:00<19:29,  6.72s/it]
 66%|██████▌   | 339/512 [38:07<19:21,  6.71s/it]
                                                 
{'loss': '8.605e-06', 'grad_norm': '0.0004508', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.73', 'tokens/total': 10259104, 'tokens/trainable': 154924, 'epoch': '1.324'}
-
 66%|██████▌   | 339/512 [38:07<19:21,  6.71s/it]
 66%|██████▋   | 340/512 [38:14<19:12,  6.70s/it]
                                                 
{'loss': '4.251e-05', 'grad_norm': '0.002413', 'learning_rate': '3.523e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.76', 'tokens/total': 10289520, 'tokens/trainable': 155373, 'epoch': '1.328'}
-
 66%|██████▋   | 340/512 [38:14<19:12,  6.70s/it]
 67%|██████▋   | 341/512 [38:20<19:08,  6.72s/it]
                                                 
{'loss': '0.0004415', 'grad_norm': '0.0478', 'learning_rate': '3.497e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 10320016, 'tokens/trainable': 155873, 'epoch': '1.332'}
-
 67%|██████▋   | 341/512 [38:20<19:08,  6.72s/it]
 67%|██████▋   | 342/512 [38:27<18:58,  6.70s/it]
                                                 
{'loss': '3.104e-05', 'grad_norm': '0.001358', 'learning_rate': '3.471e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.34', 'tokens/total': 10350368, 'tokens/trainable': 156278, 'epoch': '1.336'}
-
 67%|██████▋   | 342/512 [38:27<18:58,  6.70s/it]
 67%|██████▋   | 343/512 [38:34<18:52,  6.70s/it]
                                                 
{'loss': '9.817e-05', 'grad_norm': '0.0139', 'learning_rate': '3.445e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.31', 'tokens/total': 10380624, 'tokens/trainable': 156733, 'epoch': '1.34'}
-
 67%|██████▋   | 343/512 [38:34<18:52,  6.70s/it]
 67%|██████▋   | 344/512 [38:40<18:48,  6.72s/it]
                                                 
{'loss': '4.33e-05', 'grad_norm': '0.003103', 'learning_rate': '3.42e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 10411024, 'tokens/trainable': 157203, 'epoch': '1.344'}
-
 67%|██████▋   | 344/512 [38:40<18:48,  6.72s/it]
 67%|██████▋   | 345/512 [38:47<18:42,  6.72s/it]
                                                 
{'loss': '0.000103', 'grad_norm': '0.009028', 'learning_rate': '3.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 10441248, 'tokens/trainable': 157707, 'epoch': '1.348'}
-
 67%|██████▋   | 345/512 [38:47<18:42,  6.72s/it]
 68%|██████▊   | 346/512 [38:54<18:39,  6.74s/it]
                                                 
{'loss': '4.62e-05', 'grad_norm': '0.003629', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '34.96', 'tokens/total': 10471712, 'tokens/trainable': 158180, 'epoch': '1.352'}
-
 68%|██████▊   | 346/512 [38:54<18:39,  6.74s/it]
 68%|██████▊   | 347/512 [39:01<18:31,  6.74s/it]
                                                 
{'loss': '0.0009502', 'grad_norm': '0.08972', 'learning_rate': '3.343e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 10502288, 'tokens/trainable': 158646, 'epoch': '1.355'}
-
 68%|██████▊   | 347/512 [39:01<18:31,  6.74s/it]
 68%|██████▊   | 348/512 [39:07<18:18,  6.70s/it]
                                                 
{'loss': '7.784e-05', 'grad_norm': '0.008904', 'learning_rate': '3.317e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '35.59', 'tokens/total': 10532304, 'tokens/trainable': 159115, 'epoch': '1.359'}
-
 68%|██████▊   | 348/512 [39:07<18:18,  6.70s/it]
 68%|██████▊   | 349/512 [39:14<18:10,  6.69s/it]
                                                 
{'loss': '0.0002163', 'grad_norm': '0.02332', 'learning_rate': '3.292e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 10562608, 'tokens/trainable': 159538, 'epoch': '1.363'}
-
 68%|██████▊   | 349/512 [39:14<18:10,  6.69s/it]
 68%|██████▊   | 350/512 [39:21<18:04,  6.69s/it]
                                                 
{'loss': '0.0004488', 'grad_norm': '0.04788', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 10592992, 'tokens/trainable': 160016, 'epoch': '1.367'}
-
 68%|██████▊   | 350/512 [39:21<18:04,  6.69s/it]
 69%|██████▊   | 351/512 [39:27<17:56,  6.69s/it]
                                                 
{'loss': '5.017e-05', 'grad_norm': '0.008666', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 10623328, 'tokens/trainable': 160475, 'epoch': '1.371'}
-
 69%|██████▊   | 351/512 [39:27<17:56,  6.69s/it]
 69%|██████▉   | 352/512 [39:34<17:50,  6.69s/it]
                                                 
{'loss': '0.0007272', 'grad_norm': '0.06453', 'learning_rate': '3.217e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 10653632, 'tokens/trainable': 160952, 'epoch': '1.375'}
-
 69%|██████▉   | 352/512 [39:34<17:50,  6.69s/it][2026-08-18 13:23:22,545] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
-
 69%|██████▉   | 353/512 [39:43<19:15,  7.27s/it]
                                                 
{'loss': '0.007551', 'grad_norm': '0.3706', 'learning_rate': '3.192e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 10684176, 'tokens/trainable': 161420, 'epoch': '1.379'}
-
 69%|██████▉   | 353/512 [39:43<19:15,  7.27s/it]
 69%|██████▉   | 354/512 [39:49<18:38,  7.08s/it]
                                                 
{'loss': '2.345e-05', 'grad_norm': '0.003138', 'learning_rate': '3.167e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 10714352, 'tokens/trainable': 161868, 'epoch': '1.383'}
-
 69%|██████▉   | 354/512 [39:49<18:38,  7.08s/it]
 69%|██████▉   | 355/512 [39:56<18:12,  6.96s/it]
                                                 
{'loss': '2.583e-05', 'grad_norm': '0.00214', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 10744464, 'tokens/trainable': 162318, 'epoch': '1.387'}
-
 69%|██████▉   | 355/512 [39:56<18:12,  6.96s/it]
 70%|██████▉   | 356/512 [40:03<17:55,  6.89s/it]
                                                 
{'loss': '3.729e-05', 'grad_norm': '0.002815', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 10774976, 'tokens/trainable': 162789, 'epoch': '1.391'}
-
 70%|██████▉   | 356/512 [40:03<17:55,  6.89s/it]
 70%|██████▉   | 357/512 [40:09<17:40,  6.84s/it]
                                                 
{'loss': '0.002785', 'grad_norm': '0.1642', 'learning_rate': '3.093e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 10805600, 'tokens/trainable': 163254, 'epoch': '1.395'}
-
 70%|██████▉   | 357/512 [40:09<17:40,  6.84s/it]
 70%|██████▉   | 358/512 [40:16<17:26,  6.79s/it]
                                                 
{'loss': '0.0002488', 'grad_norm': '0.06654', 'learning_rate': '3.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.41', 'tokens/total': 10835920, 'tokens/trainable': 163715, 'epoch': '1.398'}
-
 70%|██████▉   | 358/512 [40:16<17:26,  6.79s/it]
 70%|███████   | 359/512 [40:23<17:15,  6.77s/it]
                                                 
{'loss': '7.841e-05', 'grad_norm': '0.006618', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.4', 'tokens/total': 10866048, 'tokens/trainable': 164197, 'epoch': '1.402'}
-
 70%|███████   | 359/512 [40:23<17:15,  6.77s/it]
 70%|███████   | 360/512 [40:30<17:06,  6.75s/it]
                                                 
{'loss': '0.0001049', 'grad_norm': '0.01041', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.6', 'tokens/total': 10896288, 'tokens/trainable': 164661, 'epoch': '1.406'}
-
 70%|███████   | 360/512 [40:30<17:06,  6.75s/it]
 71%|███████   | 361/512 [40:36<16:56,  6.73s/it]
                                                 
{'loss': '4.034e-05', 'grad_norm': '0.002426', 'learning_rate': '2.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 10926816, 'tokens/trainable': 165111, 'epoch': '1.41'}
-
 71%|███████   | 361/512 [40:36<16:56,  6.73s/it]
 71%|███████   | 362/512 [40:42<16:24,  6.56s/it]
                                                 
{'loss': '0.0001499', 'grad_norm': '0.01567', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '30.34', 'tokens/total': 10954928, 'tokens/trainable': 165552, 'epoch': '1.414'}
-
 71%|███████   | 362/512 [40:42<16:24,  6.56s/it]
 71%|███████   | 363/512 [40:49<16:24,  6.61s/it]
                                                 
{'loss': '0.001498', 'grad_norm': '0.2696', 'learning_rate': '2.947e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 10985328, 'tokens/trainable': 166026, 'epoch': '1.418'}
-
 71%|███████   | 363/512 [40:49<16:24,  6.61s/it]
 71%|███████   | 364/512 [40:56<16:22,  6.64s/it]
                                                 
{'loss': '0.0003051', 'grad_norm': '0.03586', 'learning_rate': '2.924e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '37.86', 'tokens/total': 11015664, 'tokens/trainable': 166486, 'epoch': '1.422'}
-
 71%|███████   | 364/512 [40:56<16:22,  6.64s/it]
 71%|███████▏  | 365/512 [41:02<16:18,  6.66s/it]
                                                 
{'loss': '2.413e-05', 'grad_norm': '0.001931', 'learning_rate': '2.9e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 11046256, 'tokens/trainable': 166959, 'epoch': '1.426'}
-
 71%|███████▏  | 365/512 [41:03<16:18,  6.66s/it]
 71%|███████▏  | 366/512 [41:09<16:10,  6.65s/it]
                                                 
{'loss': '2.824e-05', 'grad_norm': '0.005941', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '33.64', 'tokens/total': 11076288, 'tokens/trainable': 167394, 'epoch': '1.43'}
-
 71%|███████▏  | 366/512 [41:09<16:10,  6.65s/it]
 72%|███████▏  | 367/512 [41:16<16:03,  6.64s/it]
                                                 
{'loss': '1.603e-05', 'grad_norm': '0.002126', 'learning_rate': '2.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '28.5', 'tokens/total': 11106416, 'tokens/trainable': 167832, 'epoch': '1.434'}
-
 72%|███████▏  | 367/512 [41:16<16:03,  6.64s/it]
 72%|███████▏  | 368/512 [41:22<15:58,  6.65s/it]
                                                 
{'loss': '0.000104', 'grad_norm': '0.251', 'learning_rate': '2.829e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.86', 'tokens/total': 11136848, 'tokens/trainable': 168242, 'epoch': '1.438'}
-
 72%|███████▏  | 368/512 [41:22<15:58,  6.65s/it]
 72%|███████▏  | 369/512 [41:29<15:52,  6.66s/it]
                                                 
{'loss': '2.642e-05', 'grad_norm': '0.001904', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.63', 'tokens/total': 11167376, 'tokens/trainable': 168715, 'epoch': '1.441'}
-
 72%|███████▏  | 369/512 [41:29<15:52,  6.66s/it]
 72%|███████▏  | 370/512 [41:36<15:48,  6.68s/it]
                                                 
{'loss': '2.997e-05', 'grad_norm': '0.002815', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.7', 'tokens/total': 11197776, 'tokens/trainable': 169147, 'epoch': '1.445'}
-
 72%|███████▏  | 370/512 [41:36<15:48,  6.68s/it]
 72%|███████▏  | 371/512 [41:43<15:41,  6.68s/it]
                                                 
{'loss': '1.342e-05', 'grad_norm': '0.001082', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 11227840, 'tokens/trainable': 169628, 'epoch': '1.449'}
-
 72%|███████▏  | 371/512 [41:43<15:41,  6.68s/it]
 73%|███████▎  | 372/512 [41:49<15:36,  6.69s/it]
                                                 
{'loss': '6.656e-05', 'grad_norm': '0.005368', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 11258176, 'tokens/trainable': 170085, 'epoch': '1.453'}
-
 73%|███████▎  | 372/512 [41:49<15:36,  6.69s/it]
 73%|███████▎  | 373/512 [41:56<15:28,  6.68s/it]
                                                 
{'loss': '0.0003471', 'grad_norm': '0.04757', 'learning_rate': '2.714e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 11288336, 'tokens/trainable': 170584, 'epoch': '1.457'}
-
 73%|███████▎  | 373/512 [41:56<15:28,  6.68s/it]
 73%|███████▎  | 374/512 [42:03<15:21,  6.68s/it]
                                                 
{'loss': '0.000288', 'grad_norm': '0.04089', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 11318848, 'tokens/trainable': 171058, 'epoch': '1.461'}
-
 73%|███████▎  | 374/512 [42:03<15:21,  6.68s/it]
 73%|███████▎  | 375/512 [42:09<15:16,  6.69s/it]
                                                 
{'loss': '2.691e-05', 'grad_norm': '0.003122', 'learning_rate': '2.668e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 11349168, 'tokens/trainable': 171518, 'epoch': '1.465'}
-
 73%|███████▎  | 375/512 [42:09<15:16,  6.69s/it]
 73%|███████▎  | 376/512 [42:16<15:08,  6.68s/it]
                                                 
{'loss': '8.402e-05', 'grad_norm': '0.01218', 'learning_rate': '2.646e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '40.48', 'tokens/total': 11379328, 'tokens/trainable': 172007, 'epoch': '1.469'}
-
 73%|███████▎  | 376/512 [42:16<15:08,  6.68s/it]
 74%|███████▎  | 377/512 [42:23<15:00,  6.67s/it]
                                                 
{'loss': '0.00133', 'grad_norm': '0.2414', 'learning_rate': '2.624e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.71', 'tokens/total': 11409568, 'tokens/trainable': 172468, 'epoch': '1.473'}
-
 74%|███████▎  | 377/512 [42:23<15:00,  6.67s/it]
 74%|███████▍  | 378/512 [42:29<14:54,  6.68s/it]
                                                 
{'loss': '0.0002016', 'grad_norm': '0.02171', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.64', 'tokens/total': 11439968, 'tokens/trainable': 172923, 'epoch': '1.477'}
-
 74%|███████▍  | 378/512 [42:29<14:54,  6.68s/it]
 74%|███████▍  | 379/512 [42:36<14:47,  6.67s/it]
                                                 
{'loss': '0.01576', 'grad_norm': '0.3541', 'learning_rate': '2.579e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 11470496, 'tokens/trainable': 173370, 'epoch': '1.48'}
-
 74%|███████▍  | 379/512 [42:36<14:47,  6.67s/it]
 74%|███████▍  | 380/512 [42:43<14:40,  6.67s/it]
                                                 
{'loss': '2.579e-05', 'grad_norm': '0.005766', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 11500720, 'tokens/trainable': 173837, 'epoch': '1.484'}
-
 74%|███████▍  | 380/512 [42:43<14:40,  6.67s/it]
 74%|███████▍  | 381/512 [42:49<14:33,  6.67s/it]
                                                 
{'loss': '5.822e-05', 'grad_norm': '0.005899', 'learning_rate': '2.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.56', 'tokens/total': 11531280, 'tokens/trainable': 174294, 'epoch': '1.488'}
-
 74%|███████▍  | 381/512 [42:49<14:33,  6.67s/it]
 75%|███████▍  | 382/512 [42:56<14:25,  6.66s/it]
                                                 
{'loss': '4.018e-05', 'grad_norm': '0.005452', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.78', 'tokens/total': 11561264, 'tokens/trainable': 174701, 'epoch': '1.492'}
-
 75%|███████▍  | 382/512 [42:56<14:25,  6.66s/it]
 75%|███████▍  | 383/512 [43:03<14:19,  6.66s/it]
                                                 
{'loss': '3.404e-05', 'grad_norm': '0.00383', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.04', 'tokens/total': 11591616, 'tokens/trainable': 175131, 'epoch': '1.496'}
-
 75%|███████▍  | 383/512 [43:03<14:19,  6.66s/it]
 75%|███████▌  | 384/512 [43:09<14:13,  6.67s/it]
                                                 
{'loss': '3.562e-05', 'grad_norm': '0.004227', 'learning_rate': '2.47e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 11622000, 'tokens/trainable': 175586, 'epoch': '1.5'}
-
 75%|███████▌  | 384/512 [43:09<14:13,  6.67s/it][2026-08-18 13:26:57,790] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
-
 75%|███████▌  | 385/512 [43:18<15:18,  7.24s/it]
                                                 
{'loss': '0.0001464', 'grad_norm': '0.01336', 'learning_rate': '2.449e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 11652352, 'tokens/trainable': 176026, 'epoch': '1.504'}
-
 75%|███████▌  | 385/512 [43:18<15:18,  7.24s/it]
 75%|███████▌  | 386/512 [43:25<14:51,  7.08s/it]
                                                 
{'loss': '8.266e-05', 'grad_norm': '0.008831', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 11682736, 'tokens/trainable': 176512, 'epoch': '1.508'}
-
 75%|███████▌  | 386/512 [43:25<14:51,  7.08s/it]
 76%|███████▌  | 387/512 [43:31<14:31,  6.97s/it]
                                                 
{'loss': '0.0001227', 'grad_norm': '0.009311', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 11713136, 'tokens/trainable': 176943, 'epoch': '1.512'}
-
 76%|███████▌  | 387/512 [43:31<14:31,  6.97s/it]
 76%|███████▌  | 388/512 [43:38<14:12,  6.87s/it]
                                                 
{'loss': '2.322e-05', 'grad_norm': '0.001782', 'learning_rate': '2.385e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.23', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 11743808, 'tokens/trainable': 177370, 'epoch': '1.516'}
-
 76%|███████▌  | 388/512 [43:38<14:12,  6.87s/it]
 76%|███████▌  | 389/512 [43:45<13:58,  6.82s/it]
                                                 
{'loss': '0.0001264', 'grad_norm': '0.0236', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.58', 'tokens/train_per_sec_per_gpu': '35.14', 'tokens/total': 11774144, 'tokens/trainable': 177851, 'epoch': '1.52'}
-
 76%|███████▌  | 389/512 [43:45<13:58,  6.82s/it]
 76%|███████▌  | 390/512 [43:51<13:47,  6.78s/it]
                                                 
{'loss': '0.0004641', 'grad_norm': '0.04384', 'learning_rate': '2.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.68', 'tokens/total': 11804272, 'tokens/trainable': 178310, 'epoch': '1.523'}
-
 76%|███████▌  | 390/512 [43:51<13:47,  6.78s/it]
 76%|███████▋  | 391/512 [43:57<13:19,  6.60s/it]
                                                 
{'loss': '0.0003324', 'grad_norm': '0.06193', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.8', 'tokens/total': 11832432, 'tokens/trainable': 178736, 'epoch': '1.527'}
-
 76%|███████▋  | 391/512 [43:57<13:19,  6.60s/it]
 77%|███████▋  | 392/512 [44:04<13:18,  6.65s/it]
                                                 
{'loss': '0.0005994', 'grad_norm': '0.04984', 'learning_rate': '2.303e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 11862960, 'tokens/trainable': 179233, 'epoch': '1.531'}
-
 77%|███████▋  | 392/512 [44:04<13:18,  6.65s/it]
 77%|███████▋  | 393/512 [44:11<13:14,  6.68s/it]
                                                 
{'loss': '0.0001238', 'grad_norm': '0.01204', 'learning_rate': '2.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.49', 'tokens/total': 11893232, 'tokens/trainable': 179730, 'epoch': '1.535'}
-
 77%|███████▋  | 393/512 [44:11<13:14,  6.68s/it]
 77%|███████▋  | 394/512 [44:18<13:09,  6.69s/it]
                                                 
{'loss': '0.0001039', 'grad_norm': '0.008993', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.81', 'tokens/total': 11923664, 'tokens/trainable': 180227, 'epoch': '1.539'}
-
 77%|███████▋  | 394/512 [44:18<13:09,  6.69s/it]
 77%|███████▋  | 395/512 [44:24<13:04,  6.70s/it]
                                                 
{'loss': '0.003169', 'grad_norm': '0.2463', 'learning_rate': '2.242e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.85', 'tokens/total': 11954032, 'tokens/trainable': 180687, 'epoch': '1.543'}
-
 77%|███████▋  | 395/512 [44:24<13:04,  6.70s/it]
 77%|███████▋  | 396/512 [44:31<12:57,  6.71s/it]
                                                 
{'loss': '0.0005169', 'grad_norm': '0.06872', 'learning_rate': '2.222e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 11984512, 'tokens/trainable': 181141, 'epoch': '1.547'}
-
 77%|███████▋  | 396/512 [44:31<12:57,  6.71s/it]
 78%|███████▊  | 397/512 [44:38<12:53,  6.72s/it]
                                                 
{'loss': '0.000175', 'grad_norm': '0.01167', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 12014880, 'tokens/trainable': 181648, 'epoch': '1.551'}
-
 78%|███████▊  | 397/512 [44:38<12:53,  6.72s/it]
 78%|███████▊  | 398/512 [44:45<12:43,  6.70s/it]
                                                 
{'loss': '0.001466', 'grad_norm': '0.1461', 'learning_rate': '2.183e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 12044928, 'tokens/trainable': 182109, 'epoch': '1.555'}
-
 78%|███████▊  | 398/512 [44:45<12:43,  6.70s/it]
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
                                                 
{'loss': '7.929e-05', 'grad_norm': '0.01636', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.5', 'tokens/total': 12075344, 'tokens/trainable': 182559, 'epoch': '1.559'}
-
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
 78%|███████▊  | 400/512 [44:58<12:27,  6.68s/it]
                                                 
{'loss': '1.526e-05', 'grad_norm': '0.001048', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 12105616, 'tokens/trainable': 182985, 'epoch': '1.562'}
-
 78%|███████▊  | 400/512 [44:58<12:27,  6.68s/it]
 78%|███████▊  | 401/512 [45:05<12:20,  6.67s/it]
                                                 
{'loss': '1.217e-05', 'grad_norm': '0.0005901', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 12136080, 'tokens/trainable': 183440, 'epoch': '1.566'}
-
 78%|███████▊  | 401/512 [45:05<12:20,  6.67s/it]
 79%|███████▊  | 402/512 [45:11<12:14,  6.68s/it]
                                                 
{'loss': '0.000197', 'grad_norm': '0.01319', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '34.52', 'tokens/total': 12166240, 'tokens/trainable': 183917, 'epoch': '1.57'}
-
 79%|███████▊  | 402/512 [45:11<12:14,  6.68s/it]
 79%|███████▊  | 403/512 [45:18<12:07,  6.67s/it]
                                                 
{'loss': '0.0002239', 'grad_norm': '0.02566', 'learning_rate': '2.086e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '30.39', 'tokens/total': 12196544, 'tokens/trainable': 184348, 'epoch': '1.574'}
-
 79%|███████▊  | 403/512 [45:18<12:07,  6.67s/it]
 79%|███████▉  | 404/512 [45:25<12:00,  6.67s/it]
                                                 
{'loss': '8.35e-05', 'grad_norm': '0.008275', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 12226784, 'tokens/trainable': 184786, 'epoch': '1.578'}
-
 79%|███████▉  | 404/512 [45:25<12:00,  6.67s/it]
 79%|███████▉  | 405/512 [45:31<11:54,  6.68s/it]
                                                 
{'loss': '0.0001853', 'grad_norm': '0.03818', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 12257264, 'tokens/trainable': 185249, 'epoch': '1.582'}
-
 79%|███████▉  | 405/512 [45:31<11:54,  6.68s/it]
 79%|███████▉  | 406/512 [45:38<11:48,  6.68s/it]
                                                 
{'loss': '0.0001144', 'grad_norm': '0.007673', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 12287584, 'tokens/trainable': 185680, 'epoch': '1.586'}
-
 79%|███████▉  | 406/512 [45:38<11:48,  6.68s/it]
 79%|███████▉  | 407/512 [45:45<11:40,  6.67s/it]
                                                 
{'loss': '4.649e-05', 'grad_norm': '0.00316', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.72', 'tokens/total': 12317984, 'tokens/trainable': 186100, 'epoch': '1.59'}
-
 79%|███████▉  | 407/512 [45:45<11:40,  6.67s/it]
 80%|███████▉  | 408/512 [45:51<11:33,  6.67s/it]
                                                 
{'loss': '0.00274', 'grad_norm': '0.3274', 'learning_rate': '1.993e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.82', 'tokens/total': 12348144, 'tokens/trainable': 186565, 'epoch': '1.594'}
-
 80%|███████▉  | 408/512 [45:51<11:33,  6.67s/it]
 80%|███████▉  | 409/512 [45:58<11:26,  6.67s/it]
                                                 
{'loss': '1.297e-05', 'grad_norm': '0.0006676', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 12378544, 'tokens/trainable': 187017, 'epoch': '1.598'}
-
 80%|███████▉  | 409/512 [45:58<11:26,  6.67s/it]
 80%|████████  | 410/512 [46:05<11:21,  6.68s/it]
                                                 
{'loss': '3.858e-05', 'grad_norm': '0.003977', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 12408784, 'tokens/trainable': 187468, 'epoch': '1.602'}
-
 80%|████████  | 410/512 [46:05<11:21,  6.68s/it]
 80%|████████  | 411/512 [46:11<11:13,  6.67s/it]
                                                 
{'loss': '2.431e-05', 'grad_norm': '0.002458', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.61', 'tokens/total': 12438832, 'tokens/trainable': 187961, 'epoch': '1.605'}
-
 80%|████████  | 411/512 [46:11<11:13,  6.67s/it]
 80%|████████  | 412/512 [46:18<11:07,  6.68s/it]
                                                 
{'loss': '4.12e-05', 'grad_norm': '0.003755', 'learning_rate': '1.922e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 12469296, 'tokens/trainable': 188447, 'epoch': '1.609'}
-
 80%|████████  | 412/512 [46:18<11:07,  6.68s/it]
 81%|████████  | 413/512 [46:25<11:00,  6.67s/it]
                                                 
{'loss': '0.004735', 'grad_norm': '0.3583', 'learning_rate': '1.904e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 12499392, 'tokens/trainable': 188903, 'epoch': '1.613'}
-
 81%|████████  | 413/512 [46:25<11:00,  6.67s/it]
 81%|████████  | 414/512 [46:31<10:57,  6.70s/it]
                                                 
{'loss': '0.000194', 'grad_norm': '0.04949', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 12529952, 'tokens/trainable': 189398, 'epoch': '1.617'}
-
 81%|████████  | 414/512 [46:31<10:57,  6.70s/it]
 81%|████████  | 415/512 [46:38<10:50,  6.70s/it]
                                                 
{'loss': '2.004e-05', 'grad_norm': '0.003045', 'learning_rate': '1.87e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 12560240, 'tokens/trainable': 189850, 'epoch': '1.621'}
-
 81%|████████  | 415/512 [46:38<10:50,  6.70s/it]
 81%|████████▏ | 416/512 [46:45<10:43,  6.71s/it]
                                                 
{'loss': '0.00162', 'grad_norm': '0.2542', 'learning_rate': '1.853e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '32.64', 'tokens/total': 12590736, 'tokens/trainable': 190306, 'epoch': '1.625'}
-
 81%|████████▏ | 416/512 [46:45<10:43,  6.71s/it][2026-08-18 13:30:33,327] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
-
 81%|████████▏ | 417/512 [46:53<11:28,  7.25s/it]
                                                 
{'loss': '4.322e-05', 'grad_norm': '0.006711', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 12621168, 'tokens/trainable': 190746, 'epoch': '1.629'}
-
 81%|████████▏ | 417/512 [46:53<11:28,  7.25s/it]
 82%|████████▏ | 418/512 [47:00<11:04,  7.07s/it]
                                                 
{'loss': '4.74e-05', 'grad_norm': '0.004932', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '31.07', 'tokens/total': 12651456, 'tokens/trainable': 191156, 'epoch': '1.633'}
-
 82%|████████▏ | 418/512 [47:00<11:04,  7.07s/it]
 82%|████████▏ | 419/512 [47:07<10:46,  6.95s/it]
                                                 
{'loss': '7.545e-05', 'grad_norm': '0.007677', 'learning_rate': '1.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '29.47', 'tokens/total': 12681760, 'tokens/trainable': 191573, 'epoch': '1.637'}
-
 82%|████████▏ | 419/512 [47:07<10:46,  6.95s/it]
 82%|████████▏ | 420/512 [47:13<10:32,  6.87s/it]
                                                 
{'loss': '0.0001045', 'grad_norm': '0.01545', 'learning_rate': '1.786e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 12712144, 'tokens/trainable': 192014, 'epoch': '1.641'}
-
 82%|████████▏ | 420/512 [47:13<10:32,  6.87s/it]
 82%|████████▏ | 421/512 [47:20<10:19,  6.81s/it]
                                                 
{'loss': '0.002415', 'grad_norm': '0.1688', 'learning_rate': '1.77e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.82', 'tokens/total': 12742400, 'tokens/trainable': 192453, 'epoch': '1.645'}
-
 82%|████████▏ | 421/512 [47:20<10:19,  6.81s/it]
 82%|████████▏ | 422/512 [47:27<10:08,  6.76s/it]
                                                 
{'loss': '3.337e-05', 'grad_norm': '0.002461', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 12772288, 'tokens/trainable': 192886, 'epoch': '1.648'}
-
 82%|████████▏ | 422/512 [47:27<10:08,  6.76s/it]
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
                                                 
{'loss': '8.696e-06', 'grad_norm': '0.0005388', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '29.85', 'tokens/total': 12802880, 'tokens/trainable': 193323, 'epoch': '1.652'}
-
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
 83%|████████▎ | 424/512 [47:40<09:56,  6.78s/it]
                                                 
{'loss': '4.861e-05', 'grad_norm': '0.00501', 'learning_rate': '1.722e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.6', 'tokens/total': 12833456, 'tokens/trainable': 193753, 'epoch': '1.656'}
-
 83%|████████▎ | 424/512 [47:40<09:56,  6.78s/it]
 83%|████████▎ | 425/512 [47:47<09:47,  6.75s/it]
                                                 
{'loss': '3.262e-05', 'grad_norm': '0.001982', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 12863952, 'tokens/trainable': 194240, 'epoch': '1.66'}
-
 83%|████████▎ | 425/512 [47:47<09:47,  6.75s/it]
 83%|████████▎ | 426/512 [47:54<09:38,  6.73s/it]
                                                 
{'loss': '0.0005229', 'grad_norm': '0.05306', 'learning_rate': '1.69e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 12894352, 'tokens/trainable': 194680, 'epoch': '1.664'}
-
 83%|████████▎ | 426/512 [47:54<09:38,  6.73s/it]
 83%|████████▎ | 427/512 [48:00<09:31,  6.72s/it]
                                                 
{'loss': '1.543e-05', 'grad_norm': '0.001265', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '38.73', 'tokens/total': 12924688, 'tokens/trainable': 195186, 'epoch': '1.668'}
-
 83%|████████▎ | 427/512 [48:00<09:31,  6.72s/it]
 84%|████████▎ | 428/512 [48:07<09:25,  6.74s/it]
                                                 
{'loss': '2.778e-05', 'grad_norm': '0.001626', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.51', 'tokens/total': 12955072, 'tokens/trainable': 195693, 'epoch': '1.672'}
-
 84%|████████▎ | 428/512 [48:07<09:25,  6.74s/it]
 84%|████████▍ | 429/512 [48:14<09:17,  6.71s/it]
                                                 
{'loss': '6.972e-05', 'grad_norm': '0.01333', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.93', 'tokens/total': 12985328, 'tokens/trainable': 196130, 'epoch': '1.676'}
-
 84%|████████▍ | 429/512 [48:14<09:17,  6.71s/it]
 84%|████████▍ | 430/512 [48:20<09:08,  6.69s/it]
                                                 
{'loss': '4.5e-05', 'grad_norm': '0.003462', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.58', 'tokens/total': 13015600, 'tokens/trainable': 196530, 'epoch': '1.68'}
-
 84%|████████▍ | 430/512 [48:20<09:08,  6.69s/it]
 84%|████████▍ | 431/512 [48:27<09:01,  6.68s/it]
                                                 
{'loss': '0.0002949', 'grad_norm': '0.03424', 'learning_rate': '1.615e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.88', 'tokens/total': 13045728, 'tokens/trainable': 196974, 'epoch': '1.684'}
-
 84%|████████▍ | 431/512 [48:27<09:01,  6.68s/it]
 84%|████████▍ | 432/512 [48:34<08:54,  6.69s/it]
                                                 
{'loss': '7.788e-05', 'grad_norm': '0.00645', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 13075920, 'tokens/trainable': 197454, 'epoch': '1.688'}
-
 84%|████████▍ | 432/512 [48:34<08:54,  6.69s/it]
 85%|████████▍ | 433/512 [48:40<08:48,  6.69s/it]
                                                 
{'loss': '6.603e-06', 'grad_norm': '0.0002191', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '31.4', 'tokens/total': 13106128, 'tokens/trainable': 197902, 'epoch': '1.691'}
-
 85%|████████▍ | 433/512 [48:40<08:48,  6.69s/it]
 85%|████████▍ | 434/512 [48:47<08:40,  6.67s/it]
                                                 
{'loss': '6.426e-06', 'grad_norm': '0.0002612', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '36', 'tokens/total': 13136368, 'tokens/trainable': 198374, 'epoch': '1.695'}
-
 85%|████████▍ | 434/512 [48:47<08:40,  6.67s/it]
 85%|████████▍ | 435/512 [48:54<08:34,  6.68s/it]
                                                 
{'loss': '2.549e-05', 'grad_norm': '0.00164', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.54', 'tokens/total': 13166704, 'tokens/trainable': 198846, 'epoch': '1.699'}
-
 85%|████████▍ | 435/512 [48:54<08:34,  6.68s/it]
 85%|████████▌ | 436/512 [49:00<08:29,  6.70s/it]
                                                 
{'loss': '0.0002141', 'grad_norm': '0.01888', 'learning_rate': '1.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 13197344, 'tokens/trainable': 199320, 'epoch': '1.703'}
-
 85%|████████▌ | 436/512 [49:00<08:29,  6.70s/it]
 85%|████████▌ | 437/512 [49:07<08:21,  6.69s/it]
                                                 
{'loss': '2.316e-05', 'grad_norm': '0.001229', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.43', 'tokens/total': 13227664, 'tokens/trainable': 199777, 'epoch': '1.707'}
-
 85%|████████▌ | 437/512 [49:07<08:21,  6.69s/it]
 86%|████████▌ | 438/512 [49:14<08:13,  6.67s/it]
                                                 
{'loss': '4.594e-05', 'grad_norm': '0.005586', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '30.82', 'tokens/total': 13257632, 'tokens/trainable': 200194, 'epoch': '1.711'}
-
 86%|████████▌ | 438/512 [49:14<08:13,  6.67s/it]
 86%|████████▌ | 439/512 [49:21<08:09,  6.71s/it]
                                                 
{'loss': '2.285e-05', 'grad_norm': '0.001554', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 13288240, 'tokens/trainable': 200663, 'epoch': '1.715'}
-
 86%|████████▌ | 439/512 [49:21<08:09,  6.71s/it]
 86%|████████▌ | 440/512 [49:27<08:05,  6.74s/it]
                                                 
{'loss': '4.093e-05', 'grad_norm': '0.002979', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.79', 'tokens/total': 13318992, 'tokens/trainable': 201121, 'epoch': '1.719'}
-
 86%|████████▌ | 440/512 [49:27<08:05,  6.74s/it]
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
                                                 
{'loss': '7.297e-06', 'grad_norm': '0.001739', 'learning_rate': '1.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.74', 'tokens/total': 13349392, 'tokens/trainable': 201598, 'epoch': '1.723'}
-
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
 86%|████████▋ | 442/512 [49:41<07:50,  6.72s/it]
                                                 
{'loss': '1.716e-05', 'grad_norm': '0.0009943', 'learning_rate': '1.464e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 13379648, 'tokens/trainable': 202068, 'epoch': '1.727'}
-
 86%|████████▋ | 442/512 [49:41<07:50,  6.72s/it]
 87%|████████▋ | 443/512 [49:47<07:41,  6.70s/it]
                                                 
{'loss': '2.382e-05', 'grad_norm': '0.001147', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 13409936, 'tokens/trainable': 202524, 'epoch': '1.73'}
-
 87%|████████▋ | 443/512 [49:47<07:41,  6.70s/it]
 87%|████████▋ | 444/512 [49:54<07:35,  6.69s/it]
                                                 
{'loss': '6.557e-05', 'grad_norm': '0.006799', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 13440288, 'tokens/trainable': 202996, 'epoch': '1.734'}
-
 87%|████████▋ | 444/512 [49:54<07:35,  6.69s/it]
 87%|████████▋ | 445/512 [50:00<07:19,  6.56s/it]
                                                 
{'loss': '1.839e-05', 'grad_norm': '0.001165', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.14', 'tokens/total': 13468880, 'tokens/trainable': 203470, 'epoch': '1.738'}
-
 87%|████████▋ | 445/512 [50:00<07:19,  6.56s/it]
 87%|████████▋ | 446/512 [50:07<07:16,  6.61s/it]
                                                 
{'loss': '0.000164', 'grad_norm': '0.02089', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 13499184, 'tokens/trainable': 203919, 'epoch': '1.742'}
-
 87%|████████▋ | 446/512 [50:07<07:16,  6.61s/it]
 87%|████████▋ | 447/512 [50:14<07:10,  6.62s/it]
                                                 
{'loss': '0.001068', 'grad_norm': '0.09616', 'learning_rate': '1.402e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.3', 'tokens/total': 13529632, 'tokens/trainable': 204376, 'epoch': '1.746'}
-
 87%|████████▋ | 447/512 [50:14<07:10,  6.62s/it]
 88%|████████▊ | 448/512 [50:20<07:05,  6.64s/it]
                                                 
{'loss': '0.001196', 'grad_norm': '0.09744', 'learning_rate': '1.39e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32', 'tokens/total': 13560080, 'tokens/trainable': 204821, 'epoch': '1.75'}
-
 88%|████████▊ | 448/512 [50:20<07:05,  6.64s/it][2026-08-18 13:34:08,932] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
-
 88%|████████▊ | 449/512 [50:29<07:36,  7.24s/it]
                                                 
{'loss': '8.02e-06', 'grad_norm': '0.0007811', 'learning_rate': '1.378e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.63', 'tokens/total': 13590496, 'tokens/trainable': 205271, 'epoch': '1.754'}
-
 88%|████████▊ | 449/512 [50:29<07:36,  7.24s/it]
 88%|████████▊ | 450/512 [50:36<07:18,  7.07s/it]
                                                 
{'loss': '0.0001024', 'grad_norm': '0.03409', 'learning_rate': '1.367e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '30.95', 'tokens/total': 13620960, 'tokens/trainable': 205723, 'epoch': '1.758'}
-
 88%|████████▊ | 450/512 [50:36<07:18,  7.07s/it]
 88%|████████▊ | 451/512 [50:42<07:04,  6.96s/it]
                                                 
{'loss': '0.0002193', 'grad_norm': '0.02931', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.42', 'tokens/total': 13651344, 'tokens/trainable': 206169, 'epoch': '1.762'}
-
 88%|████████▊ | 451/512 [50:42<07:04,  6.96s/it]
 88%|████████▊ | 452/512 [50:49<06:54,  6.90s/it]
                                                 
{'loss': '8e-05', 'grad_norm': '0.01096', 'learning_rate': '1.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 13681792, 'tokens/trainable': 206592, 'epoch': '1.766'}
-
 88%|████████▊ | 452/512 [50:49<06:54,  6.90s/it]
 88%|████████▊ | 453/512 [50:56<06:43,  6.83s/it]
                                                 
{'loss': '3.79e-05', 'grad_norm': '0.004453', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 13712256, 'tokens/trainable': 207055, 'epoch': '1.77'}
-
 88%|████████▊ | 453/512 [50:56<06:43,  6.83s/it]
 89%|████████▊ | 454/512 [51:02<06:31,  6.76s/it]
                                                 
{'loss': '1.224e-05', 'grad_norm': '0.000565', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.65', 'tokens/total': 13742128, 'tokens/trainable': 207494, 'epoch': '1.773'}
-
 89%|████████▊ | 454/512 [51:02<06:31,  6.76s/it]
 89%|████████▉ | 455/512 [51:09<06:23,  6.72s/it]
                                                 
{'loss': '1.357e-05', 'grad_norm': '0.0007141', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 13772160, 'tokens/trainable': 207956, 'epoch': '1.777'}
-
 89%|████████▉ | 455/512 [51:09<06:23,  6.72s/it]
 89%|████████▉ | 456/512 [51:16<06:15,  6.71s/it]
                                                 
{'loss': '1.166e-05', 'grad_norm': '0.0008097', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 13802448, 'tokens/trainable': 208408, 'epoch': '1.781'}
-
 89%|████████▉ | 456/512 [51:16<06:15,  6.71s/it]
 89%|████████▉ | 457/512 [51:22<06:08,  6.70s/it]
                                                 
{'loss': '9.201e-06', 'grad_norm': '0.001265', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '36', 'tokens/total': 13832880, 'tokens/trainable': 208889, 'epoch': '1.785'}
-
 89%|████████▉ | 457/512 [51:22<06:08,  6.70s/it]
 89%|████████▉ | 458/512 [51:29<06:00,  6.68s/it]
                                                 
{'loss': '0.0001209', 'grad_norm': '0.01498', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.08', 'tokens/total': 13862880, 'tokens/trainable': 209336, 'epoch': '1.789'}
-
 89%|████████▉ | 458/512 [51:29<06:00,  6.68s/it]
 90%|████████▉ | 459/512 [51:36<05:54,  6.68s/it]
                                                 
{'loss': '3.241e-05', 'grad_norm': '0.004875', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.26', 'tokens/total': 13893440, 'tokens/trainable': 209797, 'epoch': '1.793'}
-
 90%|████████▉ | 459/512 [51:36<05:54,  6.68s/it]
 90%|████████▉ | 460/512 [51:42<05:47,  6.68s/it]
                                                 
{'loss': '1.98e-05', 'grad_norm': '0.0009628', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.12', 'tokens/total': 13923824, 'tokens/trainable': 210240, 'epoch': '1.797'}
-
 90%|████████▉ | 460/512 [51:42<05:47,  6.68s/it]
 90%|█████████ | 461/512 [51:49<05:41,  6.69s/it]
                                                 
{'loss': '0.0001068', 'grad_norm': '0.01991', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 13954208, 'tokens/trainable': 210702, 'epoch': '1.801'}
-
 90%|█████████ | 461/512 [51:49<05:41,  6.69s/it]
 90%|█████████ | 462/512 [51:56<05:34,  6.69s/it]
                                                 
{'loss': '1.108e-05', 'grad_norm': '0.000537', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 13984784, 'tokens/trainable': 211140, 'epoch': '1.805'}
-
 90%|█████████ | 462/512 [51:56<05:34,  6.69s/it]
 90%|█████████ | 463/512 [52:02<05:27,  6.68s/it]
                                                 
{'loss': '1.551e-05', 'grad_norm': '0.0008154', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.6', 'tokens/total': 14014896, 'tokens/trainable': 211575, 'epoch': '1.809'}
-
 90%|█████████ | 463/512 [52:02<05:27,  6.68s/it]
 91%|█████████ | 464/512 [52:09<05:21,  6.69s/it]
                                                 
{'loss': '0.0001314', 'grad_norm': '0.02876', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '39.03', 'tokens/total': 14045312, 'tokens/trainable': 212083, 'epoch': '1.812'}
-
 91%|█████████ | 464/512 [52:09<05:21,  6.69s/it]
 91%|█████████ | 465/512 [52:16<05:15,  6.71s/it]
                                                 
{'loss': '0.000297', 'grad_norm': '0.04559', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 14075840, 'tokens/trainable': 212558, 'epoch': '1.816'}
-
 91%|█████████ | 465/512 [52:16<05:15,  6.71s/it]
 91%|█████████ | 466/512 [52:23<05:07,  6.69s/it]
                                                 
{'loss': '9.938e-06', 'grad_norm': '0.0004123', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 14106016, 'tokens/trainable': 213014, 'epoch': '1.82'}
-
 91%|█████████ | 466/512 [52:23<05:07,  6.69s/it]
 91%|█████████ | 467/512 [52:29<05:01,  6.70s/it]
                                                 
{'loss': '1.558e-05', 'grad_norm': '0.001011', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.99', 'tokens/train_per_sec_per_gpu': '34.03', 'tokens/total': 14136512, 'tokens/trainable': 213465, 'epoch': '1.824'}
-
 91%|█████████ | 467/512 [52:29<05:01,  6.70s/it]
 91%|█████████▏| 468/512 [52:36<04:54,  6.70s/it]
                                                 
{'loss': '3.261e-05', 'grad_norm': '0.00431', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.62', 'tokens/total': 14166832, 'tokens/trainable': 213940, 'epoch': '1.828'}
-
 91%|█████████▏| 468/512 [52:36<04:54,  6.70s/it]
 92%|█████████▏| 469/512 [52:43<04:49,  6.72s/it]
                                                 
{'loss': '2.145e-05', 'grad_norm': '0.005317', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 14197360, 'tokens/trainable': 214415, 'epoch': '1.832'}
-
 92%|█████████▏| 469/512 [52:43<04:49,  6.72s/it]
 92%|█████████▏| 470/512 [52:50<04:42,  6.72s/it]
                                                 
{'loss': '5.694e-06', 'grad_norm': '0.0003265', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 14227600, 'tokens/trainable': 214880, 'epoch': '1.836'}
-
 92%|█████████▏| 470/512 [52:50<04:42,  6.72s/it]
 92%|█████████▏| 471/512 [52:56<04:35,  6.71s/it]
                                                 
{'loss': '0.0001062', 'grad_norm': '0.01372', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 14257888, 'tokens/trainable': 215334, 'epoch': '1.84'}
-
 92%|█████████▏| 471/512 [52:56<04:35,  6.71s/it]
 92%|█████████▏| 472/512 [53:03<04:28,  6.71s/it]
                                                 
{'loss': '4.587e-06', 'grad_norm': '0.0002091', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 14288208, 'tokens/trainable': 215822, 'epoch': '1.844'}
-
 92%|█████████▏| 472/512 [53:03<04:28,  6.71s/it]
 92%|█████████▏| 473/512 [53:10<04:21,  6.70s/it]
                                                 
{'loss': '0.0006915', 'grad_norm': '0.1067', 'learning_rate': '1.149e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '30.86', 'tokens/total': 14318448, 'tokens/trainable': 216239, 'epoch': '1.848'}
-
 92%|█████████▏| 473/512 [53:10<04:21,  6.70s/it]
 93%|█████████▎| 474/512 [53:16<04:14,  6.69s/it]
                                                 
{'loss': '6.862e-05', 'grad_norm': '0.01328', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.37', 'tokens/total': 14348912, 'tokens/trainable': 216691, 'epoch': '1.852'}
-
 93%|█████████▎| 474/512 [53:16<04:14,  6.69s/it]
 93%|█████████▎| 475/512 [53:23<04:07,  6.68s/it]
                                                 
{'loss': '8.334e-05', 'grad_norm': '0.007407', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 14379440, 'tokens/trainable': 217162, 'epoch': '1.855'}
-
 93%|█████████▎| 475/512 [53:23<04:07,  6.68s/it]
 93%|█████████▎| 476/512 [53:30<04:01,  6.71s/it]
                                                 
{'loss': '1.804e-05', 'grad_norm': '0.001777', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 14409840, 'tokens/trainable': 217638, 'epoch': '1.859'}
-
 93%|█████████▎| 476/512 [53:30<04:01,  6.71s/it]
 93%|█████████▎| 477/512 [53:36<03:54,  6.71s/it]
                                                 
{'loss': '9.578e-06', 'grad_norm': '0.0004483', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 14440256, 'tokens/trainable': 218053, 'epoch': '1.863'}
-
 93%|█████████▎| 477/512 [53:36<03:54,  6.71s/it]
 93%|█████████▎| 478/512 [53:43<03:47,  6.70s/it]
                                                 
{'loss': '1.095e-05', 'grad_norm': '0.000956', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 14470704, 'tokens/trainable': 218503, 'epoch': '1.867'}
-
 93%|█████████▎| 478/512 [53:43<03:47,  6.70s/it]
 94%|█████████▎| 479/512 [53:50<03:40,  6.70s/it]
                                                 
{'loss': '0.00123', 'grad_norm': '0.2087', 'learning_rate': '1.108e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.19', 'tokens/total': 14501072, 'tokens/trainable': 218964, 'epoch': '1.871'}
-
 94%|█████████▎| 479/512 [53:50<03:40,  6.70s/it]
 94%|█████████▍| 480/512 [53:56<03:33,  6.68s/it]
                                                 
{'loss': '0.0001162', 'grad_norm': '0.01922', 'learning_rate': '1.102e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 14531200, 'tokens/trainable': 219412, 'epoch': '1.875'}
-
 94%|█████████▍| 480/512 [53:56<03:33,  6.68s/it][2026-08-18 13:37:44,940] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
-
 94%|█████████▍| 481/512 [54:05<03:44,  7.24s/it]
                                                 
{'loss': '2.362e-05', 'grad_norm': '0.002075', 'learning_rate': '1.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '38.29', 'tokens/total': 14561152, 'tokens/trainable': 219886, 'epoch': '1.879'}
-
 94%|█████████▍| 481/512 [54:05<03:44,  7.24s/it]
 94%|█████████▍| 482/512 [54:12<03:32,  7.07s/it]
                                                 
{'loss': '5.528e-05', 'grad_norm': '0.01115', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 14591264, 'tokens/trainable': 220343, 'epoch': '1.883'}
-
 94%|█████████▍| 482/512 [54:12<03:32,  7.07s/it]
 94%|█████████▍| 483/512 [54:18<03:21,  6.96s/it]
                                                 
{'loss': '6.712e-05', 'grad_norm': '0.01091', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 14621568, 'tokens/trainable': 220825, 'epoch': '1.887'}
-
 94%|█████████▍| 483/512 [54:18<03:21,  6.96s/it]
 95%|█████████▍| 484/512 [54:25<03:12,  6.88s/it]
                                                 
{'loss': '0.002836', 'grad_norm': '0.3236', 'learning_rate': '1.079e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '32.52', 'tokens/total': 14652064, 'tokens/trainable': 221282, 'epoch': '1.891'}
-
 95%|█████████▍| 484/512 [54:25<03:12,  6.88s/it]
 95%|█████████▍| 485/512 [54:32<03:04,  6.83s/it]
                                                 
{'loss': '5.301e-06', 'grad_norm': '0.0002806', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.86', 'tokens/total': 14682608, 'tokens/trainable': 221780, 'epoch': '1.895'}
-
 95%|█████████▍| 485/512 [54:32<03:04,  6.83s/it]
 95%|█████████▍| 486/512 [54:39<02:57,  6.81s/it]
                                                 
{'loss': '3.665e-05', 'grad_norm': '0.009112', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 14713168, 'tokens/trainable': 222210, 'epoch': '1.898'}
-
 95%|█████████▍| 486/512 [54:39<02:57,  6.81s/it]
 95%|█████████▌| 487/512 [54:45<02:49,  6.77s/it]
                                                 
{'loss': '0.0002056', 'grad_norm': '0.04485', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.33', 'tokens/total': 14743536, 'tokens/trainable': 222646, 'epoch': '1.902'}
-
 95%|█████████▌| 487/512 [54:45<02:49,  6.77s/it]
 95%|█████████▌| 488/512 [54:52<02:41,  6.74s/it]
                                                 
{'loss': '0.008499', 'grad_norm': '0.2785', 'learning_rate': '1.058e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 14773728, 'tokens/trainable': 223066, 'epoch': '1.906'}
-
 95%|█████████▌| 488/512 [54:52<02:41,  6.74s/it]
 96%|█████████▌| 489/512 [54:59<02:34,  6.72s/it]
                                                 
{'loss': '9.457e-06', 'grad_norm': '0.0006374', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 14803744, 'tokens/trainable': 223512, 'epoch': '1.91'}
-
 96%|█████████▌| 489/512 [54:59<02:34,  6.72s/it]
 96%|█████████▌| 490/512 [55:05<02:24,  6.56s/it]
                                                 
{'loss': '6.928e-06', 'grad_norm': '0.000387', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.96', 'tokens/total': 14832112, 'tokens/trainable': 223956, 'epoch': '1.914'}
-
 96%|█████████▌| 490/512 [55:05<02:24,  6.56s/it]
 96%|█████████▌| 491/512 [55:11<02:15,  6.45s/it]
                                                 
{'loss': '2.272e-05', 'grad_norm': '0.001532', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 14860560, 'tokens/trainable': 224416, 'epoch': '1.918'}
-
 96%|█████████▌| 491/512 [55:11<02:15,  6.45s/it]
 96%|█████████▌| 492/512 [55:18<02:10,  6.52s/it]
                                                 
{'loss': '0.0001193', 'grad_norm': '0.01609', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 14890848, 'tokens/trainable': 224883, 'epoch': '1.922'}
-
 96%|█████████▌| 492/512 [55:18<02:10,  6.52s/it]
 96%|█████████▋| 493/512 [55:24<02:04,  6.56s/it]
                                                 
{'loss': '1.477e-05', 'grad_norm': '0.001622', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.47', 'tokens/total': 14921008, 'tokens/trainable': 225324, 'epoch': '1.926'}
-
 96%|█████████▋| 493/512 [55:24<02:04,  6.56s/it]
 96%|█████████▋| 494/512 [55:31<01:58,  6.60s/it]
                                                 
{'loss': '3.447e-05', 'grad_norm': '0.004443', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.9', 'tokens/total': 14951360, 'tokens/trainable': 225783, 'epoch': '1.93'}
-
 96%|█████████▋| 494/512 [55:31<01:58,  6.60s/it]
 97%|█████████▋| 495/512 [55:38<01:52,  6.63s/it]
                                                 
{'loss': '1.805e-05', 'grad_norm': '0.002437', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 14981552, 'tokens/trainable': 226281, 'epoch': '1.934'}
-
 97%|█████████▋| 495/512 [55:38<01:52,  6.63s/it]
 97%|█████████▋| 496/512 [55:44<01:45,  6.62s/it]
                                                 
{'loss': '1.692e-05', 'grad_norm': '0.00158', 'learning_rate': '1.027e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.72', 'tokens/total': 15011584, 'tokens/trainable': 226735, 'epoch': '1.938'}
-
 97%|█████████▋| 496/512 [55:44<01:45,  6.62s/it]
 97%|█████████▋| 497/512 [55:51<01:39,  6.64s/it]
                                                 
{'loss': '4.592e-05', 'grad_norm': '0.003621', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.31', 'tokens/total': 15042016, 'tokens/trainable': 227198, 'epoch': '1.941'}
-
 97%|█████████▋| 497/512 [55:51<01:39,  6.64s/it]
 97%|█████████▋| 498/512 [55:58<01:32,  6.63s/it]
                                                 
{'loss': '1.531e-05', 'grad_norm': '0.0005919', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.07', 'tokens/total': 15071984, 'tokens/trainable': 227623, 'epoch': '1.945'}
-
 97%|█████████▋| 498/512 [55:58<01:32,  6.63s/it]
 97%|█████████▋| 499/512 [56:04<01:26,  6.65s/it]
                                                 
{'loss': '1.349e-05', 'grad_norm': '0.001367', 'learning_rate': '1.018e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '32.15', 'tokens/total': 15102192, 'tokens/trainable': 228074, 'epoch': '1.949'}
-
 97%|█████████▋| 499/512 [56:04<01:26,  6.65s/it]
 98%|█████████▊| 500/512 [56:11<01:19,  6.66s/it]
                                                 
{'loss': '2.841e-05', 'grad_norm': '0.002793', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 15132560, 'tokens/trainable': 228541, 'epoch': '1.953'}
-
 98%|█████████▊| 500/512 [56:11<01:19,  6.66s/it]
 98%|█████████▊| 501/512 [56:18<01:13,  6.67s/it]
                                                 
{'loss': '3.183e-05', 'grad_norm': '0.002832', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.77', 'tokens/total': 15163152, 'tokens/trainable': 229031, 'epoch': '1.957'}
-
 98%|█████████▊| 501/512 [56:18<01:13,  6.67s/it]
 98%|█████████▊| 502/512 [56:24<01:06,  6.68s/it]
                                                 
{'loss': '2.238e-05', 'grad_norm': '0.003093', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.68', 'tokens/total': 15193456, 'tokens/trainable': 229512, 'epoch': '1.961'}
-
 98%|█████████▊| 502/512 [56:24<01:06,  6.68s/it]
 98%|█████████▊| 503/512 [56:31<01:00,  6.67s/it]
                                                 
{'loss': '0.0004141', 'grad_norm': '0.0639', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 15223680, 'tokens/trainable': 229989, 'epoch': '1.965'}
-
 98%|█████████▊| 503/512 [56:31<01:00,  6.67s/it]
 98%|█████████▊| 504/512 [56:38<00:53,  6.67s/it]
                                                 
{'loss': '4.398e-05', 'grad_norm': '0.003928', 'learning_rate': '1.008e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.81', 'tokens/total': 15254336, 'tokens/trainable': 230447, 'epoch': '1.969'}
-
 98%|█████████▊| 504/512 [56:38<00:53,  6.67s/it]
 99%|█████████▊| 505/512 [56:44<00:46,  6.68s/it]
                                                 
{'loss': '2.18e-05', 'grad_norm': '0.002279', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 15284496, 'tokens/trainable': 230879, 'epoch': '1.973'}
-
 99%|█████████▊| 505/512 [56:44<00:46,  6.68s/it]
 99%|█████████▉| 506/512 [56:51<00:40,  6.69s/it]
                                                 
{'loss': '2.01e-05', 'grad_norm': '0.002216', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 15314976, 'tokens/trainable': 231348, 'epoch': '1.977'}
-
 99%|█████████▉| 506/512 [56:51<00:40,  6.69s/it]
 99%|█████████▉| 507/512 [56:58<00:33,  6.69s/it]
                                                 
{'loss': '9.908e-05', 'grad_norm': '0.01093', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.22', 'tokens/total': 15345280, 'tokens/trainable': 231819, 'epoch': '1.98'}
-
 99%|█████████▉| 507/512 [56:58<00:33,  6.69s/it]
 99%|█████████▉| 508/512 [57:04<00:26,  6.67s/it]
                                                 
{'loss': '1.419e-05', 'grad_norm': '0.0008611', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.57', 'tokens/total': 15375424, 'tokens/trainable': 232269, 'epoch': '1.984'}
-
 99%|█████████▉| 508/512 [57:04<00:26,  6.67s/it]
 99%|█████████▉| 509/512 [57:11<00:20,  6.67s/it]
                                                 
{'loss': '0.001132', 'grad_norm': '0.1847', 'learning_rate': '1.001e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 15405872, 'tokens/trainable': 232716, 'epoch': '1.988'}
-
 99%|█████████▉| 509/512 [57:11<00:20,  6.67s/it]
100%|█████████▉| 510/512 [57:18<00:13,  6.68s/it]
                                                 
{'loss': '1.466e-05', 'grad_norm': '0.001229', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 15436128, 'tokens/trainable': 233178, 'epoch': '1.992'}
-
100%|█████████▉| 510/512 [57:18<00:13,  6.68s/it]
100%|█████████▉| 511/512 [57:24<00:06,  6.68s/it]
                                                 
{'loss': '2.096e-05', 'grad_norm': '0.003051', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 15466464, 'tokens/trainable': 233647, 'epoch': '1.996'}
-
100%|█████████▉| 511/512 [57:24<00:06,  6.68s/it]
100%|██████████| 512/512 [57:31<00:00,  6.70s/it]
                                                 
{'loss': '1.928e-05', 'grad_norm': '0.00193', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.42', 'tokens/total': 15496368, 'tokens/trainable': 234060, 'epoch': '2'}
-
100%|██████████| 512/512 [57:31<00:00,  6.70s/it][2026-08-18 13:41:19,700] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
-
                                                 
{'train_runtime': '3454', 'train_samples_per_second': '4.744', 'train_steps_per_second': '0.148', 'train_loss': '0.007604', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '38.44', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
-
100%|██████████| 512/512 [57:33<00:00,  6.70s/it]
100%|██████████| 512/512 [57:33<00:00,  6.75s/it]
-[2026-08-18 13:41:21,525] [INFO] [axolotl.train] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
-[2026-08-18 13:41:21,960] [INFO] [axolotl.train] Model successfully saved to /workspace/wave/training/checkpoints
+
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
                                                 
{'loss': '0.1025', 'grad_norm': '0.8591', 'learning_rate': '0', 'ppl': '1.108', 'memory/max_active (GiB)': '32.98', 'memory/max_allocated (GiB)': '32.98', 'memory/device_reserved (GiB)': '34.94', 'tokens/train_per_sec_per_gpu': '30.01', 'tokens/total': 30464, 'tokens/trainable': 470, 'epoch': '0.003906'}
+
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
  0%|          | 2/512 [00:14<1:01:50,  7.28s/it]
                                                 
{'loss': '0.09678', 'grad_norm': '0.8034', 'learning_rate': '4e-06', 'ppl': '1.102', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.83', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 60800, 'tokens/trainable': 922, 'epoch': '0.007812'}
+
  0%|          | 2/512 [00:14<1:01:50,  7.28s/it]
  1%|          | 3/512 [00:21<59:21,  7.00s/it]  
                                               
{'loss': '0.1088', 'grad_norm': '1.01', 'learning_rate': '8e-06', 'ppl': '1.115', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '34.67', 'tokens/total': 91136, 'tokens/trainable': 1396, 'epoch': '0.01172'}
+
  1%|          | 3/512 [00:21<59:21,  7.00s/it]
  1%|          | 4/512 [00:28<58:17,  6.88s/it]
                                               
{'loss': '0.1031', 'grad_norm': '2.204', 'learning_rate': '1.2e-05', 'ppl': '1.109', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '38', 'tokens/total': 121552, 'tokens/trainable': 1850, 'epoch': '0.01562'}
+
  1%|          | 4/512 [00:28<58:17,  6.88s/it]
  1%|          | 5/512 [00:35<57:58,  6.86s/it]
                                               
{'loss': '0.1097', 'grad_norm': '2.576', 'learning_rate': '1.6e-05', 'ppl': '1.116', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 152304, 'tokens/trainable': 2302, 'epoch': '0.01953'}
+
  1%|          | 5/512 [00:35<57:58,  6.86s/it]
  1%|          | 6/512 [00:41<57:18,  6.80s/it]
                                               
{'loss': '0.08722', 'grad_norm': '1.498', 'learning_rate': '2e-05', 'ppl': '1.091', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '31.65', 'tokens/total': 182448, 'tokens/trainable': 2742, 'epoch': '0.02344'}
+
  1%|          | 6/512 [00:41<57:18,  6.80s/it]
  1%|▏         | 7/512 [00:48<56:56,  6.77s/it]
                                               
{'loss': '0.07384', 'grad_norm': '1.28', 'learning_rate': '2.4e-05', 'ppl': '1.077', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.85', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 212736, 'tokens/trainable': 3208, 'epoch': '0.02734'}
+
  1%|▏         | 7/512 [00:48<56:56,  6.77s/it]
  2%|▏         | 8/512 [00:55<56:34,  6.74s/it]
                                               
{'loss': '0.08001', 'grad_norm': '1.695', 'learning_rate': '2.8e-05', 'ppl': '1.083', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 243024, 'tokens/trainable': 3655, 'epoch': '0.03125'}
+
  2%|▏         | 8/512 [00:55<56:34,  6.74s/it]
  2%|▏         | 9/512 [01:01<55:00,  6.56s/it]
                                               
{'loss': '0.04736', 'grad_norm': '1.222', 'learning_rate': '3.2e-05', 'ppl': '1.048', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '35.78', 'tokens/total': 271264, 'tokens/trainable': 4091, 'epoch': '0.03516'}
+
  2%|▏         | 9/512 [01:01<55:00,  6.56s/it]
  2%|▏         | 10/512 [01:07<55:18,  6.61s/it]
                                                
{'loss': '0.0957', 'grad_norm': '2.902', 'learning_rate': '3.6e-05', 'ppl': '1.1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.07', 'tokens/train_per_sec_per_gpu': '34.98', 'tokens/total': 301520, 'tokens/trainable': 4553, 'epoch': '0.03906'}
+
  2%|▏         | 10/512 [01:07<55:18,  6.61s/it]
  2%|▏         | 11/512 [01:14<55:21,  6.63s/it]
                                                
{'loss': '0.08283', 'grad_norm': '2.177', 'learning_rate': '4e-05', 'ppl': '1.086', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.08', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 331808, 'tokens/trainable': 4992, 'epoch': '0.04297'}
+
  2%|▏         | 11/512 [01:14<55:21,  6.63s/it]
  2%|▏         | 12/512 [01:21<55:35,  6.67s/it]
                                                
{'loss': '0.1214', 'grad_norm': '3.152', 'learning_rate': '4.4e-05', 'ppl': '1.129', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.08', 'tokens/train_per_sec_per_gpu': '38.32', 'tokens/total': 362224, 'tokens/trainable': 5494, 'epoch': '0.04688'}
+
  2%|▏         | 12/512 [01:21<55:35,  6.67s/it]
  3%|▎         | 13/512 [01:28<55:17,  6.65s/it]
                                                
{'loss': '0.03794', 'grad_norm': '2.383', 'learning_rate': '4.8e-05', 'ppl': '1.039', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.04', 'tokens/total': 392432, 'tokens/trainable': 5933, 'epoch': '0.05078'}
+
  3%|▎         | 13/512 [01:28<55:17,  6.65s/it]
  3%|▎         | 14/512 [01:34<55:19,  6.67s/it]
                                                
{'loss': '0.06515', 'grad_norm': '3.259', 'learning_rate': '5.2e-05', 'ppl': '1.067', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '29.67', 'tokens/total': 422784, 'tokens/trainable': 6369, 'epoch': '0.05469'}
+
  3%|▎         | 14/512 [01:34<55:19,  6.67s/it]
  3%|▎         | 15/512 [01:41<55:11,  6.66s/it]
                                                
{'loss': '0.05657', 'grad_norm': '1.582', 'learning_rate': '5.6e-05', 'ppl': '1.058', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 453376, 'tokens/trainable': 6820, 'epoch': '0.05859'}
+
  3%|▎         | 15/512 [01:41<55:11,  6.66s/it]
  3%|▎         | 16/512 [01:48<55:02,  6.66s/it]
                                                
{'loss': '0.0607', 'grad_norm': '1.422', 'learning_rate': '6e-05', 'ppl': '1.063', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.85', 'tokens/total': 483504, 'tokens/trainable': 7258, 'epoch': '0.0625'}
+
  3%|▎         | 16/512 [01:48<55:02,  6.66s/it]
  3%|▎         | 17/512 [01:54<55:08,  6.68s/it]
                                                
{'loss': '0.05027', 'grad_norm': '1.307', 'learning_rate': '6.4e-05', 'ppl': '1.052', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 514032, 'tokens/trainable': 7710, 'epoch': '0.06641'}
+
  3%|▎         | 17/512 [01:54<55:08,  6.68s/it]
  4%|▎         | 18/512 [02:01<55:05,  6.69s/it]
                                                
{'loss': '0.0285', 'grad_norm': '0.6123', 'learning_rate': '6.8e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '31.76', 'tokens/total': 544432, 'tokens/trainable': 8174, 'epoch': '0.07031'}
+
  4%|▎         | 18/512 [02:01<55:05,  6.69s/it]
  4%|▎         | 19/512 [02:08<54:56,  6.69s/it]
                                                
{'loss': '0.03114', 'grad_norm': '0.6484', 'learning_rate': '7.2e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 574880, 'tokens/trainable': 8617, 'epoch': '0.07422'}
+
  4%|▎         | 19/512 [02:08<54:56,  6.69s/it]
  4%|▍         | 20/512 [02:14<54:45,  6.68s/it]
                                                
{'loss': '0.03075', 'grad_norm': '0.6737', 'learning_rate': '7.6e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '30.37', 'tokens/total': 605408, 'tokens/trainable': 9037, 'epoch': '0.07812'}
+
  4%|▍         | 20/512 [02:14<54:45,  6.68s/it]
  4%|▍         | 21/512 [02:21<54:40,  6.68s/it]
                                                
{'loss': '0.02451', 'grad_norm': '0.7465', 'learning_rate': '8e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.63', 'tokens/total': 635584, 'tokens/trainable': 9503, 'epoch': '0.08203'}
+
  4%|▍         | 21/512 [02:21<54:40,  6.68s/it]
  4%|▍         | 22/512 [02:28<54:38,  6.69s/it]
                                                
{'loss': '0.01763', 'grad_norm': '0.9436', 'learning_rate': '8.4e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '36.79', 'tokens/total': 665952, 'tokens/trainable': 9972, 'epoch': '0.08594'}
+
  4%|▍         | 22/512 [02:28<54:38,  6.69s/it]
  4%|▍         | 23/512 [02:34<54:39,  6.71s/it]
                                                
{'loss': '0.01396', 'grad_norm': '0.637', 'learning_rate': '8.8e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.21', 'tokens/train_per_sec_per_gpu': '37.54', 'tokens/total': 696240, 'tokens/trainable': 10447, 'epoch': '0.08984'}
+
  4%|▍         | 23/512 [02:34<54:39,  6.71s/it]
  5%|▍         | 24/512 [02:41<54:22,  6.69s/it]
                                                
{'loss': '0.03303', 'grad_norm': '1.067', 'learning_rate': '9.2e-05', 'ppl': '1.034', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '37.23', 'tokens/total': 726464, 'tokens/trainable': 10899, 'epoch': '0.09375'}
+
  5%|▍         | 24/512 [02:41<54:22,  6.69s/it]
  5%|▍         | 25/512 [02:48<54:19,  6.69s/it]
                                                
{'loss': '0.03084', 'grad_norm': '1.049', 'learning_rate': '9.6e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 756704, 'tokens/trainable': 11360, 'epoch': '0.09766'}
+
  5%|▍         | 25/512 [02:48<54:19,  6.69s/it]
  5%|▌         | 26/512 [02:54<54:12,  6.69s/it]
                                                
{'loss': '0.03408', 'grad_norm': '0.8108', 'learning_rate': '0.0001', 'ppl': '1.035', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '29.3', 'tokens/total': 786912, 'tokens/trainable': 11775, 'epoch': '0.1016'}
+
  5%|▌         | 26/512 [02:54<54:12,  6.69s/it]
  5%|▌         | 27/512 [03:01<52:53,  6.54s/it]
                                                
{'loss': '0.014', 'grad_norm': '0.507', 'learning_rate': '0.0001', 'ppl': '1.014', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.89', 'tokens/total': 815424, 'tokens/trainable': 12242, 'epoch': '0.1055'}
+
  5%|▌         | 27/512 [03:01<52:53,  6.54s/it]
  5%|▌         | 28/512 [03:07<53:04,  6.58s/it]
                                                
{'loss': '0.02229', 'grad_norm': '0.6185', 'learning_rate': '0.0001', 'ppl': '1.023', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.48', 'tokens/total': 845584, 'tokens/trainable': 12696, 'epoch': '0.1094'}
+
  5%|▌         | 28/512 [03:07<53:04,  6.58s/it]
  6%|▌         | 29/512 [03:14<53:15,  6.62s/it]
                                                
{'loss': '0.02593', 'grad_norm': '0.4989', 'learning_rate': '9.999e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 875808, 'tokens/trainable': 13153, 'epoch': '0.1133'}
+
  6%|▌         | 29/512 [03:14<53:15,  6.62s/it]
  6%|▌         | 30/512 [03:20<52:19,  6.51s/it]
                                                
{'loss': '0.01448', 'grad_norm': '0.3578', 'learning_rate': '9.999e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '39.7', 'tokens/total': 904096, 'tokens/trainable': 13624, 'epoch': '0.1172'}
+
  6%|▌         | 30/512 [03:20<52:19,  6.51s/it]
  6%|▌         | 31/512 [03:27<52:42,  6.57s/it]
                                                
{'loss': '0.01593', 'grad_norm': '0.4404', 'learning_rate': '9.998e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 934400, 'tokens/trainable': 14064, 'epoch': '0.1211'}
+
  6%|▌         | 31/512 [03:27<52:42,  6.57s/it]
  6%|▋         | 32/512 [03:34<52:58,  6.62s/it]
                                                
{'loss': '0.01941', 'grad_norm': '0.5914', 'learning_rate': '9.997e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.92', 'tokens/total': 964832, 'tokens/trainable': 14565, 'epoch': '0.125'}
+
  6%|▋         | 32/512 [03:34<52:58,  6.62s/it][2026-08-18 14:08:37,375] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
+
  6%|▋         | 33/512 [03:43<58:41,  7.35s/it]
                                                
{'loss': '0.007002', 'grad_norm': '0.2796', 'learning_rate': '9.995e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '33.77', 'tokens/total': 995040, 'tokens/trainable': 15005, 'epoch': '0.1289'}
+
  6%|▋         | 33/512 [03:43<58:41,  7.35s/it]
  7%|▋         | 34/512 [03:49<56:49,  7.13s/it]
                                                
{'loss': '0.02145', 'grad_norm': '1.176', 'learning_rate': '9.994e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 1024896, 'tokens/trainable': 15459, 'epoch': '0.1328'}
+
  7%|▋         | 34/512 [03:49<56:49,  7.13s/it]
  7%|▋         | 35/512 [03:56<55:36,  7.00s/it]
                                                
{'loss': '0.01894', 'grad_norm': '0.386', 'learning_rate': '9.992e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '38.21', 'tokens/total': 1054992, 'tokens/trainable': 15960, 'epoch': '0.1367'}
+
  7%|▋         | 35/512 [03:56<55:36,  7.00s/it]
  7%|▋         | 36/512 [04:03<54:40,  6.89s/it]
                                                
{'loss': '0.01381', 'grad_norm': '0.4804', 'learning_rate': '9.991e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '34.59', 'tokens/total': 1085280, 'tokens/trainable': 16428, 'epoch': '0.1406'}
+
  7%|▋         | 36/512 [04:03<54:40,  6.89s/it]
  7%|▋         | 37/512 [04:09<53:54,  6.81s/it]
                                                
{'loss': '0.008235', 'grad_norm': '0.3357', 'learning_rate': '9.989e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '31.89', 'tokens/total': 1115504, 'tokens/trainable': 16884, 'epoch': '0.1445'}
+
  7%|▋         | 37/512 [04:09<53:54,  6.81s/it]
  7%|▋         | 38/512 [04:16<53:34,  6.78s/it]
                                                
{'loss': '0.01956', 'grad_norm': '0.9475', 'learning_rate': '9.987e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '32.02', 'tokens/total': 1145712, 'tokens/trainable': 17341, 'epoch': '0.1484'}
+
  7%|▋         | 38/512 [04:16<53:34,  6.78s/it]
  8%|▊         | 39/512 [04:23<53:13,  6.75s/it]
                                                
{'loss': '0.01788', 'grad_norm': '1.102', 'learning_rate': '9.984e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 1176128, 'tokens/trainable': 17786, 'epoch': '0.1523'}
+
  8%|▊         | 39/512 [04:23<53:13,  6.75s/it]
  8%|▊         | 40/512 [04:30<53:05,  6.75s/it]
                                                
{'loss': '0.01934', 'grad_norm': '0.7563', 'learning_rate': '9.982e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 1206576, 'tokens/trainable': 18262, 'epoch': '0.1562'}
+
  8%|▊         | 40/512 [04:30<53:05,  6.75s/it]
  8%|▊         | 41/512 [04:36<52:54,  6.74s/it]
                                                
{'loss': '0.01162', 'grad_norm': '0.4158', 'learning_rate': '9.979e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 1236848, 'tokens/trainable': 18716, 'epoch': '0.1602'}
+
  8%|▊         | 41/512 [04:36<52:54,  6.74s/it]
  8%|▊         | 42/512 [04:43<52:41,  6.73s/it]
                                                
{'loss': '0.04662', 'grad_norm': '1.181', 'learning_rate': '9.976e-05', 'ppl': '1.048', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 1267088, 'tokens/trainable': 19157, 'epoch': '0.1641'}
+
  8%|▊         | 42/512 [04:43<52:41,  6.73s/it]
  8%|▊         | 43/512 [04:50<52:36,  6.73s/it]
                                                
{'loss': '0.02922', 'grad_norm': '0.7583', 'learning_rate': '9.973e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 1297360, 'tokens/trainable': 19647, 'epoch': '0.168'}
+
  8%|▊         | 43/512 [04:50<52:36,  6.73s/it]
  9%|▊         | 44/512 [04:56<52:25,  6.72s/it]
                                                
{'loss': '0.007799', 'grad_norm': '0.1886', 'learning_rate': '9.97e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 1327744, 'tokens/trainable': 20119, 'epoch': '0.1719'}
+
  9%|▊         | 44/512 [04:56<52:25,  6.72s/it]
  9%|▉         | 45/512 [05:03<52:14,  6.71s/it]
                                                
{'loss': '0.01204', 'grad_norm': '0.351', 'learning_rate': '9.966e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 1358096, 'tokens/trainable': 20560, 'epoch': '0.1758'}
+
  9%|▉         | 45/512 [05:03<52:14,  6.71s/it]
  9%|▉         | 46/512 [05:10<51:59,  6.69s/it]
                                                
{'loss': '0.01786', 'grad_norm': '0.5071', 'learning_rate': '9.963e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 1388240, 'tokens/trainable': 20992, 'epoch': '0.1797'}
+
  9%|▉         | 46/512 [05:10<51:59,  6.69s/it]
  9%|▉         | 47/512 [05:16<50:37,  6.53s/it]
                                                
{'loss': '0.01944', 'grad_norm': '0.557', 'learning_rate': '9.959e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.26', 'memory/max_allocated (GiB)': '33.26', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.55', 'tokens/total': 1416240, 'tokens/trainable': 21441, 'epoch': '0.1836'}
+
  9%|▉         | 47/512 [05:16<50:37,  6.53s/it]
  9%|▉         | 48/512 [05:23<50:51,  6.58s/it]
                                                
{'loss': '0.01302', 'grad_norm': '0.4206', 'learning_rate': '9.955e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.68', 'tokens/total': 1446880, 'tokens/trainable': 21901, 'epoch': '0.1875'}
+
  9%|▉         | 48/512 [05:23<50:51,  6.58s/it]
 10%|▉         | 49/512 [05:29<50:57,  6.60s/it]
                                                
{'loss': '0.007749', 'grad_norm': '0.304', 'learning_rate': '9.951e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 1477168, 'tokens/trainable': 22341, 'epoch': '0.1914'}
+
 10%|▉         | 49/512 [05:29<50:57,  6.60s/it]
 10%|▉         | 50/512 [05:36<51:07,  6.64s/it]
                                                
{'loss': '0.01445', 'grad_norm': '2.045', 'learning_rate': '9.946e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 1507712, 'tokens/trainable': 22833, 'epoch': '0.1953'}
+
 10%|▉         | 50/512 [05:36<51:07,  6.64s/it]
 10%|▉         | 51/512 [05:43<51:06,  6.65s/it]
                                                
{'loss': '0.01272', 'grad_norm': '0.9358', 'learning_rate': '9.942e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.72', 'tokens/total': 1537936, 'tokens/trainable': 23277, 'epoch': '0.1992'}
+
 10%|▉         | 51/512 [05:43<51:06,  6.65s/it]
 10%|█         | 52/512 [05:49<51:07,  6.67s/it]
                                                
{'loss': '0.005878', 'grad_norm': '0.2882', 'learning_rate': '9.937e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.63', 'tokens/total': 1568352, 'tokens/trainable': 23759, 'epoch': '0.2031'}
+
 10%|█         | 52/512 [05:49<51:07,  6.67s/it]
 10%|█         | 53/512 [05:56<51:10,  6.69s/it]
                                                
{'loss': '0.01452', 'grad_norm': '0.7578', 'learning_rate': '9.932e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 1598992, 'tokens/trainable': 24193, 'epoch': '0.207'}
+
 10%|█         | 53/512 [05:56<51:10,  6.69s/it]
 11%|█         | 54/512 [06:03<50:57,  6.68s/it]
                                                
{'loss': '0.006856', 'grad_norm': '0.5201', 'learning_rate': '9.927e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.59', 'tokens/total': 1629488, 'tokens/trainable': 24619, 'epoch': '0.2109'}
+
 11%|█         | 54/512 [06:03<50:57,  6.68s/it]
 11%|█         | 55/512 [06:09<50:53,  6.68s/it]
                                                
{'loss': '0.03797', 'grad_norm': '0.84', 'learning_rate': '9.921e-05', 'ppl': '1.039', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.85', 'tokens/total': 1659696, 'tokens/trainable': 25075, 'epoch': '0.2148'}
+
 11%|█         | 55/512 [06:09<50:53,  6.68s/it]
 11%|█         | 56/512 [06:16<50:43,  6.68s/it]
                                                
{'loss': '0.01281', 'grad_norm': '0.8204', 'learning_rate': '9.916e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.55', 'tokens/total': 1689872, 'tokens/trainable': 25519, 'epoch': '0.2188'}
+
 11%|█         | 56/512 [06:16<50:43,  6.68s/it]
 11%|█         | 57/512 [06:23<50:40,  6.68s/it]
                                                
{'loss': '0.00539', 'grad_norm': '0.2087', 'learning_rate': '9.91e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 1720144, 'tokens/trainable': 25966, 'epoch': '0.2227'}
+
 11%|█         | 57/512 [06:23<50:40,  6.68s/it]
 11%|█▏        | 58/512 [06:29<50:34,  6.68s/it]
                                                
{'loss': '0.009391', 'grad_norm': '0.4428', 'learning_rate': '9.904e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 1750448, 'tokens/trainable': 26423, 'epoch': '0.2266'}
+
 11%|█▏        | 58/512 [06:29<50:34,  6.68s/it]
 12%|█▏        | 59/512 [06:36<50:37,  6.70s/it]
                                                
{'loss': '0.01953', 'grad_norm': '0.7458', 'learning_rate': '9.898e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.51', 'tokens/total': 1781088, 'tokens/trainable': 26934, 'epoch': '0.2305'}
+
 12%|█▏        | 59/512 [06:36<50:37,  6.70s/it]
 12%|█▏        | 60/512 [06:43<50:30,  6.71s/it]
                                                
{'loss': '0.001454', 'grad_norm': '0.134', 'learning_rate': '9.892e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 1811344, 'tokens/trainable': 27393, 'epoch': '0.2344'}
+
 12%|█▏        | 60/512 [06:43<50:30,  6.71s/it]
 12%|█▏        | 61/512 [06:50<50:20,  6.70s/it]
                                                
{'loss': '0.01992', 'grad_norm': '1.039', 'learning_rate': '9.886e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 1841600, 'tokens/trainable': 27852, 'epoch': '0.2383'}
+
 12%|█▏        | 61/512 [06:50<50:20,  6.70s/it]
 12%|█▏        | 62/512 [06:56<50:11,  6.69s/it]
                                                
{'loss': '0.01588', 'grad_norm': '1.26', 'learning_rate': '9.879e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 1872048, 'tokens/trainable': 28292, 'epoch': '0.2422'}
+
 12%|█▏        | 62/512 [06:56<50:11,  6.69s/it]
 12%|█▏        | 63/512 [07:03<50:11,  6.71s/it]
                                                
{'loss': '0.002796', 'grad_norm': '0.1403', 'learning_rate': '9.872e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '37.36', 'tokens/total': 1902592, 'tokens/trainable': 28798, 'epoch': '0.2461'}
+
 12%|█▏        | 63/512 [07:03<50:11,  6.71s/it]
 12%|█▎        | 64/512 [07:10<50:05,  6.71s/it]
                                                
{'loss': '0.01712', 'grad_norm': '0.6248', 'learning_rate': '9.865e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.79', 'tokens/total': 1933088, 'tokens/trainable': 29283, 'epoch': '0.25'}
+
 12%|█▎        | 64/512 [07:10<50:05,  6.71s/it][2026-08-18 14:12:13,340] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
+
 13%|█▎        | 65/512 [07:18<53:59,  7.25s/it]
                                                
{'loss': '0.008029', 'grad_norm': '0.3513', 'learning_rate': '9.858e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.99', 'tokens/total': 1963296, 'tokens/trainable': 29745, 'epoch': '0.2539'}
+
 13%|█▎        | 65/512 [07:18<53:59,  7.25s/it]
 13%|█▎        | 66/512 [07:25<52:47,  7.10s/it]
                                                
{'loss': '0.01612', 'grad_norm': '0.2794', 'learning_rate': '9.851e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '34.87', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 1993760, 'tokens/trainable': 30186, 'epoch': '0.2578'}
+
 13%|█▎        | 66/512 [07:25<52:47,  7.10s/it]
 13%|█▎        | 67/512 [07:32<51:41,  6.97s/it]
                                                
{'loss': '0.009092', 'grad_norm': '0.3174', 'learning_rate': '9.844e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '33.36', 'tokens/total': 2024144, 'tokens/trainable': 30622, 'epoch': '0.2617'}
+
 13%|█▎        | 67/512 [07:32<51:41,  6.97s/it]
 13%|█▎        | 68/512 [07:38<50:58,  6.89s/it]
                                                
{'loss': '0.009083', 'grad_norm': '0.3539', 'learning_rate': '9.836e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.52', 'tokens/total': 2054608, 'tokens/trainable': 31097, 'epoch': '0.2656'}
+
 13%|█▎        | 68/512 [07:38<50:58,  6.89s/it]
 13%|█▎        | 69/512 [07:45<50:26,  6.83s/it]
                                                
{'loss': '0.03087', 'grad_norm': '0.6498', 'learning_rate': '9.828e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36.6', 'tokens/total': 2084912, 'tokens/trainable': 31595, 'epoch': '0.2695'}
+
 13%|█▎        | 69/512 [07:45<50:26,  6.83s/it]
 14%|█▎        | 70/512 [07:52<50:01,  6.79s/it]
                                                
{'loss': '0.003916', 'grad_norm': '0.1966', 'learning_rate': '9.82e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '30.75', 'tokens/total': 2115216, 'tokens/trainable': 32036, 'epoch': '0.2734'}
+
 14%|█▎        | 70/512 [07:52<50:01,  6.79s/it]
 14%|█▍        | 71/512 [07:58<49:37,  6.75s/it]
                                                
{'loss': '0.007547', 'grad_norm': '0.43', 'learning_rate': '9.812e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.09', 'tokens/total': 2145424, 'tokens/trainable': 32428, 'epoch': '0.2773'}
+
 14%|█▍        | 71/512 [07:58<49:37,  6.75s/it]
 14%|█▍        | 72/512 [08:05<49:23,  6.74s/it]
                                                
{'loss': '0.0004817', 'grad_norm': '0.02775', 'learning_rate': '9.803e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.11', 'tokens/total': 2175648, 'tokens/trainable': 32880, 'epoch': '0.2812'}
+
 14%|█▍        | 72/512 [08:05<49:23,  6.74s/it]
 14%|█▍        | 73/512 [08:12<49:10,  6.72s/it]
                                                
{'loss': '0.001214', 'grad_norm': '0.112', 'learning_rate': '9.795e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.51', 'tokens/total': 2205856, 'tokens/trainable': 33323, 'epoch': '0.2852'}
+
 14%|█▍        | 73/512 [08:12<49:10,  6.72s/it]
 14%|█▍        | 74/512 [08:18<48:57,  6.71s/it]
                                                
{'loss': '0.0004691', 'grad_norm': '0.02696', 'learning_rate': '9.786e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.86', 'tokens/total': 2236352, 'tokens/trainable': 33792, 'epoch': '0.2891'}
+
 14%|█▍        | 74/512 [08:18<48:57,  6.71s/it]
 15%|█▍        | 75/512 [08:25<48:45,  6.69s/it]
                                                
{'loss': '0.02965', 'grad_norm': '0.5172', 'learning_rate': '9.777e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 2266480, 'tokens/trainable': 34223, 'epoch': '0.293'}
+
 15%|█▍        | 75/512 [08:25<48:45,  6.69s/it]
 15%|█▍        | 76/512 [08:32<48:33,  6.68s/it]
                                                
{'loss': '0.00698', 'grad_norm': '0.8409', 'learning_rate': '9.768e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.61', 'tokens/total': 2296496, 'tokens/trainable': 34649, 'epoch': '0.2969'}
+
 15%|█▍        | 76/512 [08:32<48:33,  6.68s/it]
 15%|█▌        | 77/512 [08:38<48:22,  6.67s/it]
                                                
{'loss': '0.00887', 'grad_norm': '3.494', 'learning_rate': '9.759e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.64', 'tokens/total': 2327040, 'tokens/trainable': 35094, 'epoch': '0.3008'}
+
 15%|█▌        | 77/512 [08:38<48:22,  6.67s/it]
 15%|█▌        | 78/512 [08:45<48:15,  6.67s/it]
                                                
{'loss': '0.03207', 'grad_norm': '0.9529', 'learning_rate': '9.749e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.82', 'tokens/total': 2357328, 'tokens/trainable': 35534, 'epoch': '0.3047'}
+
 15%|█▌        | 78/512 [08:45<48:15,  6.67s/it]
 15%|█▌        | 79/512 [08:52<48:12,  6.68s/it]
                                                
{'loss': '0.01475', 'grad_norm': '0.5782', 'learning_rate': '9.74e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.82', 'tokens/total': 2387664, 'tokens/trainable': 36029, 'epoch': '0.3086'}
+
 15%|█▌        | 79/512 [08:52<48:12,  6.68s/it]
 16%|█▌        | 80/512 [08:58<48:01,  6.67s/it]
                                                
{'loss': '0.002579', 'grad_norm': '0.1509', 'learning_rate': '9.73e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 2418000, 'tokens/trainable': 36469, 'epoch': '0.3125'}
+
 16%|█▌        | 80/512 [08:58<48:01,  6.67s/it]
 16%|█▌        | 81/512 [09:05<48:02,  6.69s/it]
                                                
{'loss': '0.004292', 'grad_norm': '0.19', 'learning_rate': '9.72e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.75', 'tokens/total': 2448720, 'tokens/trainable': 36906, 'epoch': '0.3164'}
+
 16%|█▌        | 81/512 [09:05<48:02,  6.69s/it]
 16%|█▌        | 82/512 [09:12<47:54,  6.69s/it]
                                                
{'loss': '0.01853', 'grad_norm': '0.4506', 'learning_rate': '9.71e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '38.95', 'tokens/total': 2479248, 'tokens/trainable': 37390, 'epoch': '0.3203'}
+
 16%|█▌        | 82/512 [09:12<47:54,  6.69s/it]
 16%|█▌        | 83/512 [09:19<47:48,  6.69s/it]
                                                
{'loss': '0.01482', 'grad_norm': '0.4981', 'learning_rate': '9.699e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.8', 'tokens/total': 2509408, 'tokens/trainable': 37838, 'epoch': '0.3242'}
+
 16%|█▌        | 83/512 [09:19<47:48,  6.69s/it]
 16%|█▋        | 84/512 [09:25<47:41,  6.69s/it]
                                                
{'loss': '0.003186', 'grad_norm': '0.1638', 'learning_rate': '9.689e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 2539776, 'tokens/trainable': 38310, 'epoch': '0.3281'}
+
 16%|█▋        | 84/512 [09:25<47:41,  6.69s/it]
 17%|█▋        | 85/512 [09:32<47:33,  6.68s/it]
                                                
{'loss': '0.0208', 'grad_norm': '0.4073', 'learning_rate': '9.678e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.03', 'tokens/total': 2569840, 'tokens/trainable': 38789, 'epoch': '0.332'}
+
 17%|█▋        | 85/512 [09:32<47:33,  6.68s/it]
 17%|█▋        | 86/512 [09:39<47:30,  6.69s/it]
                                                
{'loss': '0.01293', 'grad_norm': '0.3687', 'learning_rate': '9.667e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '36.36', 'tokens/total': 2600192, 'tokens/trainable': 39277, 'epoch': '0.3359'}
+
 17%|█▋        | 86/512 [09:39<47:30,  6.69s/it]
 17%|█▋        | 87/512 [09:45<47:24,  6.69s/it]
                                                
{'loss': '0.01383', 'grad_norm': '0.3674', 'learning_rate': '9.656e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.68', 'tokens/total': 2630640, 'tokens/trainable': 39737, 'epoch': '0.3398'}
+
 17%|█▋        | 87/512 [09:45<47:24,  6.69s/it]
 17%|█▋        | 88/512 [09:52<47:13,  6.68s/it]
                                                
{'loss': '0.008718', 'grad_norm': '0.2704', 'learning_rate': '9.645e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '29.6', 'tokens/total': 2660832, 'tokens/trainable': 40179, 'epoch': '0.3438'}
+
 17%|█▋        | 88/512 [09:52<47:13,  6.68s/it]
 17%|█▋        | 89/512 [09:59<47:09,  6.69s/it]
                                                
{'loss': '0.01938', 'grad_norm': '3.375', 'learning_rate': '9.633e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.66', 'tokens/total': 2691328, 'tokens/trainable': 40649, 'epoch': '0.3477'}
+
 17%|█▋        | 89/512 [09:59<47:09,  6.69s/it]
 18%|█▊        | 90/512 [10:05<46:01,  6.54s/it]
                                                
{'loss': '0.02013', 'grad_norm': '0.469', 'learning_rate': '9.622e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 2719696, 'tokens/trainable': 41080, 'epoch': '0.3516'}
+
 18%|█▊        | 90/512 [10:05<46:01,  6.54s/it]
 18%|█▊        | 91/512 [10:12<46:20,  6.60s/it]
                                                
{'loss': '0.002885', 'grad_norm': '0.08706', 'learning_rate': '9.61e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '38.11', 'tokens/total': 2750272, 'tokens/trainable': 41599, 'epoch': '0.3555'}
+
 18%|█▊        | 91/512 [10:12<46:20,  6.60s/it]
 18%|█▊        | 92/512 [10:18<46:23,  6.63s/it]
                                                
{'loss': '0.008225', 'grad_norm': '8.198', 'learning_rate': '9.598e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 2780656, 'tokens/trainable': 42076, 'epoch': '0.3594'}
+
 18%|█▊        | 92/512 [10:18<46:23,  6.63s/it]
 18%|█▊        | 93/512 [10:25<46:23,  6.64s/it]
                                                
{'loss': '0.004234', 'grad_norm': '0.1494', 'learning_rate': '9.586e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.71', 'tokens/total': 2811088, 'tokens/trainable': 42522, 'epoch': '0.3633'}
+
 18%|█▊        | 93/512 [10:25<46:23,  6.64s/it]
 18%|█▊        | 94/512 [10:32<46:22,  6.66s/it]
                                                
{'loss': '0.00949', 'grad_norm': '0.154', 'learning_rate': '9.574e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.4', 'tokens/total': 2841520, 'tokens/trainable': 42974, 'epoch': '0.3672'}
+
 18%|█▊        | 94/512 [10:32<46:22,  6.66s/it]
 19%|█▊        | 95/512 [10:38<46:07,  6.64s/it]
                                                
{'loss': '0.02414', 'grad_norm': '0.5275', 'learning_rate': '9.562e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '30.67', 'tokens/total': 2871600, 'tokens/trainable': 43380, 'epoch': '0.3711'}
+
 19%|█▊        | 95/512 [10:38<46:07,  6.64s/it]
 19%|█▉        | 96/512 [10:45<46:13,  6.67s/it]
                                                
{'loss': '0.009927', 'grad_norm': '0.5183', 'learning_rate': '9.549e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 2902144, 'tokens/trainable': 43865, 'epoch': '0.375'}
+
 19%|█▉        | 96/512 [10:45<46:13,  6.67s/it][2026-08-18 14:15:48,624] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
+
 19%|█▉        | 97/512 [10:54<50:11,  7.26s/it]
                                                
{'loss': '0.01449', 'grad_norm': '0.5579', 'learning_rate': '9.536e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.1', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 2932272, 'tokens/trainable': 44328, 'epoch': '0.3789'}
+
 19%|█▉        | 97/512 [10:54<50:11,  7.26s/it]
 19%|█▉        | 98/512 [11:00<48:41,  7.06s/it]
                                                
{'loss': '0.002155', 'grad_norm': '0.1052', 'learning_rate': '9.523e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '30.95', 'tokens/total': 2962032, 'tokens/trainable': 44778, 'epoch': '0.3828'}
+
 19%|█▉        | 98/512 [11:00<48:41,  7.06s/it]
 19%|█▉        | 99/512 [11:07<47:44,  6.94s/it]
                                                
{'loss': '0.01703', 'grad_norm': '0.7545', 'learning_rate': '9.51e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.12', 'tokens/total': 2992256, 'tokens/trainable': 45225, 'epoch': '0.3867'}
+
 19%|█▉        | 99/512 [11:07<47:44,  6.94s/it]
 20%|█▉        | 100/512 [11:14<47:07,  6.86s/it]
                                                 
{'loss': '0.006046', 'grad_norm': '0.3857', 'learning_rate': '9.497e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.75', 'tokens/total': 3022608, 'tokens/trainable': 45678, 'epoch': '0.3906'}
+
 20%|█▉        | 100/512 [11:14<47:07,  6.86s/it]
 20%|█▉        | 101/512 [11:20<46:35,  6.80s/it]
                                                 
{'loss': '0.005754', 'grad_norm': '0.2159', 'learning_rate': '9.484e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '32.72', 'tokens/total': 3052992, 'tokens/trainable': 46125, 'epoch': '0.3945'}
+
 20%|█▉        | 101/512 [11:20<46:35,  6.80s/it]
 20%|█▉        | 102/512 [11:27<46:25,  6.79s/it]
                                                 
{'loss': '0.006994', 'grad_norm': '1.125', 'learning_rate': '9.47e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '39.6', 'tokens/total': 3083392, 'tokens/trainable': 46612, 'epoch': '0.3984'}
+
 20%|█▉        | 102/512 [11:27<46:25,  6.79s/it]
 20%|██        | 103/512 [11:34<46:14,  6.78s/it]
                                                 
{'loss': '0.01372', 'grad_norm': '0.2677', 'learning_rate': '9.456e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '32.41', 'tokens/total': 3113744, 'tokens/trainable': 47059, 'epoch': '0.4023'}
+
 20%|██        | 103/512 [11:34<46:14,  6.78s/it]
 20%|██        | 104/512 [11:41<46:17,  6.81s/it]
                                                 
{'loss': '0.01295', 'grad_norm': '0.3259', 'learning_rate': '9.442e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 3144272, 'tokens/trainable': 47536, 'epoch': '0.4062'}
+
 20%|██        | 104/512 [11:41<46:17,  6.81s/it]
 21%|██        | 105/512 [11:47<46:03,  6.79s/it]
                                                 
{'loss': '0.02244', 'grad_norm': '0.5319', 'learning_rate': '9.428e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 3174512, 'tokens/trainable': 48037, 'epoch': '0.4102'}
+
 21%|██        | 105/512 [11:47<46:03,  6.79s/it]
 21%|██        | 106/512 [11:54<45:47,  6.77s/it]
                                                 
{'loss': '0.0158', 'grad_norm': '0.3957', 'learning_rate': '9.414e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 3204896, 'tokens/trainable': 48491, 'epoch': '0.4141'}
+
 21%|██        | 106/512 [11:54<45:47,  6.77s/it]
 21%|██        | 107/512 [12:01<45:42,  6.77s/it]
                                                 
{'loss': '0.01632', 'grad_norm': '0.4242', 'learning_rate': '9.4e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 3235360, 'tokens/trainable': 48940, 'epoch': '0.418'}
+
 21%|██        | 107/512 [12:01<45:42,  6.77s/it]
 21%|██        | 108/512 [12:08<45:30,  6.76s/it]
                                                 
{'loss': '0.01079', 'grad_norm': '0.3206', 'learning_rate': '9.385e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 3265552, 'tokens/trainable': 49389, 'epoch': '0.4219'}
+
 21%|██        | 108/512 [12:08<45:30,  6.76s/it]
 21%|██▏       | 109/512 [12:14<44:10,  6.58s/it]
                                                 
{'loss': '0.008147', 'grad_norm': '0.1722', 'learning_rate': '9.37e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.91', 'tokens/total': 3293856, 'tokens/trainable': 49838, 'epoch': '0.4258'}
+
 21%|██▏       | 109/512 [12:14<44:10,  6.58s/it]
 21%|██▏       | 110/512 [12:20<44:21,  6.62s/it]
                                                 
{'loss': '0.01213', 'grad_norm': '0.3818', 'learning_rate': '9.355e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 3323984, 'tokens/trainable': 50310, 'epoch': '0.4297'}
+
 21%|██▏       | 110/512 [12:20<44:21,  6.62s/it]
 22%|██▏       | 111/512 [12:27<44:21,  6.64s/it]
                                                 
{'loss': '0.01525', 'grad_norm': '0.4984', 'learning_rate': '9.34e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.55', 'tokens/total': 3354320, 'tokens/trainable': 50755, 'epoch': '0.4336'}
+
 22%|██▏       | 111/512 [12:27<44:21,  6.64s/it]
 22%|██▏       | 112/512 [12:34<44:26,  6.67s/it]
                                                 
{'loss': '0.01127', 'grad_norm': '0.439', 'learning_rate': '9.325e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.5', 'tokens/total': 3384880, 'tokens/trainable': 51213, 'epoch': '0.4375'}
+
 22%|██▏       | 112/512 [12:34<44:26,  6.67s/it]
 22%|██▏       | 113/512 [12:41<44:16,  6.66s/it]
                                                 
{'loss': '0.002892', 'grad_norm': '0.567', 'learning_rate': '9.31e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '33.77', 'tokens/total': 3415104, 'tokens/trainable': 51660, 'epoch': '0.4414'}
+
 22%|██▏       | 113/512 [12:41<44:16,  6.66s/it]
 22%|██▏       | 114/512 [12:47<44:16,  6.67s/it]
                                                 
{'loss': '0.002557', 'grad_norm': '0.1595', 'learning_rate': '9.294e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.34', 'tokens/total': 3445504, 'tokens/trainable': 52133, 'epoch': '0.4453'}
+
 22%|██▏       | 114/512 [12:47<44:16,  6.67s/it]
 22%|██▏       | 115/512 [12:54<44:17,  6.69s/it]
                                                 
{'loss': '0.02446', 'grad_norm': '0.7951', 'learning_rate': '9.278e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.25', 'tokens/total': 3475744, 'tokens/trainable': 52580, 'epoch': '0.4492'}
+
 22%|██▏       | 115/512 [12:54<44:17,  6.69s/it]
 23%|██▎       | 116/512 [13:01<44:16,  6.71s/it]
                                                 
{'loss': '0.004517', 'grad_norm': '0.3832', 'learning_rate': '9.263e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.74', 'tokens/total': 3506016, 'tokens/trainable': 53033, 'epoch': '0.4531'}
+
 23%|██▎       | 116/512 [13:01<44:16,  6.71s/it]
 23%|██▎       | 117/512 [13:07<44:12,  6.71s/it]
                                                 
{'loss': '0.02197', 'grad_norm': '0.3037', 'learning_rate': '9.247e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.22', 'tokens/total': 3536256, 'tokens/trainable': 53517, 'epoch': '0.457'}
+
 23%|██▎       | 117/512 [13:07<44:12,  6.71s/it]
 23%|██▎       | 118/512 [13:14<43:54,  6.69s/it]
                                                 
{'loss': '0.009211', 'grad_norm': '1.39', 'learning_rate': '9.23e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '34.59', 'tokens/total': 3566480, 'tokens/trainable': 53967, 'epoch': '0.4609'}
+
 23%|██▎       | 118/512 [13:14<43:54,  6.69s/it]
 23%|██▎       | 119/512 [13:21<43:47,  6.69s/it]
                                                 
{'loss': '0.002145', 'grad_norm': '0.1572', 'learning_rate': '9.214e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.33', 'tokens/total': 3596624, 'tokens/trainable': 54430, 'epoch': '0.4648'}
+
 23%|██▎       | 119/512 [13:21<43:47,  6.69s/it]
 23%|██▎       | 120/512 [13:27<43:40,  6.69s/it]
                                                 
{'loss': '0.007881', 'grad_norm': '0.4', 'learning_rate': '9.198e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 3627168, 'tokens/trainable': 54896, 'epoch': '0.4688'}
+
 23%|██▎       | 120/512 [13:27<43:40,  6.69s/it]
 24%|██▎       | 121/512 [13:34<43:38,  6.70s/it]
                                                 
{'loss': '0.01013', 'grad_norm': '0.3812', 'learning_rate': '9.181e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 3657632, 'tokens/trainable': 55379, 'epoch': '0.4727'}
+
 24%|██▎       | 121/512 [13:34<43:38,  6.70s/it]
 24%|██▍       | 122/512 [13:41<43:23,  6.68s/it]
                                                 
{'loss': '0.000794', 'grad_norm': '0.05129', 'learning_rate': '9.164e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 3687824, 'tokens/trainable': 55803, 'epoch': '0.4766'}
+
 24%|██▍       | 122/512 [13:41<43:23,  6.68s/it]
 24%|██▍       | 123/512 [13:47<43:16,  6.67s/it]
                                                 
{'loss': '0.002327', 'grad_norm': '0.1332', 'learning_rate': '9.147e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 3718320, 'tokens/trainable': 56249, 'epoch': '0.4805'}
+
 24%|██▍       | 123/512 [13:47<43:16,  6.67s/it]
 24%|██▍       | 124/512 [13:54<43:15,  6.69s/it]
                                                 
{'loss': '0.003759', 'grad_norm': '0.2302', 'learning_rate': '9.13e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.62', 'tokens/train_per_sec_per_gpu': '32.43', 'tokens/total': 3748672, 'tokens/trainable': 56713, 'epoch': '0.4844'}
+
 24%|██▍       | 124/512 [13:54<43:15,  6.69s/it]
 24%|██▍       | 125/512 [14:01<43:13,  6.70s/it]
                                                 
{'loss': '0.004737', 'grad_norm': '0.3058', 'learning_rate': '9.113e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '39.1', 'tokens/total': 3778976, 'tokens/trainable': 57196, 'epoch': '0.4883'}
+
 24%|██▍       | 125/512 [14:01<43:13,  6.70s/it]
 25%|██▍       | 126/512 [14:08<43:06,  6.70s/it]
                                                 
{'loss': '0.005812', 'grad_norm': '0.3715', 'learning_rate': '9.096e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 3809296, 'tokens/trainable': 57680, 'epoch': '0.4922'}
+
 25%|██▍       | 126/512 [14:08<43:06,  6.70s/it]
 25%|██▍       | 127/512 [14:14<42:50,  6.68s/it]
                                                 
{'loss': '0.003203', 'grad_norm': '0.1105', 'learning_rate': '9.078e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.88', 'tokens/total': 3839328, 'tokens/trainable': 58110, 'epoch': '0.4961'}
+
 25%|██▍       | 127/512 [14:14<42:50,  6.68s/it]
 25%|██▌       | 128/512 [14:21<42:44,  6.68s/it]
                                                 
{'loss': '0.001725', 'grad_norm': '0.09251', 'learning_rate': '9.06e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '31.74', 'tokens/total': 3869824, 'tokens/trainable': 58534, 'epoch': '0.5'}
+
 25%|██▌       | 128/512 [14:21<42:44,  6.68s/it][2026-08-18 14:19:24,541] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
+
 25%|██▌       | 129/512 [14:29<46:14,  7.25s/it]
                                                 
{'loss': '0.004688', 'grad_norm': '0.2276', 'learning_rate': '9.043e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.76', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 3900080, 'tokens/trainable': 58998, 'epoch': '0.5039'}
+
 25%|██▌       | 129/512 [14:29<46:14,  7.25s/it]
 25%|██▌       | 130/512 [14:36<45:04,  7.08s/it]
                                                 
{'loss': '0.008267', 'grad_norm': '0.2882', 'learning_rate': '9.025e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 3930560, 'tokens/trainable': 59448, 'epoch': '0.5078'}
+
 25%|██▌       | 130/512 [14:36<45:04,  7.08s/it]
 26%|██▌       | 131/512 [14:43<44:07,  6.95s/it]
                                                 
{'loss': '0.002837', 'grad_norm': '0.188', 'learning_rate': '9.007e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '31.92', 'tokens/total': 3960496, 'tokens/trainable': 59906, 'epoch': '0.5117'}
+
 26%|██▌       | 131/512 [14:43<44:07,  6.95s/it]
 26%|██▌       | 132/512 [14:50<43:30,  6.87s/it]
                                                 
{'loss': '0.01204', 'grad_norm': '0.675', 'learning_rate': '8.988e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 3990928, 'tokens/trainable': 60350, 'epoch': '0.5156'}
+
 26%|██▌       | 132/512 [14:50<43:30,  6.87s/it]
 26%|██▌       | 133/512 [14:56<43:03,  6.82s/it]
                                                 
{'loss': '0.01131', 'grad_norm': '0.5789', 'learning_rate': '8.97e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.16', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 4021264, 'tokens/trainable': 60831, 'epoch': '0.5195'}
+
 26%|██▌       | 133/512 [14:56<43:03,  6.82s/it]
 26%|██▌       | 134/512 [15:03<42:42,  6.78s/it]
                                                 
{'loss': '0.03088', 'grad_norm': '0.4706', 'learning_rate': '8.951e-05', 'ppl': '1.031', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 4051728, 'tokens/trainable': 61305, 'epoch': '0.5234'}
+
 26%|██▌       | 134/512 [15:03<42:42,  6.78s/it]
 26%|██▋       | 135/512 [15:10<42:29,  6.76s/it]
                                                 
{'loss': '0.005536', 'grad_norm': '0.1663', 'learning_rate': '8.933e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.62', 'tokens/total': 4081920, 'tokens/trainable': 61792, 'epoch': '0.5273'}
+
 26%|██▋       | 135/512 [15:10<42:29,  6.76s/it]
 27%|██▋       | 136/512 [15:16<42:18,  6.75s/it]
                                                 
{'loss': '0.002974', 'grad_norm': '0.107', 'learning_rate': '8.914e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.37', 'tokens/total': 4112304, 'tokens/trainable': 62253, 'epoch': '0.5312'}
+
 27%|██▋       | 136/512 [15:16<42:18,  6.75s/it]
 27%|██▋       | 137/512 [15:23<42:06,  6.74s/it]
                                                 
{'loss': '0.0117', 'grad_norm': '0.292', 'learning_rate': '8.895e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.55', 'tokens/train_per_sec_per_gpu': '31.65', 'tokens/total': 4142512, 'tokens/trainable': 62707, 'epoch': '0.5352'}
+
 27%|██▋       | 137/512 [15:23<42:06,  6.74s/it]
 27%|██▋       | 138/512 [15:30<41:49,  6.71s/it]
                                                 
{'loss': '0.01136', 'grad_norm': '0.3187', 'learning_rate': '8.876e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '29.32', 'tokens/total': 4172992, 'tokens/trainable': 63145, 'epoch': '0.5391'}
+
 27%|██▋       | 138/512 [15:30<41:49,  6.71s/it]
 27%|██▋       | 139/512 [15:36<41:40,  6.70s/it]
                                                 
{'loss': '0.006681', 'grad_norm': '0.1859', 'learning_rate': '8.856e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 4203136, 'tokens/trainable': 63587, 'epoch': '0.543'}
+
 27%|██▋       | 139/512 [15:36<41:40,  6.70s/it]
 27%|██▋       | 140/512 [15:43<41:32,  6.70s/it]
                                                 
{'loss': '0.01478', 'grad_norm': '0.276', 'learning_rate': '8.837e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '36.74', 'tokens/total': 4233456, 'tokens/trainable': 64052, 'epoch': '0.5469'}
+
 27%|██▋       | 140/512 [15:43<41:32,  6.70s/it]
 28%|██▊       | 141/512 [15:50<41:25,  6.70s/it]
                                                 
{'loss': '0.01677', 'grad_norm': '0.4988', 'learning_rate': '8.817e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 4263728, 'tokens/trainable': 64526, 'epoch': '0.5508'}
+
 28%|██▊       | 141/512 [15:50<41:25,  6.70s/it]
 28%|██▊       | 142/512 [15:57<41:32,  6.74s/it]
                                                 
{'loss': '0.006472', 'grad_norm': '0.1818', 'learning_rate': '8.798e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '34.58', 'tokens/total': 4294432, 'tokens/trainable': 64983, 'epoch': '0.5547'}
+
 28%|██▊       | 142/512 [15:57<41:32,  6.74s/it]
 28%|██▊       | 143/512 [16:03<41:31,  6.75s/it]
                                                 
{'loss': '0.006493', 'grad_norm': '0.2096', 'learning_rate': '8.778e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '37.58', 'tokens/total': 4324960, 'tokens/trainable': 65492, 'epoch': '0.5586'}
+
 28%|██▊       | 143/512 [16:03<41:31,  6.75s/it]
 28%|██▊       | 144/512 [16:10<41:12,  6.72s/it]
                                                 
{'loss': '0.003702', 'grad_norm': '0.1336', 'learning_rate': '8.758e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '35.72', 'tokens/total': 4355424, 'tokens/trainable': 65925, 'epoch': '0.5625'}
+
 28%|██▊       | 144/512 [16:10<41:12,  6.72s/it]
 28%|██▊       | 145/512 [16:17<41:02,  6.71s/it]
                                                 
{'loss': '0.00649', 'grad_norm': '0.164', 'learning_rate': '8.738e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '31.39', 'tokens/total': 4385712, 'tokens/trainable': 66383, 'epoch': '0.5664'}
+
 28%|██▊       | 145/512 [16:17<41:02,  6.71s/it]
 29%|██▊       | 146/512 [16:23<40:53,  6.70s/it]
                                                 
{'loss': '0.002659', 'grad_norm': '0.1365', 'learning_rate': '8.718e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.56', 'tokens/train_per_sec_per_gpu': '31.39', 'tokens/total': 4416016, 'tokens/trainable': 66840, 'epoch': '0.5703'}
+
 29%|██▊       | 146/512 [16:23<40:53,  6.70s/it]
 29%|██▊       | 147/512 [16:30<40:44,  6.70s/it]
                                                 
{'loss': '0.01334', 'grad_norm': '0.4221', 'learning_rate': '8.697e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.76', 'tokens/total': 4446368, 'tokens/trainable': 67297, 'epoch': '0.5742'}
+
 29%|██▊       | 147/512 [16:30<40:44,  6.70s/it]
 29%|██▉       | 148/512 [16:37<40:39,  6.70s/it]
                                                 
{'loss': '0.01279', 'grad_norm': '0.3735', 'learning_rate': '8.677e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '37.36', 'tokens/total': 4476880, 'tokens/trainable': 67769, 'epoch': '0.5781'}
+
 29%|██▉       | 148/512 [16:37<40:39,  6.70s/it]
 29%|██▉       | 149/512 [16:44<40:33,  6.70s/it]
                                                 
{'loss': '0.01249', 'grad_norm': '0.3093', 'learning_rate': '8.656e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.83', 'tokens/total': 4507232, 'tokens/trainable': 68257, 'epoch': '0.582'}
+
 29%|██▉       | 149/512 [16:44<40:33,  6.70s/it]
 29%|██▉       | 150/512 [16:50<40:20,  6.69s/it]
                                                 
{'loss': '0.003457', 'grad_norm': '0.2454', 'learning_rate': '8.635e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.52', 'tokens/total': 4537376, 'tokens/trainable': 68706, 'epoch': '0.5859'}
+
 29%|██▉       | 150/512 [16:50<40:20,  6.69s/it]
 29%|██▉       | 151/512 [16:57<40:11,  6.68s/it]
                                                 
{'loss': '0.009631', 'grad_norm': '0.3378', 'learning_rate': '8.615e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.3', 'tokens/total': 4567584, 'tokens/trainable': 69131, 'epoch': '0.5898'}
+
 29%|██▉       | 151/512 [16:57<40:11,  6.68s/it]
 30%|██▉       | 152/512 [17:03<40:00,  6.67s/it]
                                                 
{'loss': '0.02038', 'grad_norm': '0.348', 'learning_rate': '8.594e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.93', 'tokens/total': 4597792, 'tokens/trainable': 69586, 'epoch': '0.5938'}
+
 30%|██▉       | 152/512 [17:03<40:00,  6.67s/it]
 30%|██▉       | 153/512 [17:10<39:58,  6.68s/it]
                                                 
{'loss': '0.0008243', 'grad_norm': '0.03586', 'learning_rate': '8.572e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 4627936, 'tokens/trainable': 70061, 'epoch': '0.5977'}
+
 30%|██▉       | 153/512 [17:10<39:58,  6.68s/it]
 30%|███       | 154/512 [17:17<39:49,  6.67s/it]
                                                 
{'loss': '0.00389', 'grad_norm': '0.187', 'learning_rate': '8.551e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '27.95', 'tokens/total': 4658352, 'tokens/trainable': 70467, 'epoch': '0.6016'}
+
 30%|███       | 154/512 [17:17<39:49,  6.67s/it]
 30%|███       | 155/512 [17:24<39:46,  6.69s/it]
                                                 
{'loss': '0.005027', 'grad_norm': '0.2884', 'learning_rate': '8.53e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 4688896, 'tokens/trainable': 70933, 'epoch': '0.6055'}
+
 30%|███       | 155/512 [17:24<39:46,  6.69s/it]
 30%|███       | 156/512 [17:30<39:42,  6.69s/it]
                                                 
{'loss': '0.00246', 'grad_norm': '0.07868', 'learning_rate': '8.508e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 4719168, 'tokens/trainable': 71382, 'epoch': '0.6094'}
+
 30%|███       | 156/512 [17:30<39:42,  6.69s/it]
 31%|███       | 157/512 [17:37<39:24,  6.66s/it]
                                                 
{'loss': '0.001206', 'grad_norm': '0.05601', 'learning_rate': '8.487e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '30.83', 'tokens/total': 4749136, 'tokens/trainable': 71808, 'epoch': '0.6133'}
+
 31%|███       | 157/512 [17:37<39:24,  6.66s/it]
 31%|███       | 158/512 [17:44<39:21,  6.67s/it]
                                                 
{'loss': '0.007983', 'grad_norm': '0.3424', 'learning_rate': '8.465e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 4779472, 'tokens/trainable': 72249, 'epoch': '0.6172'}
+
 31%|███       | 158/512 [17:44<39:21,  6.67s/it]
 31%|███       | 159/512 [17:50<39:18,  6.68s/it]
                                                 
{'loss': '0.01179', 'grad_norm': '0.3605', 'learning_rate': '8.443e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 4809632, 'tokens/trainable': 72726, 'epoch': '0.6211'}
+
 31%|███       | 159/512 [17:50<39:18,  6.68s/it]
 31%|███▏      | 160/512 [17:57<39:11,  6.68s/it]
                                                 
{'loss': '0.01866', 'grad_norm': '0.3767', 'learning_rate': '8.421e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 4840112, 'tokens/trainable': 73148, 'epoch': '0.625'}
+
 31%|███▏      | 160/512 [17:57<39:11,  6.68s/it][2026-08-18 14:23:00,532] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
+
 31%|███▏      | 161/512 [18:06<42:32,  7.27s/it]
                                                 
{'loss': '0.007014', 'grad_norm': '0.2013', 'learning_rate': '8.399e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '38.65', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 4870656, 'tokens/trainable': 73600, 'epoch': '0.6289'}
+
 31%|███▏      | 161/512 [18:06<42:32,  7.27s/it]
 32%|███▏      | 162/512 [18:12<41:14,  7.07s/it]
                                                 
{'loss': '0.008246', 'grad_norm': '0.2462', 'learning_rate': '8.376e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.76', 'tokens/total': 4900608, 'tokens/trainable': 74068, 'epoch': '0.6328'}
+
 32%|███▏      | 162/512 [18:12<41:14,  7.07s/it]
 32%|███▏      | 163/512 [18:19<40:24,  6.95s/it]
                                                 
{'loss': '0.007023', 'grad_norm': '0.2069', 'learning_rate': '8.354e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '35.75', 'tokens/total': 4930752, 'tokens/trainable': 74527, 'epoch': '0.6367'}
+
 32%|███▏      | 163/512 [18:19<40:24,  6.95s/it]
 32%|███▏      | 164/512 [18:25<39:46,  6.86s/it]
                                                 
{'loss': '0.008201', 'grad_norm': '0.1888', 'learning_rate': '8.332e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '31.87', 'tokens/total': 4961008, 'tokens/trainable': 74992, 'epoch': '0.6406'}
+
 32%|███▏      | 164/512 [18:25<39:46,  6.86s/it]
 32%|███▏      | 165/512 [18:32<39:22,  6.81s/it]
                                                 
{'loss': '0.004637', 'grad_norm': '0.1746', 'learning_rate': '8.309e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 4991200, 'tokens/trainable': 75442, 'epoch': '0.6445'}
+
 32%|███▏      | 165/512 [18:32<39:22,  6.81s/it]
 32%|███▏      | 166/512 [18:39<39:06,  6.78s/it]
                                                 
{'loss': '0.007356', 'grad_norm': '0.2601', 'learning_rate': '8.286e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.27', 'tokens/total': 5021600, 'tokens/trainable': 75887, 'epoch': '0.6484'}
+
 32%|███▏      | 166/512 [18:39<39:06,  6.78s/it]
 33%|███▎      | 167/512 [18:46<38:50,  6.76s/it]
                                                 
{'loss': '0.01041', 'grad_norm': '0.254', 'learning_rate': '8.263e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 5052032, 'tokens/trainable': 76331, 'epoch': '0.6523'}
+
 33%|███▎      | 167/512 [18:46<38:50,  6.76s/it]
 33%|███▎      | 168/512 [18:52<37:44,  6.58s/it]
                                                 
{'loss': '0.002282', 'grad_norm': '0.1233', 'learning_rate': '8.24e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.92', 'tokens/total': 5080256, 'tokens/trainable': 76745, 'epoch': '0.6562'}
+
 33%|███▎      | 168/512 [18:52<37:44,  6.58s/it]
 33%|███▎      | 169/512 [18:58<37:45,  6.61s/it]
                                                 
{'loss': '0.003871', 'grad_norm': '0.11', 'learning_rate': '8.217e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 5110784, 'tokens/trainable': 77201, 'epoch': '0.6602'}
+
 33%|███▎      | 169/512 [18:58<37:45,  6.61s/it]
 33%|███▎      | 170/512 [19:05<37:46,  6.63s/it]
                                                 
{'loss': '0.0008502', 'grad_norm': '0.0368', 'learning_rate': '8.194e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.4', 'tokens/total': 5141200, 'tokens/trainable': 77655, 'epoch': '0.6641'}
+
 33%|███▎      | 170/512 [19:05<37:46,  6.63s/it]
 33%|███▎      | 171/512 [19:12<37:45,  6.64s/it]
                                                 
{'loss': '0.001807', 'grad_norm': '0.1271', 'learning_rate': '8.171e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '39.23', 'tokens/total': 5171760, 'tokens/trainable': 78122, 'epoch': '0.668'}
+
 33%|███▎      | 171/512 [19:12<37:45,  6.64s/it]
 34%|███▎      | 172/512 [19:18<37:45,  6.66s/it]
                                                 
{'loss': '0.001766', 'grad_norm': '0.1453', 'learning_rate': '8.147e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.53', 'tokens/total': 5202272, 'tokens/trainable': 78576, 'epoch': '0.6719'}
+
 34%|███▎      | 172/512 [19:19<37:45,  6.66s/it]
 34%|███▍      | 173/512 [19:25<37:42,  6.67s/it]
                                                 
{'loss': '0.002328', 'grad_norm': '0.2125', 'learning_rate': '8.124e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 5232800, 'tokens/trainable': 79059, 'epoch': '0.6758'}
+
 34%|███▍      | 173/512 [19:25<37:42,  6.67s/it]
 34%|███▍      | 174/512 [19:32<37:27,  6.65s/it]
                                                 
{'loss': '0.00771', 'grad_norm': '0.6374', 'learning_rate': '8.1e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '29.72', 'tokens/total': 5262672, 'tokens/trainable': 79498, 'epoch': '0.6797'}
+
 34%|███▍      | 174/512 [19:32<37:27,  6.65s/it]
 34%|███▍      | 175/512 [19:38<37:19,  6.65s/it]
                                                 
{'loss': '0.0007811', 'grad_norm': '0.06158', 'learning_rate': '8.076e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '30.45', 'tokens/total': 5293072, 'tokens/trainable': 79922, 'epoch': '0.6836'}
+
 34%|███▍      | 175/512 [19:38<37:19,  6.65s/it]
 34%|███▍      | 176/512 [19:45<36:26,  6.51s/it]
                                                 
{'loss': '0.0007725', 'grad_norm': '0.04236', 'learning_rate': '8.053e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.46', 'memory/max_allocated (GiB)': '33.46', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 5321520, 'tokens/trainable': 80383, 'epoch': '0.6875'}
+
 34%|███▍      | 176/512 [19:45<36:26,  6.51s/it]
 35%|███▍      | 177/512 [19:51<36:37,  6.56s/it]
                                                 
{'loss': '0.00232', 'grad_norm': '0.08921', 'learning_rate': '8.029e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 5351904, 'tokens/trainable': 80824, 'epoch': '0.6914'}
+
 35%|███▍      | 177/512 [19:51<36:37,  6.56s/it]
 35%|███▍      | 178/512 [19:58<36:39,  6.58s/it]
                                                 
{'loss': '0.008408', 'grad_norm': '0.6793', 'learning_rate': '8.005e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '32.23', 'tokens/total': 5382432, 'tokens/trainable': 81243, 'epoch': '0.6953'}
+
 35%|███▍      | 178/512 [19:58<36:39,  6.58s/it]
 35%|███▍      | 179/512 [20:05<36:42,  6.61s/it]
                                                 
{'loss': '0.0004587', 'grad_norm': '0.05572', 'learning_rate': '7.98e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 5412960, 'tokens/trainable': 81716, 'epoch': '0.6992'}
+
 35%|███▍      | 179/512 [20:05<36:42,  6.61s/it]
 35%|███▌      | 180/512 [20:11<36:46,  6.65s/it]
                                                 
{'loss': '0.00478', 'grad_norm': '0.3616', 'learning_rate': '7.956e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 5443232, 'tokens/trainable': 82181, 'epoch': '0.7031'}
+
 35%|███▌      | 180/512 [20:11<36:46,  6.65s/it]
 35%|███▌      | 181/512 [20:18<35:53,  6.51s/it]
                                                 
{'loss': '0.004499', 'grad_norm': '0.3551', 'learning_rate': '7.932e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.29', 'memory/max_allocated (GiB)': '33.29', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.65', 'tokens/total': 5471440, 'tokens/trainable': 82626, 'epoch': '0.707'}
+
 35%|███▌      | 181/512 [20:18<35:53,  6.51s/it]
 36%|███▌      | 182/512 [20:24<36:04,  6.56s/it]
                                                 
{'loss': '0.0001246', 'grad_norm': '0.007323', 'learning_rate': '7.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 5501744, 'tokens/trainable': 83089, 'epoch': '0.7109'}
+
 36%|███▌      | 182/512 [20:24<36:04,  6.56s/it]
 36%|███▌      | 183/512 [20:31<36:11,  6.60s/it]
                                                 
{'loss': '0.000127', 'grad_norm': '0.005601', 'learning_rate': '7.883e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 5532272, 'tokens/trainable': 83590, 'epoch': '0.7148'}
+
 36%|███▌      | 183/512 [20:31<36:11,  6.60s/it]
 36%|███▌      | 184/512 [20:38<36:17,  6.64s/it]
                                                 
{'loss': '0.0002659', 'grad_norm': '0.03299', 'learning_rate': '7.858e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 5562784, 'tokens/trainable': 84046, 'epoch': '0.7188'}
+
 36%|███▌      | 184/512 [20:38<36:17,  6.64s/it]
 36%|███▌      | 185/512 [20:44<36:10,  6.64s/it]
                                                 
{'loss': '0.0002492', 'grad_norm': '0.0247', 'learning_rate': '7.833e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 5592800, 'tokens/trainable': 84517, 'epoch': '0.7227'}
+
 36%|███▌      | 185/512 [20:44<36:10,  6.64s/it]
 36%|███▋      | 186/512 [20:51<36:12,  6.66s/it]
                                                 
{'loss': '0.01673', 'grad_norm': '0.6944', 'learning_rate': '7.808e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 5623088, 'tokens/trainable': 84974, 'epoch': '0.7266'}
+
 36%|███▋      | 186/512 [20:51<36:12,  6.66s/it]
 37%|███▋      | 187/512 [20:58<36:12,  6.68s/it]
                                                 
{'loss': '0.0001727', 'grad_norm': '0.01172', 'learning_rate': '7.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '35.07', 'tokens/total': 5653344, 'tokens/trainable': 85460, 'epoch': '0.7305'}
+
 37%|███▋      | 187/512 [20:58<36:12,  6.68s/it]
 37%|███▋      | 188/512 [21:04<36:05,  6.68s/it]
                                                 
{'loss': '0.004549', 'grad_norm': '0.3487', 'learning_rate': '7.758e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 5683600, 'tokens/trainable': 85939, 'epoch': '0.7344'}
+
 37%|███▋      | 188/512 [21:04<36:05,  6.68s/it]
 37%|███▋      | 189/512 [21:11<35:06,  6.52s/it]
                                                 
{'loss': '0.00147', 'grad_norm': '0.0911', 'learning_rate': '7.733e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '38.43', 'tokens/total': 5711952, 'tokens/trainable': 86377, 'epoch': '0.7383'}
+
 37%|███▋      | 189/512 [21:11<35:06,  6.52s/it]
 37%|███▋      | 190/512 [21:17<35:17,  6.58s/it]
                                                 
{'loss': '0.01117', 'grad_norm': '0.6382', 'learning_rate': '7.708e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 5742336, 'tokens/trainable': 86834, 'epoch': '0.7422'}
+
 37%|███▋      | 190/512 [21:17<35:17,  6.58s/it]
 37%|███▋      | 191/512 [21:24<35:17,  6.60s/it]
                                                 
{'loss': '0.002249', 'grad_norm': '0.1397', 'learning_rate': '7.683e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 5772560, 'tokens/trainable': 87265, 'epoch': '0.7461'}
+
 37%|███▋      | 191/512 [21:24<35:17,  6.60s/it]
 38%|███▊      | 192/512 [21:31<35:28,  6.65s/it]
                                                 
{'loss': '0.0009342', 'grad_norm': '0.09119', 'learning_rate': '7.657e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 5803136, 'tokens/trainable': 87747, 'epoch': '0.75'}
+
 38%|███▊      | 192/512 [21:31<35:28,  6.65s/it][2026-08-18 14:26:34,276] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
+
 38%|███▊      | 193/512 [21:39<38:27,  7.23s/it]
                                                 
{'loss': '0.001603', 'grad_norm': '0.1008', 'learning_rate': '7.632e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.58', 'tokens/train_per_sec_per_gpu': '36.21', 'tokens/total': 5833440, 'tokens/trainable': 88239, 'epoch': '0.7539'}
+
 38%|███▊      | 193/512 [21:39<38:27,  7.23s/it]
 38%|███▊      | 194/512 [21:46<37:30,  7.08s/it]
                                                 
{'loss': '0.0009742', 'grad_norm': '0.05941', 'learning_rate': '7.606e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 5863552, 'tokens/trainable': 88718, 'epoch': '0.7578'}
+
 38%|███▊      | 194/512 [21:46<37:30,  7.08s/it]
 38%|███▊      | 195/512 [21:53<36:42,  6.95s/it]
                                                 
{'loss': '0.001538', 'grad_norm': '0.1252', 'learning_rate': '7.58e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '31.29', 'tokens/total': 5893680, 'tokens/trainable': 89129, 'epoch': '0.7617'}
+
 38%|███▊      | 195/512 [21:53<36:42,  6.95s/it]
 38%|███▊      | 196/512 [21:59<36:08,  6.86s/it]
                                                 
{'loss': '0.003383', 'grad_norm': '0.3236', 'learning_rate': '7.555e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '32.29', 'tokens/total': 5923744, 'tokens/trainable': 89567, 'epoch': '0.7656'}
+
 38%|███▊      | 196/512 [21:59<36:08,  6.86s/it]
 38%|███▊      | 197/512 [22:06<35:41,  6.80s/it]
                                                 
{'loss': '0.009017', 'grad_norm': '0.3034', 'learning_rate': '7.529e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 5954128, 'tokens/trainable': 89988, 'epoch': '0.7695'}
+
 38%|███▊      | 197/512 [22:06<35:41,  6.80s/it]
 39%|███▊      | 198/512 [22:13<35:24,  6.77s/it]
                                                 
{'loss': '0.01389', 'grad_norm': '0.5301', 'learning_rate': '7.503e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.56', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 5984352, 'tokens/trainable': 90434, 'epoch': '0.7734'}
+
 39%|███▊      | 198/512 [22:13<35:24,  6.77s/it]
 39%|███▉      | 199/512 [22:19<35:12,  6.75s/it]
                                                 
{'loss': '0.0002993', 'grad_norm': '0.02488', 'learning_rate': '7.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.6', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 6014704, 'tokens/trainable': 90913, 'epoch': '0.7773'}
+
 39%|███▉      | 199/512 [22:19<35:12,  6.75s/it]
 39%|███▉      | 200/512 [22:26<34:55,  6.72s/it]
                                                 
{'loss': '0.001377', 'grad_norm': '0.07655', 'learning_rate': '7.451e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.66', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 6045056, 'tokens/trainable': 91355, 'epoch': '0.7812'}
+
 39%|███▉      | 200/512 [22:26<34:55,  6.72s/it]
 39%|███▉      | 201/512 [22:33<34:56,  6.74s/it]
                                                 
{'loss': '0.001292', 'grad_norm': '0.08619', 'learning_rate': '7.424e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.66', 'tokens/train_per_sec_per_gpu': '37.38', 'tokens/total': 6075504, 'tokens/trainable': 91856, 'epoch': '0.7852'}
+
 39%|███▉      | 201/512 [22:33<34:56,  6.74s/it]
 39%|███▉      | 202/512 [22:39<34:42,  6.72s/it]
                                                 
{'loss': '0.0007855', 'grad_norm': '0.07494', 'learning_rate': '7.398e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 6105904, 'tokens/trainable': 92311, 'epoch': '0.7891'}
+
 39%|███▉      | 202/512 [22:39<34:42,  6.72s/it]
 40%|███▉      | 203/512 [22:46<34:34,  6.71s/it]
                                                 
{'loss': '0.002631', 'grad_norm': '0.1252', 'learning_rate': '7.372e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '34.73', 'tokens/total': 6136272, 'tokens/trainable': 92764, 'epoch': '0.793'}
+
 40%|███▉      | 203/512 [22:46<34:34,  6.71s/it]
 40%|███▉      | 204/512 [22:53<34:19,  6.69s/it]
                                                 
{'loss': '0.004027', 'grad_norm': '0.1421', 'learning_rate': '7.345e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 6166336, 'tokens/trainable': 93227, 'epoch': '0.7969'}
+
 40%|███▉      | 204/512 [22:53<34:19,  6.69s/it]
 40%|████      | 205/512 [22:59<34:09,  6.67s/it]
                                                 
{'loss': '0.01281', 'grad_norm': '0.9565', 'learning_rate': '7.319e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 6196720, 'tokens/trainable': 93675, 'epoch': '0.8008'}
+
 40%|████      | 205/512 [22:59<34:09,  6.67s/it]
 40%|████      | 206/512 [23:06<34:09,  6.70s/it]
                                                 
{'loss': '0.003737', 'grad_norm': '0.131', 'learning_rate': '7.292e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.3', 'tokens/total': 6227424, 'tokens/trainable': 94120, 'epoch': '0.8047'}
+
 40%|████      | 206/512 [23:06<34:09,  6.70s/it]
 40%|████      | 207/512 [23:13<33:58,  6.68s/it]
                                                 
{'loss': '0.001748', 'grad_norm': '0.0838', 'learning_rate': '7.266e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 6257616, 'tokens/trainable': 94557, 'epoch': '0.8086'}
+
 40%|████      | 207/512 [23:13<33:58,  6.68s/it]
 41%|████      | 208/512 [23:20<33:54,  6.69s/it]
                                                 
{'loss': '0.001855', 'grad_norm': '0.05349', 'learning_rate': '7.239e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '40.66', 'tokens/total': 6288000, 'tokens/trainable': 95035, 'epoch': '0.8125'}
+
 41%|████      | 208/512 [23:20<33:54,  6.69s/it]
 41%|████      | 209/512 [23:26<33:46,  6.69s/it]
                                                 
{'loss': '0.002475', 'grad_norm': '0.1101', 'learning_rate': '7.212e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '37.59', 'tokens/total': 6318336, 'tokens/trainable': 95517, 'epoch': '0.8164'}
+
 41%|████      | 209/512 [23:26<33:46,  6.69s/it]
 41%|████      | 210/512 [23:33<33:39,  6.69s/it]
                                                 
{'loss': '0.001984', 'grad_norm': '0.08477', 'learning_rate': '7.185e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 6348656, 'tokens/trainable': 96026, 'epoch': '0.8203'}
+
 41%|████      | 210/512 [23:33<33:39,  6.69s/it]
 41%|████      | 211/512 [23:40<33:29,  6.68s/it]
                                                 
{'loss': '0.001189', 'grad_norm': '0.06026', 'learning_rate': '7.158e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '35.64', 'tokens/total': 6379040, 'tokens/trainable': 96477, 'epoch': '0.8242'}
+
 41%|████      | 211/512 [23:40<33:29,  6.68s/it]
 41%|████▏     | 212/512 [23:46<33:21,  6.67s/it]
                                                 
{'loss': '0.001786', 'grad_norm': '0.07388', 'learning_rate': '7.131e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.17', 'tokens/total': 6409312, 'tokens/trainable': 96951, 'epoch': '0.8281'}
+
 41%|████▏     | 212/512 [23:46<33:21,  6.67s/it]
 42%|████▏     | 213/512 [23:53<33:13,  6.67s/it]
                                                 
{'loss': '0.00198', 'grad_norm': '0.1013', 'learning_rate': '7.104e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '30.81', 'tokens/total': 6439520, 'tokens/trainable': 97350, 'epoch': '0.832'}
+
 42%|████▏     | 213/512 [23:53<33:13,  6.67s/it]
 42%|████▏     | 214/512 [24:00<33:05,  6.66s/it]
                                                 
{'loss': '0.0001322', 'grad_norm': '0.009713', 'learning_rate': '7.077e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '29.91', 'tokens/total': 6469888, 'tokens/trainable': 97792, 'epoch': '0.8359'}
+
 42%|████▏     | 214/512 [24:00<33:05,  6.66s/it]
 42%|████▏     | 215/512 [24:06<33:02,  6.67s/it]
                                                 
{'loss': '0.00157', 'grad_norm': '0.1457', 'learning_rate': '7.05e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 6500128, 'tokens/trainable': 98257, 'epoch': '0.8398'}
+
 42%|████▏     | 215/512 [24:06<33:02,  6.67s/it]
 42%|████▏     | 216/512 [24:13<33:02,  6.70s/it]
                                                 
{'loss': '0.0003705', 'grad_norm': '0.02122', 'learning_rate': '7.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 6530832, 'tokens/trainable': 98731, 'epoch': '0.8438'}
+
 42%|████▏     | 216/512 [24:13<33:02,  6.70s/it]
 42%|████▏     | 217/512 [24:20<32:54,  6.69s/it]
                                                 
{'loss': '0.003762', 'grad_norm': '0.2245', 'learning_rate': '6.995e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.5', 'tokens/total': 6561248, 'tokens/trainable': 99212, 'epoch': '0.8477'}
+
 42%|████▏     | 217/512 [24:20<32:54,  6.69s/it]
 43%|████▎     | 218/512 [24:26<32:41,  6.67s/it]
                                                 
{'loss': '0.001126', 'grad_norm': '0.6341', 'learning_rate': '6.968e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 6591648, 'tokens/trainable': 99654, 'epoch': '0.8516'}
+
 43%|████▎     | 218/512 [24:26<32:41,  6.67s/it]
 43%|████▎     | 219/512 [24:33<32:39,  6.69s/it]
                                                 
{'loss': '0.001478', 'grad_norm': '0.1625', 'learning_rate': '6.94e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 6622368, 'tokens/trainable': 100086, 'epoch': '0.8555'}
+
 43%|████▎     | 219/512 [24:33<32:39,  6.69s/it]
 43%|████▎     | 220/512 [24:40<32:30,  6.68s/it]
                                                 
{'loss': '0.0001898', 'grad_norm': '0.01793', 'learning_rate': '6.913e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.34', 'tokens/total': 6652848, 'tokens/trainable': 100524, 'epoch': '0.8594'}
+
 43%|████▎     | 220/512 [24:40<32:30,  6.68s/it]
 43%|████▎     | 221/512 [24:46<32:30,  6.70s/it]
                                                 
{'loss': '0.0008955', 'grad_norm': '0.1249', 'learning_rate': '6.885e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.46', 'tokens/total': 6683392, 'tokens/trainable': 100996, 'epoch': '0.8633'}
+
 43%|████▎     | 221/512 [24:46<32:30,  6.70s/it]
 43%|████▎     | 222/512 [24:53<32:19,  6.69s/it]
                                                 
{'loss': '0.0001049', 'grad_norm': '0.01509', 'learning_rate': '6.857e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 6713712, 'tokens/trainable': 101449, 'epoch': '0.8672'}
+
 43%|████▎     | 222/512 [24:53<32:19,  6.69s/it]
 44%|████▎     | 223/512 [25:00<32:08,  6.67s/it]
                                                 
{'loss': '0.01887', 'grad_norm': '0.7131', 'learning_rate': '6.83e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.14', 'tokens/total': 6744176, 'tokens/trainable': 101905, 'epoch': '0.8711'}
+
 44%|████▎     | 223/512 [25:00<32:08,  6.67s/it]
 44%|████▍     | 224/512 [25:06<32:06,  6.69s/it]
                                                 
{'loss': '7.858e-05', 'grad_norm': '0.008568', 'learning_rate': '6.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '36.43', 'tokens/total': 6774416, 'tokens/trainable': 102411, 'epoch': '0.875'}
+
 44%|████▍     | 224/512 [25:06<32:06,  6.69s/it][2026-08-18 14:30:10,034] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
+
 44%|████▍     | 225/512 [25:15<34:39,  7.24s/it]
                                                 
{'loss': '0.01011', 'grad_norm': '0.6807', 'learning_rate': '6.774e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.19', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 6804592, 'tokens/trainable': 102882, 'epoch': '0.8789'}
+
 44%|████▍     | 225/512 [25:15<34:39,  7.24s/it]
 44%|████▍     | 226/512 [25:22<33:41,  7.07s/it]
                                                 
{'loss': '5.371e-05', 'grad_norm': '0.004204', 'learning_rate': '6.746e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.49', 'tokens/train_per_sec_per_gpu': '34.69', 'tokens/total': 6834976, 'tokens/trainable': 103332, 'epoch': '0.8828'}
+
 44%|████▍     | 226/512 [25:22<33:41,  7.07s/it]
 44%|████▍     | 227/512 [25:28<33:03,  6.96s/it]
                                                 
{'loss': '0.0003807', 'grad_norm': '0.0261', 'learning_rate': '6.718e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 6865408, 'tokens/trainable': 103790, 'epoch': '0.8867'}
+
 44%|████▍     | 227/512 [25:28<33:03,  6.96s/it]
 45%|████▍     | 228/512 [25:35<32:33,  6.88s/it]
                                                 
{'loss': '6.497e-05', 'grad_norm': '0.002807', 'learning_rate': '6.69e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 6895664, 'tokens/trainable': 104246, 'epoch': '0.8906'}
+
 45%|████▍     | 228/512 [25:35<32:33,  6.88s/it]
 45%|████▍     | 229/512 [25:42<32:09,  6.82s/it]
                                                 
{'loss': '0.01367', 'grad_norm': '0.8277', 'learning_rate': '6.662e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.72', 'tokens/total': 6926016, 'tokens/trainable': 104707, 'epoch': '0.8945'}
+
 45%|████▍     | 229/512 [25:42<32:09,  6.82s/it]
 45%|████▍     | 230/512 [25:48<31:53,  6.79s/it]
                                                 
{'loss': '0.0005469', 'grad_norm': '0.08311', 'learning_rate': '6.634e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.84', 'tokens/total': 6956384, 'tokens/trainable': 105167, 'epoch': '0.8984'}
+
 45%|████▍     | 230/512 [25:48<31:53,  6.79s/it]
 45%|████▌     | 231/512 [25:55<31:33,  6.74s/it]
                                                 
{'loss': '0.0001304', 'grad_norm': '0.008128', 'learning_rate': '6.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 6986768, 'tokens/trainable': 105576, 'epoch': '0.9023'}
+
 45%|████▌     | 231/512 [25:55<31:33,  6.74s/it]
 45%|████▌     | 232/512 [26:02<31:20,  6.72s/it]
                                                 
{'loss': '0.0003516', 'grad_norm': '0.03313', 'learning_rate': '6.578e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.76', 'tokens/total': 7017232, 'tokens/trainable': 106043, 'epoch': '0.9062'}
+
 45%|████▌     | 232/512 [26:02<31:20,  6.72s/it]
 46%|████▌     | 233/512 [26:08<31:11,  6.71s/it]
                                                 
{'loss': '0.001037', 'grad_norm': '0.0869', 'learning_rate': '6.55e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 7047568, 'tokens/trainable': 106506, 'epoch': '0.9102'}
+
 46%|████▌     | 233/512 [26:08<31:11,  6.71s/it]
 46%|████▌     | 234/512 [26:15<30:59,  6.69s/it]
                                                 
{'loss': '0.001943', 'grad_norm': '0.129', 'learning_rate': '6.522e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 7077760, 'tokens/trainable': 106951, 'epoch': '0.9141'}
+
 46%|████▌     | 234/512 [26:15<30:59,  6.69s/it]
 46%|████▌     | 235/512 [26:22<30:46,  6.66s/it]
                                                 
{'loss': '0.002509', 'grad_norm': '0.2505', 'learning_rate': '6.493e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.73', 'tokens/total': 7107760, 'tokens/trainable': 107382, 'epoch': '0.918'}
+
 46%|████▌     | 235/512 [26:22<30:46,  6.66s/it]
 46%|████▌     | 236/512 [26:28<30:42,  6.68s/it]
                                                 
{'loss': '0.008919', 'grad_norm': '0.5062', 'learning_rate': '6.465e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.82', 'tokens/total': 7138144, 'tokens/trainable': 107814, 'epoch': '0.9219'}
+
 46%|████▌     | 236/512 [26:28<30:42,  6.68s/it]
 46%|████▋     | 237/512 [26:35<30:37,  6.68s/it]
                                                 
{'loss': '0.004196', 'grad_norm': '0.3209', 'learning_rate': '6.437e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 7168352, 'tokens/trainable': 108274, 'epoch': '0.9258'}
+
 46%|████▋     | 237/512 [26:35<30:37,  6.68s/it]
 46%|████▋     | 238/512 [26:42<30:32,  6.69s/it]
                                                 
{'loss': '0.003744', 'grad_norm': '0.3213', 'learning_rate': '6.408e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.09', 'tokens/total': 7198624, 'tokens/trainable': 108714, 'epoch': '0.9297'}
+
 46%|████▋     | 238/512 [26:42<30:32,  6.69s/it]
 47%|████▋     | 239/512 [26:48<30:27,  6.69s/it]
                                                 
{'loss': '0.0003515', 'grad_norm': '0.02197', 'learning_rate': '6.38e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 7229216, 'tokens/trainable': 109220, 'epoch': '0.9336'}
+
 47%|████▋     | 239/512 [26:48<30:27,  6.69s/it]
 47%|████▋     | 240/512 [26:55<30:20,  6.69s/it]
                                                 
{'loss': '0.001268', 'grad_norm': '0.1164', 'learning_rate': '6.351e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32', 'tokens/total': 7259648, 'tokens/trainable': 109672, 'epoch': '0.9375'}
+
 47%|████▋     | 240/512 [26:55<30:20,  6.69s/it]
 47%|████▋     | 241/512 [27:02<30:14,  6.69s/it]
                                                 
{'loss': '0.002464', 'grad_norm': '0.1797', 'learning_rate': '6.323e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.09', 'tokens/total': 7289824, 'tokens/trainable': 110135, 'epoch': '0.9414'}
+
 47%|████▋     | 241/512 [27:02<30:14,  6.69s/it]
 47%|████▋     | 242/512 [27:09<30:08,  6.70s/it]
                                                 
{'loss': '0.01327', 'grad_norm': '0.5711', 'learning_rate': '6.294e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.6', 'tokens/total': 7320288, 'tokens/trainable': 110563, 'epoch': '0.9453'}
+
 47%|████▋     | 242/512 [27:09<30:08,  6.70s/it]
 47%|████▋     | 243/512 [27:15<30:04,  6.71s/it]
                                                 
{'loss': '0.01847', 'grad_norm': '0.782', 'learning_rate': '6.266e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.84', 'tokens/total': 7350736, 'tokens/trainable': 111049, 'epoch': '0.9492'}
+
 47%|████▋     | 243/512 [27:15<30:04,  6.71s/it]
 48%|████▊     | 244/512 [27:22<30:06,  6.74s/it]
                                                 
{'loss': '0.003869', 'grad_norm': '0.2531', 'learning_rate': '6.237e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 7381280, 'tokens/trainable': 111495, 'epoch': '0.9531'}
+
 48%|████▊     | 244/512 [27:22<30:06,  6.74s/it]
 48%|████▊     | 245/512 [27:29<29:58,  6.74s/it]
                                                 
{'loss': '0.001095', 'grad_norm': '0.09388', 'learning_rate': '6.208e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.68', 'tokens/total': 7411632, 'tokens/trainable': 111968, 'epoch': '0.957'}
+
 48%|████▊     | 245/512 [27:29<29:58,  6.74s/it]
 48%|████▊     | 246/512 [27:35<29:44,  6.71s/it]
                                                 
{'loss': '0.0006574', 'grad_norm': '0.06202', 'learning_rate': '6.18e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 7441904, 'tokens/trainable': 112416, 'epoch': '0.9609'}
+
 48%|████▊     | 246/512 [27:35<29:44,  6.71s/it]
 48%|████▊     | 247/512 [27:42<29:41,  6.72s/it]
                                                 
{'loss': '0.002766', 'grad_norm': '0.2164', 'learning_rate': '6.151e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 7472368, 'tokens/trainable': 112882, 'epoch': '0.9648'}
+
 48%|████▊     | 247/512 [27:42<29:41,  6.72s/it]
 48%|████▊     | 248/512 [27:49<29:34,  6.72s/it]
                                                 
{'loss': '0.001269', 'grad_norm': '0.04909', 'learning_rate': '6.122e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '40.02', 'tokens/total': 7502656, 'tokens/trainable': 113390, 'epoch': '0.9688'}
+
 48%|████▊     | 248/512 [27:49<29:34,  6.72s/it]
 49%|████▊     | 249/512 [27:56<29:26,  6.72s/it]
                                                 
{'loss': '0.0115', 'grad_norm': '0.2183', 'learning_rate': '6.093e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 7532800, 'tokens/trainable': 113904, 'epoch': '0.9727'}
+
 49%|████▊     | 249/512 [27:56<29:26,  6.72s/it]
 49%|████▉     | 250/512 [28:02<29:20,  6.72s/it]
                                                 
{'loss': '0.0009117', 'grad_norm': '0.05114', 'learning_rate': '6.065e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 7563264, 'tokens/trainable': 114351, 'epoch': '0.9766'}
+
 49%|████▉     | 250/512 [28:02<29:20,  6.72s/it]
 49%|████▉     | 251/512 [28:09<29:14,  6.72s/it]
                                                 
{'loss': '0.001624', 'grad_norm': '0.1029', 'learning_rate': '6.036e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 7593840, 'tokens/trainable': 114842, 'epoch': '0.9805'}
+
 49%|████▉     | 251/512 [28:09<29:14,  6.72s/it]
 49%|████▉     | 252/512 [28:16<29:04,  6.71s/it]
                                                 
{'loss': '0.001105', 'grad_norm': '0.0578', 'learning_rate': '6.007e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '28.88', 'tokens/total': 7624416, 'tokens/trainable': 115263, 'epoch': '0.9844'}
+
 49%|████▉     | 252/512 [28:16<29:04,  6.71s/it]
 49%|████▉     | 253/512 [28:22<28:57,  6.71s/it]
                                                 
{'loss': '0.0003236', 'grad_norm': '0.0175', 'learning_rate': '5.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '32.66', 'tokens/total': 7654688, 'tokens/trainable': 115703, 'epoch': '0.9883'}
+
 49%|████▉     | 253/512 [28:22<28:57,  6.71s/it]
 50%|████▉     | 254/512 [28:29<28:46,  6.69s/it]
                                                 
{'loss': '0.003511', 'grad_norm': '0.2686', 'learning_rate': '5.949e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '33.26', 'tokens/total': 7685088, 'tokens/trainable': 116157, 'epoch': '0.9922'}
+
 50%|████▉     | 254/512 [28:29<28:46,  6.69s/it]
 50%|████▉     | 255/512 [28:36<28:37,  6.68s/it]
                                                 
{'loss': '0.002814', 'grad_norm': '0.2503', 'learning_rate': '5.92e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '31.95', 'tokens/total': 7715552, 'tokens/trainable': 116567, 'epoch': '0.9961'}
+
 50%|████▉     | 255/512 [28:36<28:37,  6.68s/it]
 50%|█████     | 256/512 [28:43<28:43,  6.73s/it]
                                                 
{'loss': '0.0002322', 'grad_norm': '0.01063', 'learning_rate': '5.891e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '31.11', 'tokens/total': 7745872, 'tokens/trainable': 117030, 'epoch': '1'}
+
 50%|█████     | 256/512 [28:43<28:43,  6.73s/it][2026-08-18 14:33:46,258] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
+
 50%|█████     | 257/512 [28:52<31:38,  7.44s/it]
                                                 
{'loss': '8.973e-05', 'grad_norm': '0.004323', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.23', 'tokens/train_per_sec_per_gpu': '34.73', 'tokens/total': 7776208, 'tokens/trainable': 117517, 'epoch': '1.004'}
+
 50%|█████     | 257/512 [28:52<31:38,  7.44s/it]
 50%|█████     | 258/512 [28:58<30:34,  7.22s/it]
                                                 
{'loss': '0.001043', 'grad_norm': '0.04856', 'learning_rate': '5.834e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '38.33', 'tokens/total': 7806480, 'tokens/trainable': 118015, 'epoch': '1.008'}
+
 50%|█████     | 258/512 [28:58<30:34,  7.22s/it]
 51%|█████     | 259/512 [29:05<29:48,  7.07s/it]
                                                 
{'loss': '0.0002961', 'grad_norm': '0.01527', 'learning_rate': '5.805e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.62', 'tokens/train_per_sec_per_gpu': '36.2', 'tokens/total': 7836800, 'tokens/trainable': 118520, 'epoch': '1.012'}
+
 51%|█████     | 259/512 [29:05<29:48,  7.07s/it]
 51%|█████     | 260/512 [29:12<29:15,  6.97s/it]
                                                 
{'loss': '0.001453', 'grad_norm': '0.1239', 'learning_rate': '5.776e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 7867040, 'tokens/trainable': 118992, 'epoch': '1.016'}
+
 51%|█████     | 260/512 [29:12<29:15,  6.97s/it]
 51%|█████     | 261/512 [29:19<28:49,  6.89s/it]
                                                 
{'loss': '0.0002353', 'grad_norm': '0.008926', 'learning_rate': '5.747e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.65', 'tokens/total': 7897408, 'tokens/trainable': 119438, 'epoch': '1.02'}
+
 51%|█████     | 261/512 [29:19<28:49,  6.89s/it]
 51%|█████     | 262/512 [29:25<28:26,  6.82s/it]
                                                 
{'loss': '0.001416', 'grad_norm': '0.08671', 'learning_rate': '5.718e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 7927648, 'tokens/trainable': 119881, 'epoch': '1.023'}
+
 51%|█████     | 262/512 [29:25<28:26,  6.82s/it]
 51%|█████▏    | 263/512 [29:32<28:08,  6.78s/it]
                                                 
{'loss': '0.0004944', 'grad_norm': '0.03644', 'learning_rate': '5.689e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '27.72', 'tokens/total': 7957968, 'tokens/trainable': 120285, 'epoch': '1.027'}
+
 51%|█████▏    | 263/512 [29:32<28:08,  6.78s/it]
 52%|█████▏    | 264/512 [29:39<27:52,  6.74s/it]
                                                 
{'loss': '0.005234', 'grad_norm': '0.4136', 'learning_rate': '5.66e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.65', 'memory/max_allocated (GiB)': '33.65', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 7987888, 'tokens/trainable': 120755, 'epoch': '1.031'}
+
 52%|█████▏    | 264/512 [29:39<27:52,  6.74s/it]
 52%|█████▏    | 265/512 [29:45<27:39,  6.72s/it]
                                                 
{'loss': '0.0001969', 'grad_norm': '0.01111', 'learning_rate': '5.631e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 8018048, 'tokens/trainable': 121194, 'epoch': '1.035'}
+
 52%|█████▏    | 265/512 [29:45<27:39,  6.72s/it]
 52%|█████▏    | 266/512 [29:52<27:32,  6.72s/it]
                                                 
{'loss': '0.005357', 'grad_norm': '0.2199', 'learning_rate': '5.602e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.54', 'tokens/total': 8048448, 'tokens/trainable': 121670, 'epoch': '1.039'}
+
 52%|█████▏    | 266/512 [29:52<27:32,  6.72s/it]
 52%|█████▏    | 267/512 [29:59<27:28,  6.73s/it]
                                                 
{'loss': '0.002412', 'grad_norm': '0.151', 'learning_rate': '5.573e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '33.58', 'tokens/total': 8078848, 'tokens/trainable': 122142, 'epoch': '1.043'}
+
 52%|█████▏    | 267/512 [29:59<27:28,  6.73s/it]
 52%|█████▏    | 268/512 [30:06<27:24,  6.74s/it]
                                                 
{'loss': '4.619e-05', 'grad_norm': '0.002497', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.36', 'tokens/total': 8109360, 'tokens/trainable': 122585, 'epoch': '1.047'}
+
 52%|█████▏    | 268/512 [30:06<27:24,  6.74s/it]
 53%|█████▎    | 269/512 [30:12<27:17,  6.74s/it]
                                                 
{'loss': '0.0002445', 'grad_norm': '0.0187', 'learning_rate': '5.515e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '38.78', 'tokens/total': 8139888, 'tokens/trainable': 123091, 'epoch': '1.051'}
+
 53%|█████▎    | 269/512 [30:12<27:17,  6.74s/it]
 53%|█████▎    | 270/512 [30:19<27:17,  6.77s/it]
                                                 
{'loss': '0.000507', 'grad_norm': '0.04384', 'learning_rate': '5.485e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '30.91', 'tokens/total': 8170480, 'tokens/trainable': 123505, 'epoch': '1.055'}
+
 53%|█████▎    | 270/512 [30:19<27:17,  6.77s/it]
 53%|█████▎    | 271/512 [30:25<26:31,  6.61s/it]
                                                 
{'loss': '0.0002814', 'grad_norm': '0.01803', 'learning_rate': '5.456e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.41', 'memory/max_allocated (GiB)': '33.41', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '35.31', 'tokens/total': 8198848, 'tokens/trainable': 123953, 'epoch': '1.059'}
+
 53%|█████▎    | 271/512 [30:25<26:31,  6.61s/it]
 53%|█████▎    | 272/512 [30:32<26:35,  6.65s/it]
                                                 
{'loss': '0.0001505', 'grad_norm': '0.01605', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '28.61', 'tokens/total': 8229200, 'tokens/trainable': 124400, 'epoch': '1.062'}
+
 53%|█████▎    | 272/512 [30:32<26:35,  6.65s/it]
 53%|█████▎    | 273/512 [30:39<26:32,  6.66s/it]
                                                 
{'loss': '7.019e-05', 'grad_norm': '0.004128', 'learning_rate': '5.398e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 8259536, 'tokens/trainable': 124870, 'epoch': '1.066'}
+
 53%|█████▎    | 273/512 [30:39<26:32,  6.66s/it]
 54%|█████▎    | 274/512 [30:45<26:26,  6.67s/it]
                                                 
{'loss': '0.0002327', 'grad_norm': '0.01637', 'learning_rate': '5.369e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 8289904, 'tokens/trainable': 125333, 'epoch': '1.07'}
+
 54%|█████▎    | 274/512 [30:45<26:26,  6.67s/it]
 54%|█████▎    | 275/512 [30:52<26:25,  6.69s/it]
                                                 
{'loss': '2.309e-05', 'grad_norm': '0.0009976', 'learning_rate': '5.34e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 8320176, 'tokens/trainable': 125834, 'epoch': '1.074'}
+
 54%|█████▎    | 275/512 [30:52<26:25,  6.69s/it]
 54%|█████▍    | 276/512 [30:59<26:19,  6.69s/it]
                                                 
{'loss': '2.291e-05', 'grad_norm': '0.002002', 'learning_rate': '5.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.02', 'tokens/total': 8350832, 'tokens/trainable': 126316, 'epoch': '1.078'}
+
 54%|█████▍    | 276/512 [30:59<26:19,  6.69s/it]
 54%|█████▍    | 277/512 [31:06<26:14,  6.70s/it]
                                                 
{'loss': '6.588e-05', 'grad_norm': '0.005128', 'learning_rate': '5.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.33', 'tokens/total': 8380976, 'tokens/trainable': 126798, 'epoch': '1.082'}
+
 54%|█████▍    | 277/512 [31:06<26:14,  6.70s/it]
 54%|█████▍    | 278/512 [31:12<25:59,  6.66s/it]
                                                 
{'loss': '2.662e-05', 'grad_norm': '0.001301', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.64', 'tokens/total': 8411072, 'tokens/trainable': 127223, 'epoch': '1.086'}
+
 54%|█████▍    | 278/512 [31:12<25:59,  6.66s/it]
 54%|█████▍    | 279/512 [31:19<25:41,  6.62s/it]
                                                 
{'loss': '2.588e-05', 'grad_norm': '0.001754', 'learning_rate': '5.224e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 8440736, 'tokens/trainable': 127672, 'epoch': '1.09'}
+
 54%|█████▍    | 279/512 [31:19<25:41,  6.62s/it]
 55%|█████▍    | 280/512 [31:25<25:36,  6.62s/it]
                                                 
{'loss': '8.053e-05', 'grad_norm': '0.005083', 'learning_rate': '5.195e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 8470944, 'tokens/trainable': 128116, 'epoch': '1.094'}
+
 55%|█████▍    | 280/512 [31:25<25:36,  6.62s/it]
 55%|█████▍    | 281/512 [31:32<25:35,  6.65s/it]
                                                 
{'loss': '2.603e-05', 'grad_norm': '0.001439', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.82', 'tokens/total': 8501104, 'tokens/trainable': 128589, 'epoch': '1.098'}
+
 55%|█████▍    | 281/512 [31:32<25:35,  6.65s/it]
 55%|█████▌    | 282/512 [31:39<25:30,  6.66s/it]
                                                 
{'loss': '0.008352', 'grad_norm': '0.7785', 'learning_rate': '5.138e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 8531696, 'tokens/trainable': 129036, 'epoch': '1.102'}
+
 55%|█████▌    | 282/512 [31:39<25:30,  6.66s/it]
 55%|█████▌    | 283/512 [31:45<25:27,  6.67s/it]
                                                 
{'loss': '3.522e-05', 'grad_norm': '0.002119', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 8561936, 'tokens/trainable': 129485, 'epoch': '1.105'}
+
 55%|█████▌    | 283/512 [31:45<25:27,  6.67s/it]
 55%|█████▌    | 284/512 [31:52<25:23,  6.68s/it]
                                                 
{'loss': '0.0002057', 'grad_norm': '0.01936', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.55', 'tokens/total': 8592192, 'tokens/trainable': 129920, 'epoch': '1.109'}
+
 55%|█████▌    | 284/512 [31:52<25:23,  6.68s/it]
 56%|█████▌    | 285/512 [31:59<25:19,  6.69s/it]
                                                 
{'loss': '0.0002226', 'grad_norm': '0.03079', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.14', 'tokens/total': 8622432, 'tokens/trainable': 130417, 'epoch': '1.113'}
+
 56%|█████▌    | 285/512 [31:59<25:19,  6.69s/it]
 56%|█████▌    | 286/512 [32:06<25:13,  6.70s/it]
                                                 
{'loss': '0.002872', 'grad_norm': '0.2944', 'learning_rate': '5.022e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.95', 'tokens/total': 8652768, 'tokens/trainable': 130903, 'epoch': '1.117'}
+
 56%|█████▌    | 286/512 [32:06<25:13,  6.70s/it]
 56%|█████▌    | 287/512 [32:12<25:05,  6.69s/it]
                                                 
{'loss': '2.243e-05', 'grad_norm': '0.001503', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.5', 'tokens/total': 8683296, 'tokens/trainable': 131343, 'epoch': '1.121'}
+
 56%|█████▌    | 287/512 [32:12<25:05,  6.69s/it]
 56%|█████▋    | 288/512 [32:19<24:57,  6.69s/it]
                                                 
{'loss': '0.002656', 'grad_norm': '0.2223', 'learning_rate': '4.964e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.23', 'tokens/total': 8713504, 'tokens/trainable': 131763, 'epoch': '1.125'}
+
 56%|█████▋    | 288/512 [32:19<24:57,  6.69s/it][2026-08-18 14:37:22,492] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
+
 56%|█████▋    | 289/512 [32:28<27:16,  7.34s/it]
                                                 
{'loss': '4.355e-05', 'grad_norm': '0.019', 'learning_rate': '4.935e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 8743888, 'tokens/trainable': 132229, 'epoch': '1.129'}
+
 56%|█████▋    | 289/512 [32:28<27:16,  7.34s/it]
 57%|█████▋    | 290/512 [32:34<26:28,  7.16s/it]
                                                 
{'loss': '2.45e-05', 'grad_norm': '0.00189', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.54', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 8774336, 'tokens/trainable': 132678, 'epoch': '1.133'}
+
 57%|█████▋    | 290/512 [32:34<26:28,  7.16s/it]
 57%|█████▋    | 291/512 [32:41<25:53,  7.03s/it]
                                                 
{'loss': '2.067e-05', 'grad_norm': '0.0009668', 'learning_rate': '4.878e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.28', 'tokens/total': 8804816, 'tokens/trainable': 133136, 'epoch': '1.137'}
+
 57%|█████▋    | 291/512 [32:41<25:53,  7.03s/it]
 57%|█████▋    | 292/512 [32:48<25:26,  6.94s/it]
                                                 
{'loss': '2.282e-05', 'grad_norm': '0.001047', 'learning_rate': '4.849e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 8835024, 'tokens/trainable': 133608, 'epoch': '1.141'}
+
 57%|█████▋    | 292/512 [32:48<25:26,  6.94s/it]
 57%|█████▋    | 293/512 [32:55<25:02,  6.86s/it]
                                                 
{'loss': '0.0002425', 'grad_norm': '0.0286', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '40.43', 'tokens/total': 8865408, 'tokens/trainable': 134108, 'epoch': '1.145'}
+
 57%|█████▋    | 293/512 [32:55<25:02,  6.86s/it]
 57%|█████▋    | 294/512 [33:01<24:46,  6.82s/it]
                                                 
{'loss': '3.146e-05', 'grad_norm': '0.002422', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.68', 'tokens/total': 8895824, 'tokens/trainable': 134541, 'epoch': '1.148'}
+
 57%|█████▋    | 294/512 [33:01<24:46,  6.82s/it]
 58%|█████▊    | 295/512 [33:08<24:31,  6.78s/it]
                                                 
{'loss': '5.829e-05', 'grad_norm': '0.003736', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '29.44', 'tokens/total': 8926336, 'tokens/trainable': 134966, 'epoch': '1.152'}
+
 58%|█████▊    | 295/512 [33:08<24:31,  6.78s/it]
 58%|█████▊    | 296/512 [33:15<24:16,  6.74s/it]
                                                 
{'loss': '0.0002222', 'grad_norm': '0.02378', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '33.53', 'tokens/total': 8956656, 'tokens/trainable': 135402, 'epoch': '1.156'}
+
 58%|█████▊    | 296/512 [33:15<24:16,  6.74s/it]
 58%|█████▊    | 297/512 [33:21<24:00,  6.70s/it]
                                                 
{'loss': '0.0005197', 'grad_norm': '0.03584', 'learning_rate': '4.706e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '30.02', 'tokens/total': 8986752, 'tokens/trainable': 135804, 'epoch': '1.16'}
+
 58%|█████▊    | 297/512 [33:21<24:00,  6.70s/it]
 58%|█████▊    | 298/512 [33:28<23:48,  6.67s/it]
                                                 
{'loss': '0.0007586', 'grad_norm': '0.1083', 'learning_rate': '4.677e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '31.13', 'tokens/total': 9017024, 'tokens/trainable': 136231, 'epoch': '1.164'}
+
 58%|█████▊    | 298/512 [33:28<23:48,  6.67s/it]
 58%|█████▊    | 299/512 [33:35<23:43,  6.68s/it]
                                                 
{'loss': '1.649e-05', 'grad_norm': '0.0007778', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.55', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 9047424, 'tokens/trainable': 136705, 'epoch': '1.168'}
+
 58%|█████▊    | 299/512 [33:35<23:43,  6.68s/it]
 59%|█████▊    | 300/512 [33:41<23:34,  6.67s/it]
                                                 
{'loss': '3.358e-05', 'grad_norm': '0.002182', 'learning_rate': '4.62e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.73', 'tokens/total': 9077648, 'tokens/trainable': 137178, 'epoch': '1.172'}
+
 59%|█████▊    | 300/512 [33:41<23:34,  6.67s/it]
 59%|█████▉    | 301/512 [33:48<23:30,  6.68s/it]
                                                 
{'loss': '9.5e-05', 'grad_norm': '0.01777', 'learning_rate': '4.592e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 9108032, 'tokens/trainable': 137643, 'epoch': '1.176'}
+
 59%|█████▉    | 301/512 [33:48<23:30,  6.68s/it]
 59%|█████▉    | 302/512 [33:55<23:23,  6.68s/it]
                                                 
{'loss': '0.001084', 'grad_norm': '0.08603', 'learning_rate': '4.563e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '31.91', 'tokens/total': 9138240, 'tokens/trainable': 138078, 'epoch': '1.18'}
+
 59%|█████▉    | 302/512 [33:55<23:23,  6.68s/it]
 59%|█████▉    | 303/512 [34:01<23:15,  6.67s/it]
                                                 
{'loss': '3.201e-05', 'grad_norm': '0.001786', 'learning_rate': '4.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '35', 'tokens/total': 9168352, 'tokens/trainable': 138515, 'epoch': '1.184'}
+
 59%|█████▉    | 303/512 [34:01<23:15,  6.67s/it]
 59%|█████▉    | 304/512 [34:08<23:11,  6.69s/it]
                                                 
{'loss': '2.666e-05', 'grad_norm': '0.001189', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '35.13', 'tokens/total': 9198864, 'tokens/trainable': 138948, 'epoch': '1.188'}
+
 59%|█████▉    | 304/512 [34:08<23:11,  6.69s/it]
 60%|█████▉    | 305/512 [34:15<23:05,  6.69s/it]
                                                 
{'loss': '0.0008552', 'grad_norm': '0.08659', 'learning_rate': '4.478e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '38.05', 'tokens/total': 9229104, 'tokens/trainable': 139408, 'epoch': '1.191'}
+
 60%|█████▉    | 305/512 [34:15<23:05,  6.69s/it]
 60%|█████▉    | 306/512 [34:21<22:26,  6.54s/it]
                                                 
{'loss': '2.396e-05', 'grad_norm': '0.001476', 'learning_rate': '4.45e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '33.38', 'tokens/total': 9257344, 'tokens/trainable': 139840, 'epoch': '1.195'}
+
 60%|█████▉    | 306/512 [34:21<22:26,  6.54s/it]
 60%|█████▉    | 307/512 [34:28<22:29,  6.58s/it]
                                                 
{'loss': '0.0001232', 'grad_norm': '0.01435', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 9287520, 'tokens/trainable': 140281, 'epoch': '1.199'}
+
 60%|█████▉    | 307/512 [34:28<22:29,  6.58s/it]
 60%|██████    | 308/512 [34:34<22:30,  6.62s/it]
                                                 
{'loss': '2.204e-05', 'grad_norm': '0.001902', 'learning_rate': '4.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '37.31', 'tokens/total': 9317888, 'tokens/trainable': 140776, 'epoch': '1.203'}
+
 60%|██████    | 308/512 [34:34<22:30,  6.62s/it]
 60%|██████    | 309/512 [34:41<22:27,  6.64s/it]
                                                 
{'loss': '2.58e-05', 'grad_norm': '0.001532', 'learning_rate': '4.366e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.72', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 9348256, 'tokens/trainable': 141228, 'epoch': '1.207'}
+
 60%|██████    | 309/512 [34:41<22:27,  6.64s/it]
 61%|██████    | 310/512 [34:48<22:22,  6.65s/it]
                                                 
{'loss': '6.094e-05', 'grad_norm': '0.006136', 'learning_rate': '4.338e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '37.19', 'tokens/total': 9378560, 'tokens/trainable': 141679, 'epoch': '1.211'}
+
 61%|██████    | 310/512 [34:48<22:22,  6.65s/it]
 61%|██████    | 311/512 [34:54<22:15,  6.65s/it]
                                                 
{'loss': '2.005e-05', 'grad_norm': '0.001599', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.74', 'tokens/train_per_sec_per_gpu': '32.25', 'tokens/total': 9408848, 'tokens/trainable': 142122, 'epoch': '1.215'}
+
 61%|██████    | 311/512 [34:54<22:15,  6.65s/it]
 61%|██████    | 312/512 [35:01<22:10,  6.65s/it]
                                                 
{'loss': '1.628e-05', 'grad_norm': '0.0007549', 'learning_rate': '4.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 9439344, 'tokens/trainable': 142561, 'epoch': '1.219'}
+
 61%|██████    | 312/512 [35:01<22:10,  6.65s/it]
 61%|██████    | 313/512 [35:08<22:07,  6.67s/it]
                                                 
{'loss': '1.512e-05', 'grad_norm': '0.0008527', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 9469792, 'tokens/trainable': 143046, 'epoch': '1.223'}
+
 61%|██████    | 313/512 [35:08<22:07,  6.67s/it]
 61%|██████▏   | 314/512 [35:14<22:01,  6.67s/it]
                                                 
{'loss': '0.0001308', 'grad_norm': '0.02262', 'learning_rate': '4.226e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 9500128, 'tokens/trainable': 143493, 'epoch': '1.227'}
+
 61%|██████▏   | 314/512 [35:14<22:01,  6.67s/it]
 62%|██████▏   | 315/512 [35:21<21:54,  6.67s/it]
                                                 
{'loss': '1.917e-05', 'grad_norm': '0.001019', 'learning_rate': '4.198e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.12', 'tokens/total': 9530480, 'tokens/trainable': 143956, 'epoch': '1.23'}
+
 62%|██████▏   | 315/512 [35:21<21:54,  6.67s/it]
 62%|██████▏   | 316/512 [35:28<21:48,  6.67s/it]
                                                 
{'loss': '0.0001811', 'grad_norm': '0.02337', 'learning_rate': '4.17e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 9560912, 'tokens/trainable': 144411, 'epoch': '1.234'}
+
 62%|██████▏   | 316/512 [35:28<21:48,  6.67s/it]
 62%|██████▏   | 317/512 [35:34<21:43,  6.68s/it]
                                                 
{'loss': '0.0001281', 'grad_norm': '0.0114', 'learning_rate': '4.143e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '38.66', 'tokens/total': 9591344, 'tokens/trainable': 144892, 'epoch': '1.238'}
+
 62%|██████▏   | 317/512 [35:34<21:43,  6.68s/it]
 62%|██████▏   | 318/512 [35:41<21:34,  6.67s/it]
                                                 
{'loss': '0.006938', 'grad_norm': '0.5837', 'learning_rate': '4.115e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 9621392, 'tokens/trainable': 145339, 'epoch': '1.242'}
+
 62%|██████▏   | 318/512 [35:41<21:34,  6.67s/it]
 62%|██████▏   | 319/512 [35:48<21:31,  6.69s/it]
                                                 
{'loss': '1.689e-05', 'grad_norm': '0.000792', 'learning_rate': '4.087e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '36.38', 'tokens/total': 9651728, 'tokens/trainable': 145811, 'epoch': '1.246'}
+
 62%|██████▏   | 319/512 [35:48<21:31,  6.69s/it]
 62%|██████▎   | 320/512 [35:55<21:28,  6.71s/it]
                                                 
{'loss': '0.0002538', 'grad_norm': '0.02599', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '38.62', 'tokens/total': 9682224, 'tokens/trainable': 146314, 'epoch': '1.25'}
+
 62%|██████▎   | 320/512 [35:55<21:28,  6.71s/it][2026-08-18 14:40:58,131] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
+
 63%|██████▎   | 321/512 [36:03<23:10,  7.28s/it]
                                                 
{'loss': '1.948e-05', 'grad_norm': '0.001073', 'learning_rate': '4.032e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '37.95', 'tokens/train_per_sec_per_gpu': '31.72', 'tokens/total': 9712656, 'tokens/trainable': 146781, 'epoch': '1.254'}
+
 63%|██████▎   | 321/512 [36:03<23:10,  7.28s/it]
 63%|██████▎   | 322/512 [36:10<22:26,  7.09s/it]
                                                 
{'loss': '0.0008847', 'grad_norm': '0.08435', 'learning_rate': '4.005e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '34.89', 'tokens/train_per_sec_per_gpu': '35.55', 'tokens/total': 9743088, 'tokens/trainable': 147225, 'epoch': '1.258'}
+
 63%|██████▎   | 322/512 [36:10<22:26,  7.09s/it]
 63%|██████▎   | 323/512 [36:17<22:01,  6.99s/it]
                                                 
{'loss': '0.0004002', 'grad_norm': '0.09092', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.17', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 9773808, 'tokens/trainable': 147673, 'epoch': '1.262'}
+
 63%|██████▎   | 323/512 [36:17<22:01,  6.99s/it]
 63%|██████▎   | 324/512 [36:23<21:31,  6.87s/it]
                                                 
{'loss': '0.004064', 'grad_norm': '0.386', 'learning_rate': '3.95e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.23', 'tokens/total': 9804016, 'tokens/trainable': 148103, 'epoch': '1.266'}
+
 63%|██████▎   | 324/512 [36:23<21:31,  6.87s/it]
 63%|██████▎   | 325/512 [36:30<21:12,  6.80s/it]
                                                 
{'loss': '9.267e-06', 'grad_norm': '0.0004181', 'learning_rate': '3.923e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '31.74', 'tokens/total': 9834592, 'tokens/trainable': 148535, 'epoch': '1.27'}
+
 63%|██████▎   | 325/512 [36:30<21:12,  6.80s/it]
 64%|██████▎   | 326/512 [36:36<20:59,  6.77s/it]
                                                 
{'loss': '0.0002955', 'grad_norm': '0.03044', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 9864848, 'tokens/trainable': 148999, 'epoch': '1.273'}
+
 64%|██████▎   | 326/512 [36:36<20:59,  6.77s/it]
 64%|██████▍   | 327/512 [36:43<20:47,  6.74s/it]
                                                 
{'loss': '9.424e-06', 'grad_norm': '0.0003461', 'learning_rate': '3.869e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 9895168, 'tokens/trainable': 149431, 'epoch': '1.277'}
+
 64%|██████▍   | 327/512 [36:43<20:47,  6.74s/it]
 64%|██████▍   | 328/512 [36:50<20:35,  6.71s/it]
                                                 
{'loss': '0.004382', 'grad_norm': '0.3263', 'learning_rate': '3.842e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 9925696, 'tokens/trainable': 149886, 'epoch': '1.281'}
+
 64%|██████▍   | 328/512 [36:50<20:35,  6.71s/it]
 64%|██████▍   | 329/512 [36:56<20:21,  6.67s/it]
                                                 
{'loss': '0.0003639', 'grad_norm': '0.06806', 'learning_rate': '3.815e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '31.01', 'tokens/total': 9955648, 'tokens/trainable': 150288, 'epoch': '1.285'}
+
 64%|██████▍   | 329/512 [36:56<20:21,  6.67s/it]
 64%|██████▍   | 330/512 [37:03<20:15,  6.68s/it]
                                                 
{'loss': '3.247e-05', 'grad_norm': '0.004817', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '33.94', 'tokens/total': 9985856, 'tokens/trainable': 150733, 'epoch': '1.289'}
+
 64%|██████▍   | 330/512 [37:03<20:15,  6.68s/it]
 65%|██████▍   | 331/512 [37:10<20:09,  6.68s/it]
                                                 
{'loss': '0.0003107', 'grad_norm': '0.06924', 'learning_rate': '3.761e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 10016208, 'tokens/trainable': 151207, 'epoch': '1.293'}
+
 65%|██████▍   | 331/512 [37:10<20:09,  6.68s/it]
 65%|██████▍   | 332/512 [37:16<20:04,  6.69s/it]
                                                 
{'loss': '0.001514', 'grad_norm': '0.1473', 'learning_rate': '3.734e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '37.61', 'tokens/total': 10046512, 'tokens/trainable': 151694, 'epoch': '1.297'}
+
 65%|██████▍   | 332/512 [37:16<20:04,  6.69s/it]
 65%|██████▌   | 333/512 [37:23<20:00,  6.71s/it]
                                                 
{'loss': '7.81e-05', 'grad_norm': '0.007326', 'learning_rate': '3.708e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '38.37', 'tokens/total': 10076944, 'tokens/trainable': 152188, 'epoch': '1.301'}
+
 65%|██████▌   | 333/512 [37:23<20:00,  6.71s/it]
 65%|██████▌   | 334/512 [37:30<19:50,  6.69s/it]
                                                 
{'loss': '1.274e-05', 'grad_norm': '0.000476', 'learning_rate': '3.681e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.69', 'tokens/total': 10107296, 'tokens/trainable': 152596, 'epoch': '1.305'}
+
 65%|██████▌   | 334/512 [37:30<19:50,  6.69s/it]
 65%|██████▌   | 335/512 [37:37<19:43,  6.69s/it]
                                                 
{'loss': '0.0004747', 'grad_norm': '0.07003', 'learning_rate': '3.655e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '36.57', 'tokens/total': 10137568, 'tokens/trainable': 153052, 'epoch': '1.309'}
+
 65%|██████▌   | 335/512 [37:37<19:43,  6.69s/it]
 66%|██████▌   | 336/512 [37:43<19:38,  6.70s/it]
                                                 
{'loss': '0.0004141', 'grad_norm': '0.06498', 'learning_rate': '3.628e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '37.17', 'tokens/train_per_sec_per_gpu': '30.19', 'tokens/total': 10167952, 'tokens/trainable': 153491, 'epoch': '1.312'}
+
 66%|██████▌   | 336/512 [37:43<19:38,  6.70s/it]
 66%|██████▌   | 337/512 [37:50<19:34,  6.71s/it]
                                                 
{'loss': '7.06e-05', 'grad_norm': '0.009992', 'learning_rate': '3.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '40.43', 'tokens/total': 10198512, 'tokens/trainable': 153983, 'epoch': '1.316'}
+
 66%|██████▌   | 337/512 [37:50<19:34,  6.71s/it]
 66%|██████▌   | 338/512 [37:57<19:27,  6.71s/it]
                                                 
{'loss': '0.0002046', 'grad_norm': '0.03267', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.72', 'tokens/total': 10228752, 'tokens/trainable': 154464, 'epoch': '1.32'}
+
 66%|██████▌   | 338/512 [37:57<19:27,  6.71s/it]
 66%|██████▌   | 339/512 [38:03<19:18,  6.69s/it]
                                                 
{'loss': '8.338e-06', 'grad_norm': '0.0003953', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.87', 'tokens/total': 10259104, 'tokens/trainable': 154924, 'epoch': '1.324'}
+
 66%|██████▌   | 339/512 [38:03<19:18,  6.69s/it]
 66%|██████▋   | 340/512 [38:10<19:09,  6.68s/it]
                                                 
{'loss': '1.682e-05', 'grad_norm': '0.001057', 'learning_rate': '3.523e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.81', 'tokens/total': 10289520, 'tokens/trainable': 155373, 'epoch': '1.328'}
+
 66%|██████▋   | 340/512 [38:10<19:09,  6.68s/it]
 67%|██████▋   | 341/512 [38:17<19:04,  6.69s/it]
                                                 
{'loss': '1.703e-05', 'grad_norm': '0.0008337', 'learning_rate': '3.497e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '36.47', 'tokens/total': 10320016, 'tokens/trainable': 155873, 'epoch': '1.332'}
+
 67%|██████▋   | 341/512 [38:17<19:04,  6.69s/it]
 67%|██████▋   | 342/512 [38:23<18:55,  6.68s/it]
                                                 
{'loss': '1.627e-05', 'grad_norm': '0.0007872', 'learning_rate': '3.471e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '30.42', 'tokens/total': 10350368, 'tokens/trainable': 156278, 'epoch': '1.336'}
+
 67%|██████▋   | 342/512 [38:23<18:55,  6.68s/it]
 67%|██████▋   | 343/512 [38:30<18:49,  6.68s/it]
                                                 
{'loss': '0.0009315', 'grad_norm': '0.2122', 'learning_rate': '3.445e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.4', 'tokens/total': 10380624, 'tokens/trainable': 156733, 'epoch': '1.34'}
+
 67%|██████▋   | 343/512 [38:30<18:49,  6.68s/it]
 67%|██████▋   | 344/512 [38:37<18:45,  6.70s/it]
                                                 
{'loss': '7.496e-05', 'grad_norm': '0.01747', 'learning_rate': '3.42e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 10411024, 'tokens/trainable': 157203, 'epoch': '1.344'}
+
 67%|██████▋   | 344/512 [38:37<18:45,  6.70s/it]
 67%|██████▋   | 345/512 [38:44<18:39,  6.70s/it]
                                                 
{'loss': '0.004417', 'grad_norm': '0.2606', 'learning_rate': '3.394e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '37.8', 'tokens/total': 10441248, 'tokens/trainable': 157707, 'epoch': '1.348'}
+
 67%|██████▋   | 345/512 [38:44<18:39,  6.70s/it]
 68%|██████▊   | 346/512 [38:50<18:36,  6.73s/it]
                                                 
{'loss': '7.685e-06', 'grad_norm': '0.0003296', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 10471712, 'tokens/trainable': 158180, 'epoch': '1.352'}
+
 68%|██████▊   | 346/512 [38:50<18:36,  6.73s/it]
 68%|██████▊   | 347/512 [38:57<18:29,  6.72s/it]
                                                 
{'loss': '2.94e-05', 'grad_norm': '0.00212', 'learning_rate': '3.343e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.13', 'tokens/train_per_sec_per_gpu': '34.41', 'tokens/total': 10502288, 'tokens/trainable': 158646, 'epoch': '1.355'}
+
 68%|██████▊   | 347/512 [38:57<18:29,  6.72s/it]
 68%|██████▊   | 348/512 [39:04<18:16,  6.68s/it]
                                                 
{'loss': '9.501e-06', 'grad_norm': '0.0004069', 'learning_rate': '3.317e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '35.66', 'tokens/total': 10532304, 'tokens/trainable': 159115, 'epoch': '1.359'}
+
 68%|██████▊   | 348/512 [39:04<18:16,  6.68s/it]
 68%|██████▊   | 349/512 [39:10<18:07,  6.67s/it]
                                                 
{'loss': '9.019e-05', 'grad_norm': '0.006379', 'learning_rate': '3.292e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '31.62', 'tokens/total': 10562608, 'tokens/trainable': 159538, 'epoch': '1.363'}
+
 68%|██████▊   | 349/512 [39:10<18:07,  6.67s/it]
 68%|██████▊   | 350/512 [39:17<18:01,  6.68s/it]
                                                 
{'loss': '5.146e-05', 'grad_norm': '0.005556', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 10592992, 'tokens/trainable': 160016, 'epoch': '1.367'}
+
 68%|██████▊   | 350/512 [39:17<18:01,  6.68s/it]
 69%|██████▊   | 351/512 [39:24<17:53,  6.67s/it]
                                                 
{'loss': '9.038e-05', 'grad_norm': '0.009833', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '34.57', 'tokens/total': 10623328, 'tokens/trainable': 160475, 'epoch': '1.371'}
+
 69%|██████▊   | 351/512 [39:24<17:53,  6.67s/it]
 69%|██████▉   | 352/512 [39:30<17:47,  6.67s/it]
                                                 
{'loss': '7.51e-06', 'grad_norm': '0.0002588', 'learning_rate': '3.217e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 10653632, 'tokens/trainable': 160952, 'epoch': '1.375'}
+
 69%|██████▉   | 352/512 [39:30<17:47,  6.67s/it][2026-08-18 14:44:33,881] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
+
 69%|██████▉   | 353/512 [39:39<19:10,  7.24s/it]
                                                 
{'loss': '0.0002666', 'grad_norm': '0.03195', 'learning_rate': '3.192e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '38.4', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 10684176, 'tokens/trainable': 161420, 'epoch': '1.379'}
+
 69%|██████▉   | 353/512 [39:39<19:10,  7.24s/it]
 69%|██████▉   | 354/512 [39:45<18:34,  7.05s/it]
                                                 
{'loss': '0.0007417', 'grad_norm': '0.07013', 'learning_rate': '3.167e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 10714352, 'tokens/trainable': 161868, 'epoch': '1.383'}
+
 69%|██████▉   | 354/512 [39:45<18:34,  7.05s/it]
 69%|██████▉   | 355/512 [39:52<18:09,  6.94s/it]
                                                 
{'loss': '1.038e-05', 'grad_norm': '0.0005823', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '35.44', 'tokens/total': 10744464, 'tokens/trainable': 162318, 'epoch': '1.387'}
+
 69%|██████▉   | 355/512 [39:52<18:09,  6.94s/it]
 70%|██████▉   | 356/512 [39:59<17:52,  6.87s/it]
                                                 
{'loss': '7.898e-05', 'grad_norm': '0.01244', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.54', 'tokens/total': 10774976, 'tokens/trainable': 162789, 'epoch': '1.391'}
+
 70%|██████▉   | 356/512 [39:59<17:52,  6.87s/it]
 70%|██████▉   | 357/512 [40:06<17:37,  6.82s/it]
                                                 
{'loss': '2.485e-05', 'grad_norm': '0.00456', 'learning_rate': '3.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.43', 'tokens/total': 10805600, 'tokens/trainable': 163254, 'epoch': '1.395'}
+
 70%|██████▉   | 357/512 [40:06<17:37,  6.82s/it]
 70%|██████▉   | 358/512 [40:12<17:22,  6.77s/it]
                                                 
{'loss': '0.002299', 'grad_norm': '0.3044', 'learning_rate': '3.068e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.62', 'tokens/total': 10835920, 'tokens/trainable': 163715, 'epoch': '1.398'}
+
 70%|██████▉   | 358/512 [40:12<17:22,  6.77s/it]
 70%|███████   | 359/512 [40:19<17:12,  6.75s/it]
                                                 
{'loss': '1.963e-05', 'grad_norm': '0.002458', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 10866048, 'tokens/trainable': 164197, 'epoch': '1.402'}
+
 70%|███████   | 359/512 [40:19<17:12,  6.75s/it]
 70%|███████   | 360/512 [40:26<17:02,  6.73s/it]
                                                 
{'loss': '2.745e-05', 'grad_norm': '0.002477', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '31.74', 'tokens/total': 10896288, 'tokens/trainable': 164661, 'epoch': '1.406'}
+
 70%|███████   | 360/512 [40:26<17:02,  6.73s/it]
 71%|███████   | 361/512 [40:32<16:52,  6.70s/it]
                                                 
{'loss': '6.423e-06', 'grad_norm': '0.0002672', 'learning_rate': '2.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 10926816, 'tokens/trainable': 165111, 'epoch': '1.41'}
+
 71%|███████   | 361/512 [40:32<16:52,  6.70s/it]
 71%|███████   | 362/512 [40:38<16:19,  6.53s/it]
                                                 
{'loss': '1.355e-05', 'grad_norm': '0.0008353', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '30.52', 'tokens/total': 10954928, 'tokens/trainable': 165552, 'epoch': '1.414'}
+
 71%|███████   | 362/512 [40:38<16:19,  6.53s/it]
 71%|███████   | 363/512 [40:45<16:19,  6.57s/it]
                                                 
{'loss': '6.428e-05', 'grad_norm': '0.006071', 'learning_rate': '2.947e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 10985328, 'tokens/trainable': 166026, 'epoch': '1.418'}
+
 71%|███████   | 363/512 [40:45<16:19,  6.57s/it]
 71%|███████   | 364/512 [40:52<16:17,  6.60s/it]
                                                 
{'loss': '0.004244', 'grad_norm': '0.2542', 'learning_rate': '2.924e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '38.11', 'tokens/total': 11015664, 'tokens/trainable': 166486, 'epoch': '1.422'}
+
 71%|███████   | 364/512 [40:52<16:17,  6.60s/it]
 71%|███████▏  | 365/512 [40:58<16:12,  6.62s/it]
                                                 
{'loss': '8.404e-06', 'grad_norm': '0.0005575', 'learning_rate': '2.9e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 11046256, 'tokens/trainable': 166959, 'epoch': '1.426'}
+
 71%|███████▏  | 365/512 [40:58<16:12,  6.62s/it]
 71%|███████▏  | 366/512 [41:05<16:04,  6.61s/it]
                                                 
{'loss': '1.474e-05', 'grad_norm': '0.001434', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 11076288, 'tokens/trainable': 167394, 'epoch': '1.43'}
+
 71%|███████▏  | 366/512 [41:05<16:04,  6.61s/it]
 72%|███████▏  | 367/512 [41:12<15:58,  6.61s/it]
                                                 
{'loss': '1.328e-05', 'grad_norm': '0.001015', 'learning_rate': '2.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.01', 'tokens/train_per_sec_per_gpu': '28.58', 'tokens/total': 11106416, 'tokens/trainable': 167832, 'epoch': '1.434'}
+
 72%|███████▏  | 367/512 [41:12<15:58,  6.61s/it]
 72%|███████▏  | 368/512 [41:18<15:53,  6.62s/it]
                                                 
{'loss': '0.0006477', 'grad_norm': '0.1194', 'learning_rate': '2.829e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.02', 'tokens/total': 11136848, 'tokens/trainable': 168242, 'epoch': '1.438'}
+
 72%|███████▏  | 368/512 [41:18<15:53,  6.62s/it]
 72%|███████▏  | 369/512 [41:25<15:48,  6.63s/it]
                                                 
{'loss': '7.449e-05', 'grad_norm': '0.006878', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.74', 'tokens/total': 11167376, 'tokens/trainable': 168715, 'epoch': '1.441'}
+
 72%|███████▏  | 369/512 [41:25<15:48,  6.63s/it]
 72%|███████▏  | 370/512 [41:32<15:44,  6.65s/it]
                                                 
{'loss': '1.144e-05', 'grad_norm': '0.0007151', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.86', 'tokens/total': 11197776, 'tokens/trainable': 169147, 'epoch': '1.445'}
+
 72%|███████▏  | 370/512 [41:32<15:44,  6.65s/it]
 72%|███████▏  | 371/512 [41:38<15:37,  6.65s/it]
                                                 
{'loss': '5.924e-05', 'grad_norm': '0.009094', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.35', 'tokens/total': 11227840, 'tokens/trainable': 169628, 'epoch': '1.449'}
+
 72%|███████▏  | 371/512 [41:38<15:37,  6.65s/it]
 73%|███████▎  | 372/512 [41:45<15:32,  6.66s/it]
                                                 
{'loss': '7.548e-06', 'grad_norm': '0.0002473', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 11258176, 'tokens/trainable': 170085, 'epoch': '1.453'}
+
 73%|███████▎  | 372/512 [41:45<15:32,  6.66s/it]
 73%|███████▎  | 373/512 [41:51<15:24,  6.65s/it]
                                                 
{'loss': '5.401e-06', 'grad_norm': '0.0003029', 'learning_rate': '2.714e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 11288336, 'tokens/trainable': 170584, 'epoch': '1.457'}
+
 73%|███████▎  | 373/512 [41:51<15:24,  6.65s/it]
 73%|███████▎  | 374/512 [41:58<15:18,  6.66s/it]
                                                 
{'loss': '2.381e-05', 'grad_norm': '0.001942', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.68', 'tokens/total': 11318848, 'tokens/trainable': 171058, 'epoch': '1.461'}
+
 73%|███████▎  | 374/512 [41:58<15:18,  6.66s/it]
 73%|███████▎  | 375/512 [42:05<15:13,  6.67s/it]
                                                 
{'loss': '1.523e-05', 'grad_norm': '0.001931', 'learning_rate': '2.668e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.94', 'tokens/total': 11349168, 'tokens/trainable': 171518, 'epoch': '1.465'}
+
 73%|███████▎  | 375/512 [42:05<15:13,  6.67s/it]
 73%|███████▎  | 376/512 [42:11<15:04,  6.65s/it]
                                                 
{'loss': '0.00635', 'grad_norm': '0.6044', 'learning_rate': '2.646e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '40.62', 'tokens/total': 11379328, 'tokens/trainable': 172007, 'epoch': '1.469'}
+
 73%|███████▎  | 376/512 [42:11<15:04,  6.65s/it]
 74%|███████▎  | 377/512 [42:18<14:57,  6.65s/it]
                                                 
{'loss': '0.001872', 'grad_norm': '0.2545', 'learning_rate': '2.624e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 11409568, 'tokens/trainable': 172468, 'epoch': '1.473'}
+
 74%|███████▎  | 377/512 [42:18<14:57,  6.65s/it]
 74%|███████▍  | 378/512 [42:25<14:51,  6.66s/it]
                                                 
{'loss': '3.05e-05', 'grad_norm': '0.001921', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 11439968, 'tokens/trainable': 172923, 'epoch': '1.477'}
+
 74%|███████▍  | 378/512 [42:25<14:51,  6.66s/it]
 74%|███████▍  | 379/512 [42:31<14:44,  6.65s/it]
                                                 
{'loss': '0.02674', 'grad_norm': '0.3349', 'learning_rate': '2.579e-05', 'ppl': '1.027', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 11470496, 'tokens/trainable': 173370, 'epoch': '1.48'}
+
 74%|███████▍  | 379/512 [42:31<14:44,  6.65s/it]
 74%|███████▍  | 380/512 [42:38<14:37,  6.65s/it]
                                                 
{'loss': '5.195e-05', 'grad_norm': '0.006178', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 11500720, 'tokens/trainable': 173837, 'epoch': '1.484'}
+
 74%|███████▍  | 380/512 [42:38<14:37,  6.65s/it]
 74%|███████▍  | 381/512 [42:45<14:32,  6.66s/it]
                                                 
{'loss': '0.0006311', 'grad_norm': '0.06973', 'learning_rate': '2.535e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.54', 'tokens/total': 11531280, 'tokens/trainable': 174294, 'epoch': '1.488'}
+
 74%|███████▍  | 381/512 [42:45<14:32,  6.66s/it]
 75%|███████▍  | 382/512 [42:51<14:23,  6.64s/it]
                                                 
{'loss': '4.85e-05', 'grad_norm': '0.003902', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.84', 'tokens/total': 11561264, 'tokens/trainable': 174701, 'epoch': '1.492'}
+
 75%|███████▍  | 382/512 [42:51<14:23,  6.64s/it]
 75%|███████▍  | 383/512 [42:58<14:18,  6.65s/it]
                                                 
{'loss': '2.349e-05', 'grad_norm': '0.0007975', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 11591616, 'tokens/trainable': 175131, 'epoch': '1.496'}
+
 75%|███████▍  | 383/512 [42:58<14:18,  6.65s/it]
 75%|███████▌  | 384/512 [43:05<14:12,  6.66s/it]
                                                 
{'loss': '0.000115', 'grad_norm': '0.01254', 'learning_rate': '2.47e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 11622000, 'tokens/trainable': 175586, 'epoch': '1.5'}
+
 75%|███████▌  | 384/512 [43:05<14:12,  6.66s/it][2026-08-18 14:48:08,327] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
+
 75%|███████▌  | 385/512 [43:13<15:16,  7.21s/it]
                                                 
{'loss': '6.503e-05', 'grad_norm': '0.004752', 'learning_rate': '2.449e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 11652352, 'tokens/trainable': 176026, 'epoch': '1.504'}
+
 75%|███████▌  | 385/512 [43:13<15:16,  7.21s/it]
 75%|███████▌  | 386/512 [43:20<14:49,  7.06s/it]
                                                 
{'loss': '8.075e-05', 'grad_norm': '0.01419', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 11682736, 'tokens/trainable': 176512, 'epoch': '1.508'}
+
 75%|███████▌  | 386/512 [43:20<14:49,  7.06s/it]
 76%|███████▌  | 387/512 [43:27<14:28,  6.95s/it]
                                                 
{'loss': '8.846e-05', 'grad_norm': '0.007288', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '34.98', 'tokens/train_per_sec_per_gpu': '30.45', 'tokens/total': 11713136, 'tokens/trainable': 176943, 'epoch': '1.512'}
+
 76%|███████▌  | 387/512 [43:27<14:28,  6.95s/it]
 76%|███████▌  | 388/512 [43:33<14:10,  6.86s/it]
                                                 
{'loss': '0.0007253', 'grad_norm': '0.1016', 'learning_rate': '2.385e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.23', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 11743808, 'tokens/trainable': 177370, 'epoch': '1.516'}
+
 76%|███████▌  | 388/512 [43:33<14:10,  6.86s/it]
 76%|███████▌  | 389/512 [43:40<13:57,  6.81s/it]
                                                 
{'loss': '8.653e-05', 'grad_norm': '0.005729', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.58', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 11774144, 'tokens/trainable': 177851, 'epoch': '1.52'}
+
 76%|███████▌  | 389/512 [43:40<13:57,  6.81s/it]
 76%|███████▌  | 390/512 [43:47<13:45,  6.77s/it]
                                                 
{'loss': '0.0007951', 'grad_norm': '0.07154', 'learning_rate': '2.344e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.75', 'tokens/total': 11804272, 'tokens/trainable': 178310, 'epoch': '1.523'}
+
 76%|███████▌  | 390/512 [43:47<13:45,  6.77s/it]
 76%|███████▋  | 391/512 [43:53<13:18,  6.60s/it]
                                                 
{'loss': '0.0001062', 'grad_norm': '0.009775', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 11832432, 'tokens/trainable': 178736, 'epoch': '1.527'}
+
 76%|███████▋  | 391/512 [43:53<13:18,  6.60s/it]
 77%|███████▋  | 392/512 [44:00<13:18,  6.65s/it]
                                                 
{'loss': '2.955e-05', 'grad_norm': '0.001555', 'learning_rate': '2.303e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 11862960, 'tokens/trainable': 179233, 'epoch': '1.531'}
+
 77%|███████▋  | 392/512 [44:00<13:18,  6.65s/it]
 77%|███████▋  | 393/512 [44:06<13:13,  6.67s/it]
                                                 
{'loss': '0.0002111', 'grad_norm': '0.02611', 'learning_rate': '2.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.57', 'tokens/total': 11893232, 'tokens/trainable': 179730, 'epoch': '1.535'}
+
 77%|███████▋  | 393/512 [44:06<13:13,  6.67s/it]
 77%|███████▋  | 394/512 [44:13<13:08,  6.68s/it]
                                                 
{'loss': '0.0002601', 'grad_norm': '0.02035', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.87', 'tokens/total': 11923664, 'tokens/trainable': 180227, 'epoch': '1.539'}
+
 77%|███████▋  | 394/512 [44:13<13:08,  6.68s/it]
 77%|███████▋  | 395/512 [44:20<13:02,  6.69s/it]
                                                 
{'loss': '4.929e-05', 'grad_norm': '0.00542', 'learning_rate': '2.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 11954032, 'tokens/trainable': 180687, 'epoch': '1.543'}
+
 77%|███████▋  | 395/512 [44:20<13:02,  6.69s/it]
 77%|███████▋  | 396/512 [44:26<12:56,  6.69s/it]
                                                 
{'loss': '6.144e-05', 'grad_norm': '0.0031', 'learning_rate': '2.222e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 11984512, 'tokens/trainable': 181141, 'epoch': '1.547'}
+
 77%|███████▋  | 396/512 [44:26<12:56,  6.69s/it]
 78%|███████▊  | 397/512 [44:33<12:51,  6.71s/it]
                                                 
{'loss': '9.172e-05', 'grad_norm': '0.006749', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.02', 'tokens/total': 12014880, 'tokens/trainable': 181648, 'epoch': '1.551'}
+
 78%|███████▊  | 397/512 [44:33<12:51,  6.71s/it]
 78%|███████▊  | 398/512 [44:40<12:42,  6.69s/it]
                                                 
{'loss': '0.001001', 'grad_norm': '0.08928', 'learning_rate': '2.183e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 12044928, 'tokens/trainable': 182109, 'epoch': '1.555'}
+
 78%|███████▊  | 398/512 [44:40<12:42,  6.69s/it]
 78%|███████▊  | 399/512 [44:47<12:35,  6.69s/it]
                                                 
{'loss': '5.344e-05', 'grad_norm': '0.004652', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.58', 'tokens/total': 12075344, 'tokens/trainable': 182559, 'epoch': '1.559'}
+
 78%|███████▊  | 399/512 [44:47<12:35,  6.69s/it]
 78%|███████▊  | 400/512 [44:53<12:27,  6.67s/it]
                                                 
{'loss': '4.57e-05', 'grad_norm': '0.001751', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.17', 'tokens/total': 12105616, 'tokens/trainable': 182985, 'epoch': '1.562'}
+
 78%|███████▊  | 400/512 [44:53<12:27,  6.67s/it]
 78%|███████▊  | 401/512 [45:00<12:20,  6.67s/it]
                                                 
{'loss': '0.0001243', 'grad_norm': '0.01575', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 12136080, 'tokens/trainable': 183440, 'epoch': '1.566'}
+
 78%|███████▊  | 401/512 [45:00<12:20,  6.67s/it]
 79%|███████▊  | 402/512 [45:07<12:14,  6.68s/it]
                                                 
{'loss': '0.0001223', 'grad_norm': '0.007152', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '34.51', 'tokens/total': 12166240, 'tokens/trainable': 183917, 'epoch': '1.57'}
+
 79%|███████▊  | 402/512 [45:07<12:14,  6.68s/it]
 79%|███████▊  | 403/512 [45:13<12:07,  6.67s/it]
                                                 
{'loss': '3.622e-05', 'grad_norm': '0.001534', 'learning_rate': '2.086e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '30.36', 'tokens/total': 12196544, 'tokens/trainable': 184348, 'epoch': '1.574'}
+
 79%|███████▊  | 403/512 [45:13<12:07,  6.67s/it]
 79%|███████▉  | 404/512 [45:20<12:00,  6.67s/it]
                                                 
{'loss': '6.368e-05', 'grad_norm': '0.003521', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 12226784, 'tokens/trainable': 184786, 'epoch': '1.578'}
+
 79%|███████▉  | 404/512 [45:20<12:00,  6.67s/it]
 79%|███████▉  | 405/512 [45:27<11:54,  6.68s/it]
                                                 
{'loss': '2.033e-05', 'grad_norm': '0.0007212', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 12257264, 'tokens/trainable': 185249, 'epoch': '1.582'}
+
 79%|███████▉  | 405/512 [45:27<11:54,  6.68s/it]
 79%|███████▉  | 406/512 [45:33<11:49,  6.69s/it]
                                                 
{'loss': '3.38e-05', 'grad_norm': '0.00148', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.43', 'tokens/total': 12287584, 'tokens/trainable': 185680, 'epoch': '1.586'}
+
 79%|███████▉  | 406/512 [45:33<11:49,  6.69s/it]
 79%|███████▉  | 407/512 [45:40<11:41,  6.68s/it]
                                                 
{'loss': '5.18e-05', 'grad_norm': '0.004398', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 12317984, 'tokens/trainable': 186100, 'epoch': '1.59'}
+
 79%|███████▉  | 407/512 [45:40<11:41,  6.68s/it]
 80%|███████▉  | 408/512 [45:47<11:33,  6.67s/it]
                                                 
{'loss': '3.441e-05', 'grad_norm': '0.002647', 'learning_rate': '1.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.84', 'tokens/total': 12348144, 'tokens/trainable': 186565, 'epoch': '1.594'}
+
 80%|███████▉  | 408/512 [45:47<11:33,  6.67s/it]
 80%|███████▉  | 409/512 [45:53<11:27,  6.67s/it]
                                                 
{'loss': '0.000216', 'grad_norm': '0.02899', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.78', 'tokens/total': 12378544, 'tokens/trainable': 187017, 'epoch': '1.598'}
+
 80%|███████▉  | 409/512 [45:53<11:27,  6.67s/it]
 80%|████████  | 410/512 [46:00<11:21,  6.68s/it]
                                                 
{'loss': '9.646e-05', 'grad_norm': '0.007093', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 12408784, 'tokens/trainable': 187468, 'epoch': '1.602'}
+
 80%|████████  | 410/512 [46:00<11:21,  6.68s/it]
 80%|████████  | 411/512 [46:07<11:14,  6.67s/it]
                                                 
{'loss': '0.0003458', 'grad_norm': '0.02711', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.57', 'tokens/total': 12438832, 'tokens/trainable': 187961, 'epoch': '1.605'}
+
 80%|████████  | 411/512 [46:07<11:14,  6.67s/it]
 80%|████████  | 412/512 [46:13<11:07,  6.68s/it]
                                                 
{'loss': '0.003292', 'grad_norm': '0.2949', 'learning_rate': '1.922e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.93', 'tokens/total': 12469296, 'tokens/trainable': 188447, 'epoch': '1.609'}
+
 80%|████████  | 412/512 [46:13<11:07,  6.68s/it]
 81%|████████  | 413/512 [46:20<11:00,  6.67s/it]
                                                 
{'loss': '0.003581', 'grad_norm': '0.2288', 'learning_rate': '1.904e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '35.97', 'tokens/total': 12499392, 'tokens/trainable': 188903, 'epoch': '1.613'}
+
 81%|████████  | 413/512 [46:20<11:00,  6.67s/it]
 81%|████████  | 414/512 [46:27<10:56,  6.70s/it]
                                                 
{'loss': '6.048e-05', 'grad_norm': '0.002635', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 12529952, 'tokens/trainable': 189398, 'epoch': '1.617'}
+
 81%|████████  | 414/512 [46:27<10:56,  6.70s/it]
 81%|████████  | 415/512 [46:33<10:50,  6.70s/it]
                                                 
{'loss': '4.138e-05', 'grad_norm': '0.001722', 'learning_rate': '1.87e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 12560240, 'tokens/trainable': 189850, 'epoch': '1.621'}
+
 81%|████████  | 415/512 [46:33<10:50,  6.70s/it]
 81%|████████▏ | 416/512 [46:40<10:43,  6.70s/it]
                                                 
{'loss': '0.0001111', 'grad_norm': '0.01184', 'learning_rate': '1.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 12590736, 'tokens/trainable': 190306, 'epoch': '1.625'}
+
 81%|████████▏ | 416/512 [46:40<10:43,  6.70s/it][2026-08-18 14:51:43,726] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
+
 81%|████████▏ | 417/512 [46:49<11:28,  7.25s/it]
                                                 
{'loss': '0.000122', 'grad_norm': '0.01777', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '38.05', 'tokens/train_per_sec_per_gpu': '34.21', 'tokens/total': 12621168, 'tokens/trainable': 190746, 'epoch': '1.629'}
+
 81%|████████▏ | 417/512 [46:49<11:28,  7.25s/it]
 82%|████████▏ | 418/512 [46:55<11:04,  7.07s/it]
                                                 
{'loss': '5.223e-05', 'grad_norm': '0.003219', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '31.15', 'tokens/total': 12651456, 'tokens/trainable': 191156, 'epoch': '1.633'}
+
 82%|████████▏ | 418/512 [46:55<11:04,  7.07s/it]
 82%|████████▏ | 419/512 [47:02<10:45,  6.94s/it]
                                                 
{'loss': '0.0001365', 'grad_norm': '0.01374', 'learning_rate': '1.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 12681760, 'tokens/trainable': 191573, 'epoch': '1.637'}
+
 82%|████████▏ | 419/512 [47:02<10:45,  6.94s/it]
 82%|████████▏ | 420/512 [47:09<10:31,  6.86s/it]
                                                 
{'loss': '0.002095', 'grad_norm': '0.2501', 'learning_rate': '1.786e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 12712144, 'tokens/trainable': 192014, 'epoch': '1.641'}
+
 82%|████████▏ | 420/512 [47:09<10:31,  6.86s/it]
 82%|████████▏ | 421/512 [47:15<10:19,  6.80s/it]
                                                 
{'loss': '0.001427', 'grad_norm': '0.1429', 'learning_rate': '1.77e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 12742400, 'tokens/trainable': 192453, 'epoch': '1.645'}
+
 82%|████████▏ | 421/512 [47:15<10:19,  6.80s/it]
 82%|████████▏ | 422/512 [47:22<10:07,  6.75s/it]
                                                 
{'loss': '1.823e-05', 'grad_norm': '0.0009391', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 12772288, 'tokens/trainable': 192886, 'epoch': '1.648'}
+
 82%|████████▏ | 422/512 [47:22<10:07,  6.75s/it]
 83%|████████▎ | 423/512 [47:29<10:01,  6.76s/it]
                                                 
{'loss': '6.211e-05', 'grad_norm': '0.004544', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '29.87', 'tokens/total': 12802880, 'tokens/trainable': 193323, 'epoch': '1.652'}
+
 83%|████████▎ | 423/512 [47:29<10:01,  6.76s/it]
 83%|████████▎ | 424/512 [47:35<09:56,  6.78s/it]
                                                 
{'loss': '0.0005114', 'grad_norm': '0.03986', 'learning_rate': '1.722e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 12833456, 'tokens/trainable': 193753, 'epoch': '1.656'}
+
 83%|████████▎ | 424/512 [47:35<09:56,  6.78s/it]
 83%|████████▎ | 425/512 [47:42<09:46,  6.75s/it]
                                                 
{'loss': '7.249e-05', 'grad_norm': '0.004084', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.41', 'tokens/total': 12863952, 'tokens/trainable': 194240, 'epoch': '1.66'}
+
 83%|████████▎ | 425/512 [47:42<09:46,  6.75s/it]
 83%|████████▎ | 426/512 [47:49<09:38,  6.72s/it]
                                                 
{'loss': '0.001506', 'grad_norm': '0.1312', 'learning_rate': '1.69e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '35.54', 'tokens/total': 12894352, 'tokens/trainable': 194680, 'epoch': '1.664'}
+
 83%|████████▎ | 426/512 [47:49<09:38,  6.72s/it]
 83%|████████▎ | 427/512 [47:56<09:31,  6.72s/it]
                                                 
{'loss': '8.445e-05', 'grad_norm': '0.01044', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '38.77', 'tokens/total': 12924688, 'tokens/trainable': 195186, 'epoch': '1.668'}
+
 83%|████████▎ | 427/512 [47:56<09:31,  6.72s/it]
 84%|████████▎ | 428/512 [48:02<09:25,  6.73s/it]
                                                 
{'loss': '9.946e-05', 'grad_norm': '0.007595', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 12955072, 'tokens/trainable': 195693, 'epoch': '1.672'}
+
 84%|████████▎ | 428/512 [48:02<09:25,  6.73s/it]
 84%|████████▍ | 429/512 [48:09<09:16,  6.71s/it]
                                                 
{'loss': '0.0001356', 'grad_norm': '0.006466', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.05', 'tokens/train_per_sec_per_gpu': '36.95', 'tokens/total': 12985328, 'tokens/trainable': 196130, 'epoch': '1.676'}
+
 84%|████████▍ | 429/512 [48:09<09:16,  6.71s/it]
 84%|████████▍ | 430/512 [48:16<09:08,  6.69s/it]
                                                 
{'loss': '3.711e-05', 'grad_norm': '0.002304', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 13015600, 'tokens/trainable': 196530, 'epoch': '1.68'}
+
 84%|████████▍ | 430/512 [48:16<09:08,  6.69s/it]
 84%|████████▍ | 431/512 [48:22<09:00,  6.68s/it]
                                                 
{'loss': '6.117e-05', 'grad_norm': '0.00419', 'learning_rate': '1.615e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '29.94', 'tokens/total': 13045728, 'tokens/trainable': 196974, 'epoch': '1.684'}
+
 84%|████████▍ | 431/512 [48:22<09:00,  6.68s/it]
 84%|████████▍ | 432/512 [48:29<08:54,  6.68s/it]
                                                 
{'loss': '0.0004236', 'grad_norm': '0.02952', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 13075920, 'tokens/trainable': 197454, 'epoch': '1.688'}
+
 84%|████████▍ | 432/512 [48:29<08:54,  6.68s/it]
 85%|████████▍ | 433/512 [48:36<08:47,  6.68s/it]
                                                 
{'loss': '1.119e-05', 'grad_norm': '0.0003203', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '31.43', 'tokens/total': 13106128, 'tokens/trainable': 197902, 'epoch': '1.691'}
+
 85%|████████▍ | 433/512 [48:36<08:47,  6.68s/it]
 85%|████████▍ | 434/512 [48:42<08:40,  6.67s/it]
                                                 
{'loss': '2.119e-05', 'grad_norm': '0.0006868', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '36.06', 'tokens/total': 13136368, 'tokens/trainable': 198374, 'epoch': '1.695'}
+
 85%|████████▍ | 434/512 [48:42<08:40,  6.67s/it]
 85%|████████▍ | 435/512 [48:49<08:33,  6.67s/it]
                                                 
{'loss': '4.928e-05', 'grad_norm': '0.002541', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.59', 'tokens/total': 13166704, 'tokens/trainable': 198846, 'epoch': '1.699'}
+
 85%|████████▍ | 435/512 [48:49<08:33,  6.67s/it]
 85%|████████▌ | 436/512 [48:56<08:28,  6.69s/it]
                                                 
{'loss': '3.578e-05', 'grad_norm': '0.00139', 'learning_rate': '1.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 13197344, 'tokens/trainable': 199320, 'epoch': '1.703'}
+
 85%|████████▌ | 436/512 [48:56<08:28,  6.69s/it]
 85%|████████▌ | 437/512 [49:02<08:21,  6.69s/it]
                                                 
{'loss': '6.294e-05', 'grad_norm': '0.003646', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.06', 'tokens/train_per_sec_per_gpu': '37.46', 'tokens/total': 13227664, 'tokens/trainable': 199777, 'epoch': '1.707'}
+
 85%|████████▌ | 437/512 [49:02<08:21,  6.69s/it]
 86%|████████▌ | 438/512 [49:09<08:12,  6.66s/it]
                                                 
{'loss': '0.000177', 'grad_norm': '0.03043', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '30.93', 'tokens/total': 13257632, 'tokens/trainable': 200194, 'epoch': '1.711'}
+
 86%|████████▌ | 438/512 [49:09<08:12,  6.66s/it]
 86%|████████▌ | 439/512 [49:16<08:09,  6.70s/it]
                                                 
{'loss': '3.761e-05', 'grad_norm': '0.001823', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 13288240, 'tokens/trainable': 200663, 'epoch': '1.715'}
+
 86%|████████▌ | 439/512 [49:16<08:09,  6.70s/it]
 86%|████████▌ | 440/512 [49:23<08:04,  6.73s/it]
                                                 
{'loss': '4.779e-05', 'grad_norm': '0.002938', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 13318992, 'tokens/trainable': 201121, 'epoch': '1.719'}
+
 86%|████████▌ | 440/512 [49:23<08:04,  6.73s/it]
 86%|████████▌ | 441/512 [49:29<07:56,  6.71s/it]
                                                 
{'loss': '3.15e-05', 'grad_norm': '0.001837', 'learning_rate': '1.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.81', 'tokens/total': 13349392, 'tokens/trainable': 201598, 'epoch': '1.723'}
+
 86%|████████▌ | 441/512 [49:29<07:56,  6.71s/it]
 86%|████████▋ | 442/512 [49:36<07:49,  6.70s/it]
                                                 
{'loss': '2.081e-05', 'grad_norm': '0.001543', 'learning_rate': '1.464e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 13379648, 'tokens/trainable': 202068, 'epoch': '1.727'}
+
 86%|████████▋ | 442/512 [49:36<07:49,  6.70s/it]
 87%|████████▋ | 443/512 [49:43<07:41,  6.69s/it]
                                                 
{'loss': '2.304e-05', 'grad_norm': '0.001077', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.5', 'tokens/total': 13409936, 'tokens/trainable': 202524, 'epoch': '1.73'}
+
 87%|████████▋ | 443/512 [49:43<07:41,  6.69s/it]
 87%|████████▋ | 444/512 [49:49<07:34,  6.69s/it]
                                                 
{'loss': '7.449e-05', 'grad_norm': '0.006372', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 13440288, 'tokens/trainable': 202996, 'epoch': '1.734'}
+
 87%|████████▋ | 444/512 [49:49<07:34,  6.69s/it]
 87%|████████▋ | 445/512 [49:55<07:19,  6.56s/it]
                                                 
{'loss': '4.211e-05', 'grad_norm': '0.007095', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '39.19', 'tokens/total': 13468880, 'tokens/trainable': 203470, 'epoch': '1.738'}
+
 87%|████████▋ | 445/512 [49:55<07:19,  6.56s/it]
 87%|████████▋ | 446/512 [50:02<07:15,  6.60s/it]
                                                 
{'loss': '8.662e-05', 'grad_norm': '0.006931', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 13499184, 'tokens/trainable': 203919, 'epoch': '1.742'}
+
 87%|████████▋ | 446/512 [50:02<07:15,  6.60s/it]
 87%|████████▋ | 447/512 [50:09<07:10,  6.62s/it]
                                                 
{'loss': '4.787e-05', 'grad_norm': '0.004166', 'learning_rate': '1.402e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.24', 'tokens/total': 13529632, 'tokens/trainable': 204376, 'epoch': '1.746'}
+
 87%|████████▋ | 447/512 [50:09<07:10,  6.62s/it]
 88%|████████▊ | 448/512 [50:16<07:04,  6.64s/it]
                                                 
{'loss': '0.0002709', 'grad_norm': '0.04226', 'learning_rate': '1.39e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32.06', 'tokens/total': 13560080, 'tokens/trainable': 204821, 'epoch': '1.75'}
+
 88%|████████▊ | 448/512 [50:16<07:04,  6.64s/it][2026-08-18 14:55:19,147] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
+
 88%|████████▊ | 449/512 [50:24<07:34,  7.22s/it]
                                                 
{'loss': '1.284e-05', 'grad_norm': '0.0004374', 'learning_rate': '1.378e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 13590496, 'tokens/trainable': 205271, 'epoch': '1.754'}
+
 88%|████████▊ | 449/512 [50:24<07:34,  7.22s/it]
 88%|████████▊ | 450/512 [50:31<07:17,  7.05s/it]
                                                 
{'loss': '2.11e-05', 'grad_norm': '0.001298', 'learning_rate': '1.367e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '30.92', 'tokens/total': 13620960, 'tokens/trainable': 205723, 'epoch': '1.758'}
+
 88%|████████▊ | 450/512 [50:31<07:17,  7.05s/it]
 88%|████████▊ | 451/512 [50:37<07:03,  6.95s/it]
                                                 
{'loss': '7.466e-05', 'grad_norm': '0.005675', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.48', 'tokens/total': 13651344, 'tokens/trainable': 206169, 'epoch': '1.762'}
+
 88%|████████▊ | 451/512 [50:37<07:03,  6.95s/it]
 88%|████████▊ | 452/512 [50:44<06:53,  6.89s/it]
                                                 
{'loss': '2.723e-05', 'grad_norm': '0.0014', 'learning_rate': '1.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.04', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 13681792, 'tokens/trainable': 206592, 'epoch': '1.766'}
+
 88%|████████▊ | 452/512 [50:44<06:53,  6.89s/it]
 88%|████████▊ | 453/512 [50:51<06:42,  6.83s/it]
                                                 
{'loss': '9.183e-05', 'grad_norm': '0.01249', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 13712256, 'tokens/trainable': 207055, 'epoch': '1.77'}
+
 88%|████████▊ | 453/512 [50:51<06:42,  6.83s/it]
 89%|████████▊ | 454/512 [50:57<06:31,  6.76s/it]
                                                 
{'loss': '2.162e-05', 'grad_norm': '0.0009246', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.61', 'tokens/total': 13742128, 'tokens/trainable': 207494, 'epoch': '1.773'}
+
 89%|████████▊ | 454/512 [50:57<06:31,  6.76s/it]
 89%|████████▉ | 455/512 [51:04<06:22,  6.72s/it]
                                                 
{'loss': '1.243e-05', 'grad_norm': '0.0004114', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.03', 'tokens/total': 13772160, 'tokens/trainable': 207956, 'epoch': '1.777'}
+
 89%|████████▉ | 455/512 [51:04<06:22,  6.72s/it]
 89%|████████▉ | 456/512 [51:11<06:15,  6.71s/it]
                                                 
{'loss': '3.033e-05', 'grad_norm': '0.001591', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.42', 'tokens/total': 13802448, 'tokens/trainable': 208408, 'epoch': '1.781'}
+
 89%|████████▉ | 456/512 [51:11<06:15,  6.71s/it]
 89%|████████▉ | 457/512 [51:18<06:09,  6.71s/it]
                                                 
{'loss': '3.379e-05', 'grad_norm': '0.003002', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.92', 'tokens/total': 13832880, 'tokens/trainable': 208889, 'epoch': '1.785'}
+
 89%|████████▉ | 457/512 [51:18<06:09,  6.71s/it]
 89%|████████▉ | 458/512 [51:24<06:01,  6.69s/it]
                                                 
{'loss': '3.313e-05', 'grad_norm': '0.001643', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.02', 'tokens/total': 13862880, 'tokens/trainable': 209336, 'epoch': '1.789'}
+
 89%|████████▉ | 458/512 [51:24<06:01,  6.69s/it]
 90%|████████▉ | 459/512 [51:31<05:54,  6.69s/it]
                                                 
{'loss': '1.784e-05', 'grad_norm': '0.0006427', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 13893440, 'tokens/trainable': 209797, 'epoch': '1.793'}
+
 90%|████████▉ | 459/512 [51:31<05:54,  6.69s/it]
 90%|████████▉ | 460/512 [51:38<05:47,  6.69s/it]
                                                 
{'loss': '3.284e-05', 'grad_norm': '0.001669', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.03', 'tokens/total': 13923824, 'tokens/trainable': 210240, 'epoch': '1.797'}
+
 90%|████████▉ | 460/512 [51:38<05:47,  6.69s/it]
 90%|█████████ | 461/512 [51:44<05:41,  6.69s/it]
                                                 
{'loss': '7.201e-05', 'grad_norm': '0.03455', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '34.9', 'tokens/total': 13954208, 'tokens/trainable': 210702, 'epoch': '1.801'}
+
 90%|█████████ | 461/512 [51:44<05:41,  6.69s/it]
 90%|█████████ | 462/512 [51:51<05:34,  6.70s/it]
                                                 
{'loss': '3.525e-05', 'grad_norm': '0.002149', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '31.44', 'tokens/total': 13984784, 'tokens/trainable': 211140, 'epoch': '1.805'}
+
 90%|█████████ | 462/512 [51:51<05:34,  6.70s/it]
 90%|█████████ | 463/512 [51:58<05:27,  6.69s/it]
                                                 
{'loss': '1.329e-05', 'grad_norm': '0.0007275', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.54', 'tokens/total': 14014896, 'tokens/trainable': 211575, 'epoch': '1.809'}
+
 90%|█████████ | 463/512 [51:58<05:27,  6.69s/it]
 91%|█████████ | 464/512 [52:04<05:21,  6.71s/it]
                                                 
{'loss': '0.0001454', 'grad_norm': '0.03174', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '38.86', 'tokens/total': 14045312, 'tokens/trainable': 212083, 'epoch': '1.812'}
+
 91%|█████████ | 464/512 [52:04<05:21,  6.71s/it]
 91%|█████████ | 465/512 [52:11<05:16,  6.73s/it]
                                                 
{'loss': '1.449e-05', 'grad_norm': '0.0005439', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 14075840, 'tokens/trainable': 212558, 'epoch': '1.816'}
+
 91%|█████████ | 465/512 [52:11<05:16,  6.73s/it]
 91%|█████████ | 466/512 [52:18<05:08,  6.72s/it]
                                                 
{'loss': '2.788e-05', 'grad_norm': '0.001548', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.97', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 14106016, 'tokens/trainable': 213014, 'epoch': '1.82'}
+
 91%|█████████ | 466/512 [52:18<05:08,  6.72s/it]
 91%|█████████ | 467/512 [52:25<05:02,  6.72s/it]
                                                 
{'loss': '2.111e-05', 'grad_norm': '0.0007428', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.99', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 14136512, 'tokens/trainable': 213465, 'epoch': '1.824'}
+
 91%|█████████ | 467/512 [52:25<05:02,  6.72s/it]
 91%|█████████▏| 468/512 [52:31<04:55,  6.73s/it]
                                                 
{'loss': '4.059e-05', 'grad_norm': '0.002183', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 14166832, 'tokens/trainable': 213940, 'epoch': '1.828'}
+
 91%|█████████▏| 468/512 [52:31<04:55,  6.73s/it]
 92%|█████████▏| 469/512 [52:38<04:49,  6.74s/it]
                                                 
{'loss': '3.665e-05', 'grad_norm': '0.004772', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.97', 'tokens/total': 14197360, 'tokens/trainable': 214415, 'epoch': '1.832'}
+
 92%|█████████▏| 469/512 [52:38<04:49,  6.74s/it]
 92%|█████████▏| 470/512 [52:45<04:43,  6.74s/it]
                                                 
{'loss': '1.112e-05', 'grad_norm': '0.0003869', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 14227600, 'tokens/trainable': 214880, 'epoch': '1.836'}
+
 92%|█████████▏| 470/512 [52:45<04:43,  6.74s/it]
 92%|█████████▏| 471/512 [52:52<04:35,  6.72s/it]
                                                 
{'loss': '2.091e-05', 'grad_norm': '0.000844', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.34', 'tokens/total': 14257888, 'tokens/trainable': 215334, 'epoch': '1.84'}
+
 92%|█████████▏| 471/512 [52:52<04:35,  6.72s/it]
 92%|█████████▏| 472/512 [52:58<04:29,  6.73s/it]
                                                 
{'loss': '1.179e-05', 'grad_norm': '0.0005927', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 14288208, 'tokens/trainable': 215822, 'epoch': '1.844'}
+
 92%|█████████▏| 472/512 [52:58<04:29,  6.73s/it]
 92%|█████████▏| 473/512 [53:05<04:22,  6.72s/it]
                                                 
{'loss': '5.021e-05', 'grad_norm': '0.003898', 'learning_rate': '1.149e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '30.82', 'tokens/total': 14318448, 'tokens/trainable': 216239, 'epoch': '1.848'}
+
 92%|█████████▏| 473/512 [53:05<04:22,  6.72s/it]
 93%|█████████▎| 474/512 [53:12<04:14,  6.70s/it]
                                                 
{'loss': '2.28e-05', 'grad_norm': '0.001381', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.35', 'tokens/total': 14348912, 'tokens/trainable': 216691, 'epoch': '1.852'}
+
 93%|█████████▎| 474/512 [53:12<04:14,  6.70s/it]
 93%|█████████▎| 475/512 [53:18<04:07,  6.70s/it]
                                                 
{'loss': '1.463e-05', 'grad_norm': '0.0004829', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 14379440, 'tokens/trainable': 217162, 'epoch': '1.855'}
+
 93%|█████████▎| 475/512 [53:18<04:07,  6.70s/it]
 93%|█████████▎| 476/512 [53:25<04:01,  6.71s/it]
                                                 
{'loss': '1.446e-05', 'grad_norm': '0.0004492', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.88', 'tokens/total': 14409840, 'tokens/trainable': 217638, 'epoch': '1.859'}
+
 93%|█████████▎| 476/512 [53:25<04:01,  6.71s/it]
 93%|█████████▎| 477/512 [53:32<03:54,  6.71s/it]
                                                 
{'loss': '4.934e-05', 'grad_norm': '0.002406', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 14440256, 'tokens/trainable': 218053, 'epoch': '1.863'}
+
 93%|█████████▎| 477/512 [53:32<03:54,  6.71s/it]
 93%|█████████▎| 478/512 [53:38<03:47,  6.70s/it]
                                                 
{'loss': '2.053e-05', 'grad_norm': '0.0008485', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 14470704, 'tokens/trainable': 218503, 'epoch': '1.867'}
+
 93%|█████████▎| 478/512 [53:38<03:47,  6.70s/it]
 94%|█████████▎| 479/512 [53:45<03:41,  6.70s/it]
                                                 
{'loss': '8.76e-05', 'grad_norm': '0.01195', 'learning_rate': '1.108e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '36.19', 'tokens/total': 14501072, 'tokens/trainable': 218964, 'epoch': '1.871'}
+
 94%|█████████▎| 479/512 [53:45<03:41,  6.70s/it]
 94%|█████████▍| 480/512 [53:52<03:33,  6.68s/it]
                                                 
{'loss': '7.079e-05', 'grad_norm': '0.004777', 'learning_rate': '1.102e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 14531200, 'tokens/trainable': 219412, 'epoch': '1.875'}
+
 94%|█████████▍| 480/512 [53:52<03:33,  6.68s/it][2026-08-18 14:58:55,402] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
+
 94%|█████████▍| 481/512 [54:00<03:45,  7.27s/it]
                                                 
{'loss': '1.676e-05', 'grad_norm': '0.0006404', 'learning_rate': '1.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '38.63', 'tokens/train_per_sec_per_gpu': '38.25', 'tokens/total': 14561152, 'tokens/trainable': 219886, 'epoch': '1.879'}
+
 94%|█████████▍| 481/512 [54:00<03:45,  7.27s/it]
 94%|█████████▍| 482/512 [54:07<03:32,  7.09s/it]
                                                 
{'loss': '2.334e-05', 'grad_norm': '0.005523', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '34.52', 'tokens/total': 14591264, 'tokens/trainable': 220343, 'epoch': '1.883'}
+
 94%|█████████▍| 482/512 [54:07<03:32,  7.09s/it]
 94%|█████████▍| 483/512 [54:14<03:22,  6.97s/it]
                                                 
{'loss': '2.361e-05', 'grad_norm': '0.001016', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 14621568, 'tokens/trainable': 220825, 'epoch': '1.887'}
+
 94%|█████████▍| 483/512 [54:14<03:22,  6.97s/it]
 95%|█████████▍| 484/512 [54:20<03:12,  6.89s/it]
                                                 
{'loss': '5.646e-05', 'grad_norm': '0.004034', 'learning_rate': '1.079e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.77', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 14652064, 'tokens/trainable': 221282, 'epoch': '1.891'}
+
 95%|█████████▍| 484/512 [54:20<03:12,  6.89s/it]
 95%|█████████▍| 485/512 [54:27<03:04,  6.84s/it]
                                                 
{'loss': '5.676e-05', 'grad_norm': '0.008658', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.89', 'tokens/total': 14682608, 'tokens/trainable': 221780, 'epoch': '1.895'}
+
 95%|█████████▍| 485/512 [54:27<03:04,  6.84s/it]
 95%|█████████▍| 486/512 [54:34<02:57,  6.81s/it]
                                                 
{'loss': '3.921e-05', 'grad_norm': '0.003652', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.56', 'tokens/total': 14713168, 'tokens/trainable': 222210, 'epoch': '1.898'}
+
 95%|█████████▍| 486/512 [54:34<02:57,  6.81s/it]
 95%|█████████▌| 487/512 [54:41<02:49,  6.77s/it]
                                                 
{'loss': '5.354e-05', 'grad_norm': '0.005618', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.37', 'tokens/total': 14743536, 'tokens/trainable': 222646, 'epoch': '1.902'}
+
 95%|█████████▌| 487/512 [54:41<02:49,  6.77s/it]
 95%|█████████▌| 488/512 [54:47<02:41,  6.74s/it]
                                                 
{'loss': '0.0006614', 'grad_norm': '0.07198', 'learning_rate': '1.058e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 14773728, 'tokens/trainable': 223066, 'epoch': '1.906'}
+
 95%|█████████▌| 488/512 [54:47<02:41,  6.74s/it]
 96%|█████████▌| 489/512 [54:54<02:34,  6.72s/it]
                                                 
{'loss': '1.627e-05', 'grad_norm': '0.00117', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 14803744, 'tokens/trainable': 223512, 'epoch': '1.91'}
+
 96%|█████████▌| 489/512 [54:54<02:34,  6.72s/it]
 96%|█████████▌| 490/512 [55:00<02:24,  6.56s/it]
                                                 
{'loss': '3.052e-05', 'grad_norm': '0.001455', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.95', 'tokens/total': 14832112, 'tokens/trainable': 223956, 'epoch': '1.914'}
+
 96%|█████████▌| 490/512 [55:00<02:24,  6.56s/it]
 96%|█████████▌| 491/512 [55:06<02:15,  6.45s/it]
                                                 
{'loss': '1.489e-05', 'grad_norm': '0.0007276', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 14860560, 'tokens/trainable': 224416, 'epoch': '1.918'}
+
 96%|█████████▌| 491/512 [55:06<02:15,  6.45s/it]
 96%|█████████▌| 492/512 [55:13<02:10,  6.52s/it]
                                                 
{'loss': '1.902e-05', 'grad_norm': '0.0007378', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 14890848, 'tokens/trainable': 224883, 'epoch': '1.922'}
+
 96%|█████████▌| 492/512 [55:13<02:10,  6.52s/it]
 96%|█████████▋| 493/512 [55:20<02:04,  6.56s/it]
                                                 
{'loss': '1.498e-05', 'grad_norm': '0.0006582', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.42', 'tokens/total': 14921008, 'tokens/trainable': 225324, 'epoch': '1.926'}
+
 96%|█████████▋| 493/512 [55:20<02:04,  6.56s/it]
 96%|█████████▋| 494/512 [55:26<01:58,  6.60s/it]
                                                 
{'loss': '1.008e-05', 'grad_norm': '0.0008491', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 14951360, 'tokens/trainable': 225783, 'epoch': '1.93'}
+
 96%|█████████▋| 494/512 [55:26<01:58,  6.60s/it]
 97%|█████████▋| 495/512 [55:33<01:52,  6.63s/it]
                                                 
{'loss': '2.822e-05', 'grad_norm': '0.002965', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.48', 'tokens/total': 14981552, 'tokens/trainable': 226281, 'epoch': '1.934'}
+
 97%|█████████▋| 495/512 [55:33<01:52,  6.63s/it]
 97%|█████████▋| 496/512 [55:40<01:45,  6.62s/it]
                                                 
{'loss': '2.592e-05', 'grad_norm': '0.00117', 'learning_rate': '1.027e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 15011584, 'tokens/trainable': 226735, 'epoch': '1.938'}
+
 97%|█████████▋| 496/512 [55:40<01:45,  6.62s/it]
 97%|█████████▋| 497/512 [55:46<01:39,  6.65s/it]
                                                 
{'loss': '9.912e-05', 'grad_norm': '0.006589', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 15042016, 'tokens/trainable': 227198, 'epoch': '1.941'}
+
 97%|█████████▋| 497/512 [55:46<01:39,  6.65s/it]
 97%|█████████▋| 498/512 [55:53<01:32,  6.63s/it]
                                                 
{'loss': '2.851e-05', 'grad_norm': '0.001317', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.68', 'memory/max_allocated (GiB)': '33.68', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '31.07', 'tokens/total': 15071984, 'tokens/trainable': 227623, 'epoch': '1.945'}
+
 97%|█████████▋| 498/512 [55:53<01:32,  6.63s/it]
 97%|█████████▋| 499/512 [56:00<01:26,  6.65s/it]
                                                 
{'loss': '2.48e-05', 'grad_norm': '0.001088', 'learning_rate': '1.018e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '32.17', 'tokens/total': 15102192, 'tokens/trainable': 228074, 'epoch': '1.949'}
+
 97%|█████████▋| 499/512 [56:00<01:26,  6.65s/it]
 98%|█████████▊| 500/512 [56:06<01:19,  6.66s/it]
                                                 
{'loss': '4.626e-05', 'grad_norm': '0.01028', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.59', 'tokens/total': 15132560, 'tokens/trainable': 228541, 'epoch': '1.953'}
+
 98%|█████████▊| 500/512 [56:06<01:19,  6.66s/it]
 98%|█████████▊| 501/512 [56:13<01:13,  6.67s/it]
                                                 
{'loss': '1.94e-05', 'grad_norm': '0.001143', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '37.78', 'tokens/total': 15163152, 'tokens/trainable': 229031, 'epoch': '1.957'}
+
 98%|█████████▊| 501/512 [56:13<01:13,  6.67s/it]
 98%|█████████▊| 502/512 [56:20<01:06,  6.68s/it]
                                                 
{'loss': '2.073e-05', 'grad_norm': '0.001288', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '38.76', 'tokens/total': 15193456, 'tokens/trainable': 229512, 'epoch': '1.961'}
+
 98%|█████████▊| 502/512 [56:20<01:06,  6.68s/it]
 98%|█████████▊| 503/512 [56:26<01:00,  6.67s/it]
                                                 
{'loss': '0.0001616', 'grad_norm': '0.02232', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 15223680, 'tokens/trainable': 229989, 'epoch': '1.965'}
+
 98%|█████████▊| 503/512 [56:26<01:00,  6.67s/it]
 98%|█████████▊| 504/512 [56:33<00:53,  6.67s/it]
                                                 
{'loss': '9.707e-06', 'grad_norm': '0.0003185', 'learning_rate': '1.008e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.84', 'tokens/total': 15254336, 'tokens/trainable': 230447, 'epoch': '1.969'}
+
 98%|█████████▊| 504/512 [56:33<00:53,  6.67s/it]
 99%|█████████▊| 505/512 [56:40<00:46,  6.67s/it]
                                                 
{'loss': '1.712e-05', 'grad_norm': '0.0005881', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.3', 'tokens/total': 15284496, 'tokens/trainable': 230879, 'epoch': '1.973'}
+
 99%|█████████▊| 505/512 [56:40<00:46,  6.67s/it]
 99%|█████████▉| 506/512 [56:46<00:40,  6.69s/it]
                                                 
{'loss': '1.641e-05', 'grad_norm': '0.00115', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.03', 'tokens/total': 15314976, 'tokens/trainable': 231348, 'epoch': '1.977'}
+
 99%|█████████▉| 506/512 [56:46<00:40,  6.69s/it]
 99%|█████████▉| 507/512 [56:53<00:33,  6.69s/it]
                                                 
{'loss': '2.183e-05', 'grad_norm': '0.0007461', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 15345280, 'tokens/trainable': 231819, 'epoch': '1.98'}
+
 99%|█████████▉| 507/512 [56:53<00:33,  6.69s/it]
 99%|█████████▉| 508/512 [57:00<00:26,  6.68s/it]
                                                 
{'loss': '4.144e-05', 'grad_norm': '0.003102', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 15375424, 'tokens/trainable': 232269, 'epoch': '1.984'}
+
 99%|█████████▉| 508/512 [57:00<00:26,  6.68s/it]
 99%|█████████▉| 509/512 [57:06<00:20,  6.67s/it]
                                                 
{'loss': '1.45e-05', 'grad_norm': '0.001036', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 15405872, 'tokens/trainable': 232716, 'epoch': '1.988'}
+
 99%|█████████▉| 509/512 [57:06<00:20,  6.67s/it]
100%|█████████▉| 510/512 [57:13<00:13,  6.68s/it]
                                                 
{'loss': '2.179e-05', 'grad_norm': '0.001914', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '36.04', 'tokens/total': 15436128, 'tokens/trainable': 233178, 'epoch': '1.992'}
+
100%|█████████▉| 510/512 [57:13<00:13,  6.68s/it]
100%|█████████▉| 511/512 [57:20<00:06,  6.68s/it]
                                                 
{'loss': '2.199e-05', 'grad_norm': '0.001745', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '35.93', 'tokens/total': 15466464, 'tokens/trainable': 233647, 'epoch': '1.996'}
+
100%|█████████▉| 511/512 [57:20<00:06,  6.68s/it]
100%|██████████| 512/512 [57:27<00:00,  6.70s/it]
                                                 
{'loss': '0.0005933', 'grad_norm': '0.06926', 'learning_rate': '1e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '38.44', 'tokens/train_per_sec_per_gpu': '29.4', 'tokens/total': 15496368, 'tokens/trainable': 234060, 'epoch': '2'}
+
100%|██████████| 512/512 [57:27<00:00,  6.70s/it][2026-08-18 15:02:30,225] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
+
                                                 
{'train_runtime': '3449', 'train_samples_per_second': '4.75', 'train_steps_per_second': '0.148', 'train_loss': '0.007172', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '38.44', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
+
100%|██████████| 512/512 [57:29<00:00,  6.70s/it]
100%|██████████| 512/512 [57:29<00:00,  6.74s/it]
+[2026-08-18 15:02:32,241] [INFO] [axolotl.train] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
+[2026-08-18 15:02:32,700] [INFO] [axolotl.train] Model successfully saved to /workspace/wave/training/checkpoints
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/trainer_state.final.json b/aft_wave_v2/coin_real_4x__agreement/training/trainer_state.final.json
index 2499b3b7e698733d7ad57d04aa419bb5db763c37..dd12f510a42892888ddf9cc0585226435f8e071d 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/trainer_state.final.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/trainer_state.final.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 0.870697021484375,
+      "grad_norm": 0.8591235280036926,
       "learning_rate": 0.0,
       "loss": 0.10251401364803314,
       "memory/device_reserved (GiB)": 34.94,
@@ -20,12 +20,12 @@
       "ppl": 1.10795,
       "step": 1,
       "tokens/total": 30464,
-      "tokens/train_per_sec_per_gpu": 23.98,
+      "tokens/train_per_sec_per_gpu": 30.01,
       "tokens/trainable": 470
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 0.8108459115028381,
+      "grad_norm": 0.8033757209777832,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.09678442776203156,
       "memory/device_reserved (GiB)": 35.83,
@@ -34,900 +34,900 @@
       "ppl": 1.10162,
       "step": 2,
       "tokens/total": 60800,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 922
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.8027335405349731,
+      "grad_norm": 1.0102914571762085,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.10952483862638474,
+      "loss": 0.10876177996397018,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.11575,
+      "ppl": 1.1149,
       "step": 3,
       "tokens/total": 91136,
-      "tokens/train_per_sec_per_gpu": 34.48,
+      "tokens/train_per_sec_per_gpu": 34.67,
       "tokens/trainable": 1396
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.0353018045425415,
+      "grad_norm": 2.2042534351348877,
       "learning_rate": 1.2e-05,
-      "loss": 0.10303406417369843,
+      "loss": 0.1031389832496643,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.10853,
+      "ppl": 1.10865,
       "step": 4,
       "tokens/total": 121552,
-      "tokens/train_per_sec_per_gpu": 38.01,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 1850
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.0778985023498535,
+      "grad_norm": 2.5755860805511475,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.10945924371480942,
+      "loss": 0.1097191572189331,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.11567,
+      "ppl": 1.11596,
       "step": 5,
       "tokens/total": 152304,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 34.92,
       "tokens/trainable": 2302
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.8929882645606995,
+      "grad_norm": 1.498002052307129,
       "learning_rate": 2e-05,
-      "loss": 0.08588902652263641,
+      "loss": 0.08722387254238129,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.08969,
+      "ppl": 1.09114,
       "step": 6,
       "tokens/total": 182448,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 2742
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 1.6409597396850586,
+      "grad_norm": 1.280110478401184,
       "learning_rate": 2.4e-05,
-      "loss": 0.07352827489376068,
+      "loss": 0.07383580505847931,
       "memory/device_reserved (GiB)": 35.85,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0763,
+      "ppl": 1.07663,
       "step": 7,
       "tokens/total": 212736,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 3208
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.5843520164489746,
+      "grad_norm": 1.6952791213989258,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.07798244059085846,
+      "loss": 0.08001460134983063,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0811,
+      "ppl": 1.0833,
       "step": 8,
       "tokens/total": 243024,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 3655
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 1.3725916147232056,
+      "grad_norm": 1.2223162651062012,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.04634054750204086,
+      "loss": 0.047361284494400024,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.04743,
+      "ppl": 1.0485,
       "step": 9,
       "tokens/total": 271264,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 4091
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.544938564300537,
+      "grad_norm": 2.902157783508301,
       "learning_rate": 3.6e-05,
-      "loss": 0.08677884936332703,
+      "loss": 0.09570425748825073,
       "memory/device_reserved (GiB)": 36.07,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.09066,
+      "ppl": 1.10043,
       "step": 10,
       "tokens/total": 301520,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.98,
       "tokens/trainable": 4553
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 1.6364734172821045,
+      "grad_norm": 2.1767208576202393,
       "learning_rate": 4e-05,
-      "loss": 0.07754456996917725,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.0828268975019455,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.08063,
+      "ppl": 1.08635,
       "step": 11,
       "tokens/total": 331808,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 4992
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 2.167391538619995,
+      "grad_norm": 3.15231990814209,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.11765319854021072,
-      "memory/device_reserved (GiB)": 36.07,
+      "loss": 0.12141455709934235,
+      "memory/device_reserved (GiB)": 36.08,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.12485,
+      "ppl": 1.12909,
       "step": 12,
       "tokens/total": 362224,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.32,
       "tokens/trainable": 5494
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.196911573410034,
+      "grad_norm": 2.3834526538848877,
       "learning_rate": 4.8e-05,
-      "loss": 0.03510797768831253,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.037937015295028687,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.03573,
+      "ppl": 1.03867,
       "step": 13,
       "tokens/total": 392432,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 35.04,
       "tokens/trainable": 5933
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.9654567241668701,
+      "grad_norm": 3.2587738037109375,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.061097435653209686,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.06514571607112885,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.063,
+      "ppl": 1.06731,
       "step": 14,
       "tokens/total": 422784,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.67,
       "tokens/trainable": 6369
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 1.934105396270752,
+      "grad_norm": 1.5818979740142822,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.05385493487119675,
-      "memory/device_reserved (GiB)": 36.15,
+      "loss": 0.05656517297029495,
+      "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.05533,
+      "ppl": 1.0582,
       "step": 15,
       "tokens/total": 453376,
-      "tokens/train_per_sec_per_gpu": 32.96,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 6820
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 1.4659016132354736,
+      "grad_norm": 1.4215443134307861,
       "learning_rate": 6e-05,
-      "loss": 0.052153222262859344,
+      "loss": 0.06070145219564438,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.05354,
+      "ppl": 1.06258,
       "step": 16,
       "tokens/total": 483504,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.85,
       "tokens/trainable": 7258
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 1.9650894403457642,
+      "grad_norm": 1.3065693378448486,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.05092189460992813,
+      "loss": 0.05027393624186516,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05224,
+      "ppl": 1.05156,
       "step": 17,
       "tokens/total": 514032,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 7710
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 0.6891724467277527,
+      "grad_norm": 0.6123363375663757,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.031155016273260117,
+      "loss": 0.028499452397227287,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03165,
+      "ppl": 1.02891,
       "step": 18,
       "tokens/total": 544432,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 8174
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.8662010431289673,
+      "grad_norm": 0.648410439491272,
       "learning_rate": 7.2e-05,
-      "loss": 0.03036138042807579,
+      "loss": 0.031143227592110634,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.03083,
+      "ppl": 1.03163,
       "step": 19,
       "tokens/total": 574880,
-      "tokens/train_per_sec_per_gpu": 34.37,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 8617
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 0.7999041080474854,
+      "grad_norm": 0.6737155318260193,
       "learning_rate": 7.6e-05,
-      "loss": 0.03458942472934723,
+      "loss": 0.030753308907151222,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.03519,
+      "ppl": 1.03123,
       "step": 20,
       "tokens/total": 605408,
-      "tokens/train_per_sec_per_gpu": 30.32,
+      "tokens/train_per_sec_per_gpu": 30.37,
       "tokens/trainable": 9037
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.5816919207572937,
+      "grad_norm": 0.7464718222618103,
       "learning_rate": 8e-05,
-      "loss": 0.02401110902428627,
+      "loss": 0.02451065182685852,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0243,
+      "ppl": 1.02481,
       "step": 21,
       "tokens/total": 635584,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.63,
       "tokens/trainable": 9503
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.6761882901191711,
+      "grad_norm": 0.9435750246047974,
       "learning_rate": 8.4e-05,
-      "loss": 0.01873329095542431,
+      "loss": 0.017626767978072166,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01891,
+      "ppl": 1.01778,
       "step": 22,
       "tokens/total": 665952,
-      "tokens/train_per_sec_per_gpu": 36.71,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 9972
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.9077537655830383,
+      "grad_norm": 0.6369844079017639,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.017490077763795853,
+      "loss": 0.013959845528006554,
       "memory/device_reserved (GiB)": 36.21,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01764,
+      "ppl": 1.01406,
       "step": 23,
       "tokens/total": 696240,
-      "tokens/train_per_sec_per_gpu": 37.39,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 10447
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 1.3226462602615356,
+      "grad_norm": 1.0666130781173706,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.028416279703378677,
+      "loss": 0.03303435444831848,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02882,
+      "ppl": 1.03359,
       "step": 24,
       "tokens/total": 726464,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 37.23,
       "tokens/trainable": 10899
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 0.9712215065956116,
+      "grad_norm": 1.0491507053375244,
       "learning_rate": 9.6e-05,
-      "loss": 0.025973526760935783,
+      "loss": 0.030840374529361725,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02631,
+      "ppl": 1.03132,
       "step": 25,
       "tokens/total": 756704,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 11360
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8638900518417358,
+      "grad_norm": 0.8108148574829102,
       "learning_rate": 0.0001,
-      "loss": 0.022434517741203308,
+      "loss": 0.03408072516322136,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.02269,
+      "ppl": 1.03467,
       "step": 26,
       "tokens/total": 786912,
-      "tokens/train_per_sec_per_gpu": 29.23,
+      "tokens/train_per_sec_per_gpu": 29.3,
       "tokens/trainable": 11775
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 0.6629000902175903,
+      "grad_norm": 0.507036030292511,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.014538668096065521,
+      "loss": 0.014000408351421356,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01464,
+      "ppl": 1.0141,
       "step": 27,
       "tokens/total": 815424,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.89,
       "tokens/trainable": 12242
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 0.3078136146068573,
+      "grad_norm": 0.618457555770874,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.01195458322763443,
+      "loss": 0.022290699183940887,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01203,
+      "ppl": 1.02254,
       "step": 28,
       "tokens/total": 845584,
-      "tokens/train_per_sec_per_gpu": 33.52,
+      "tokens/train_per_sec_per_gpu": 33.48,
       "tokens/trainable": 12696
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 1.1301876306533813,
+      "grad_norm": 0.4989492893218994,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.022538531571626663,
+      "loss": 0.025926023721694946,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02279,
+      "ppl": 1.02627,
       "step": 29,
       "tokens/total": 875808,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.34,
       "tokens/trainable": 13153
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 0.41090911626815796,
+      "grad_norm": 0.3578357696533203,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.011871461756527424,
+      "loss": 0.014475762844085693,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01194,
+      "ppl": 1.01458,
       "step": 30,
       "tokens/total": 904096,
-      "tokens/train_per_sec_per_gpu": 39.83,
+      "tokens/train_per_sec_per_gpu": 39.7,
       "tokens/trainable": 13624
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.6772723197937012,
+      "grad_norm": 0.4404466450214386,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.013700846582651138,
+      "loss": 0.015927618369460106,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0138,
+      "ppl": 1.01606,
       "step": 31,
       "tokens/total": 934400,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 14064
     },
     {
       "epoch": 0.125,
-      "grad_norm": 1.207811951637268,
+      "grad_norm": 0.5913511514663696,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.01185896061360836,
+      "loss": 0.019408874213695526,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01193,
+      "ppl": 1.0196,
       "step": 32,
       "tokens/total": 964832,
-      "tokens/train_per_sec_per_gpu": 38.9,
+      "tokens/train_per_sec_per_gpu": 38.92,
       "tokens/trainable": 14565
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.46513956785202026,
+      "grad_norm": 0.2795853614807129,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.012477721087634563,
+      "loss": 0.007002322003245354,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01256,
+      "ppl": 1.00703,
       "step": 33,
       "tokens/total": 995040,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 15005
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 1.7668761014938354,
+      "grad_norm": 1.1757413148880005,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.03285093232989311,
+      "loss": 0.021448152139782906,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0334,
+      "ppl": 1.02168,
       "step": 34,
       "tokens/total": 1024896,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 15459
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.7816088795661926,
+      "grad_norm": 0.3860406279563904,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.028746310621500015,
+      "loss": 0.01894465833902359,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02916,
+      "ppl": 1.01913,
       "step": 35,
       "tokens/total": 1054992,
-      "tokens/train_per_sec_per_gpu": 38.15,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 15960
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.683965265750885,
+      "grad_norm": 0.4803963005542755,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.020058901980519295,
+      "loss": 0.013810254633426666,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02026,
+      "ppl": 1.01391,
       "step": 36,
       "tokens/total": 1085280,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 16428
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.6797531843185425,
+      "grad_norm": 0.3357454836368561,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.010446591302752495,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.008235199376940727,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.0105,
+      "ppl": 1.00827,
       "step": 37,
       "tokens/total": 1115504,
-      "tokens/train_per_sec_per_gpu": 31.77,
+      "tokens/train_per_sec_per_gpu": 31.89,
       "tokens/trainable": 16884
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.8899193406105042,
+      "grad_norm": 0.9474877715110779,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.017391683533787727,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019564270973205566,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01754,
+      "ppl": 1.01976,
       "step": 38,
       "tokens/total": 1145712,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.02,
       "tokens/trainable": 17341
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.8132575750350952,
+      "grad_norm": 1.101553201675415,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.02252483367919922,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.017880277708172798,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02278,
+      "ppl": 1.01804,
       "step": 39,
       "tokens/total": 1176128,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.16,
       "tokens/trainable": 17786
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.4430502653121948,
+      "grad_norm": 0.7563315033912659,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.011966042220592499,
-      "memory/device_reserved (GiB)": 35.48,
+      "loss": 0.019336596131324768,
+      "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01204,
+      "ppl": 1.01952,
       "step": 40,
       "tokens/total": 1206576,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 18262
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.5074653029441833,
+      "grad_norm": 0.41576531529426575,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.014877484180033207,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.011620002798736095,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01499,
+      "ppl": 1.01169,
       "step": 41,
       "tokens/total": 1236848,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 18716
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 0.5221464037895203,
+      "grad_norm": 1.180986762046814,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.017510797828435898,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.04661658778786659,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01767,
+      "ppl": 1.04772,
       "step": 42,
       "tokens/total": 1267088,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 19157
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 0.6888790130615234,
+      "grad_norm": 0.7583066821098328,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.030433066189289093,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.029224852100014687,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0309,
+      "ppl": 1.02966,
       "step": 43,
       "tokens/total": 1297360,
-      "tokens/train_per_sec_per_gpu": 36.5,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 19647
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.5600388050079346,
+      "grad_norm": 0.18861345946788788,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.013720017857849598,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.007798851002007723,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01381,
+      "ppl": 1.00783,
       "step": 44,
       "tokens/total": 1327744,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 20119
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.4291926324367523,
+      "grad_norm": 0.35095125436782837,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.00872582383453846,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.012044938281178474,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00876,
+      "ppl": 1.01212,
       "step": 45,
       "tokens/total": 1358096,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 20560
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.944327712059021,
+      "grad_norm": 0.5071477890014648,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.025248996913433075,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.017856616526842117,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02557,
+      "ppl": 1.01802,
       "step": 46,
       "tokens/total": 1388240,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 20992
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.2425016313791275,
+      "grad_norm": 0.5569872260093689,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005067020654678345,
-      "memory/device_reserved (GiB)": 35.87,
+      "loss": 0.019440822303295135,
+      "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.26,
       "memory/max_allocated (GiB)": 33.26,
-      "ppl": 1.00508,
+      "ppl": 1.01963,
       "step": 47,
       "tokens/total": 1416240,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 21441
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 1.2363684177398682,
+      "grad_norm": 0.42062458395957947,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.03300227224826813,
+      "loss": 0.013018792495131493,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.03355,
+      "ppl": 1.0131,
       "step": 48,
       "tokens/total": 1446880,
-      "tokens/train_per_sec_per_gpu": 33.7,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 21901
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 0.7278413772583008,
+      "grad_norm": 0.30396750569343567,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.025975672528147697,
+      "loss": 0.0077492957934737206,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.02632,
+      "ppl": 1.00778,
       "step": 49,
       "tokens/total": 1477168,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 22341
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.37237733602523804,
+      "grad_norm": 2.044849395751953,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010315775871276855,
+      "loss": 0.014447445049881935,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01037,
+      "ppl": 1.01455,
       "step": 50,
       "tokens/total": 1507712,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 22833
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.25258293747901917,
+      "grad_norm": 0.9357694983482361,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.00485712755471468,
+      "loss": 0.012720856815576553,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00487,
+      "ppl": 1.0128,
       "step": 51,
       "tokens/total": 1537936,
-      "tokens/train_per_sec_per_gpu": 30.64,
+      "tokens/train_per_sec_per_gpu": 30.72,
       "tokens/trainable": 23277
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.738599419593811,
+      "grad_norm": 0.2881873548030853,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.01646406576037407,
+      "loss": 0.005877626594156027,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0166,
+      "ppl": 1.00589,
       "step": 52,
       "tokens/total": 1568352,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.63,
       "tokens/trainable": 23759
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 1.2733500003814697,
+      "grad_norm": 0.7577692270278931,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.019196398556232452,
+      "loss": 0.01451955921947956,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01938,
+      "ppl": 1.01463,
       "step": 53,
       "tokens/total": 1598992,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 24193
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.5316427946090698,
+      "grad_norm": 0.5201014280319214,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.006890955846756697,
+      "loss": 0.006856432184576988,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00691,
+      "ppl": 1.00688,
       "step": 54,
       "tokens/total": 1629488,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.59,
       "tokens/trainable": 24619
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.6924111843109131,
+      "grad_norm": 0.8399921655654907,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.021178584545850754,
+      "loss": 0.037967782467603683,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0214,
+      "ppl": 1.0387,
       "step": 55,
       "tokens/total": 1659696,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.85,
       "tokens/trainable": 25075
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.683601975440979,
+      "grad_norm": 0.8203506469726562,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.017589068040251732,
+      "loss": 0.012814272195100784,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.01774,
+      "ppl": 1.0129,
       "step": 56,
       "tokens/total": 1689872,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.55,
       "tokens/trainable": 25519
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.8593301177024841,
+      "grad_norm": 0.20874246954917908,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.025561584159731865,
+      "loss": 0.00539036700502038,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.02589,
+      "ppl": 1.0054,
       "step": 57,
       "tokens/total": 1720144,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.69,
       "tokens/trainable": 25966
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 0.2925783097743988,
+      "grad_norm": 0.4427756071090698,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.005790311843156815,
+      "loss": 0.009390819817781448,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00581,
+      "ppl": 1.00944,
       "step": 58,
       "tokens/total": 1750448,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 26423
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 0.6059477925300598,
+      "grad_norm": 0.7457786798477173,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.007334758993238211,
+      "loss": 0.019530881196260452,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00736,
+      "ppl": 1.01972,
       "step": 59,
       "tokens/total": 1781088,
-      "tokens/train_per_sec_per_gpu": 38.42,
+      "tokens/train_per_sec_per_gpu": 38.51,
       "tokens/trainable": 26934
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.29425033926963806,
+      "grad_norm": 0.13402195274829865,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.004620288498699665,
+      "loss": 0.0014544172445312142,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00463,
+      "ppl": 1.00146,
       "step": 60,
       "tokens/total": 1811344,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 27393
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.25868093967437744,
+      "grad_norm": 1.0385031700134277,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.0042824773117899895,
+      "loss": 0.019916968420147896,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00429,
+      "ppl": 1.02012,
       "step": 61,
       "tokens/total": 1841600,
-      "tokens/train_per_sec_per_gpu": 32.84,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 27852
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.9455181360244751,
+      "grad_norm": 1.2596747875213623,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.013508133590221405,
+      "loss": 0.015884939581155777,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0136,
+      "ppl": 1.01601,
       "step": 62,
       "tokens/total": 1872048,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 28292
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 1.149442434310913,
+      "grad_norm": 0.14031143486499786,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.020864056423306465,
+      "loss": 0.002796083688735962,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02108,
+      "ppl": 1.0028,
       "step": 63,
       "tokens/total": 1902592,
-      "tokens/train_per_sec_per_gpu": 37.27,
+      "tokens/train_per_sec_per_gpu": 37.36,
       "tokens/trainable": 28798
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.7421550750732422,
+      "grad_norm": 0.6247786283493042,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.016041038557887077,
+      "loss": 0.017117884010076523,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01617,
+      "ppl": 1.01727,
       "step": 64,
       "tokens/total": 1933088,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.79,
       "tokens/trainable": 29283
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.36109936237335205,
+      "grad_norm": 0.3513385057449341,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.005372575484216213,
+      "loss": 0.008029159158468246,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00539,
+      "ppl": 1.00806,
       "step": 65,
       "tokens/total": 1963296,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 29745
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.7074101567268372,
+      "grad_norm": 0.279448539018631,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.018492329865694046,
-      "memory/device_reserved (GiB)": 34.88,
+      "loss": 0.0161186084151268,
+      "memory/device_reserved (GiB)": 34.87,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01866,
+      "ppl": 1.01625,
       "step": 66,
       "tokens/total": 1993760,
       "tokens/train_per_sec_per_gpu": 31.19,
@@ -935,1007 +935,1007 @@
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.43113431334495544,
+      "grad_norm": 0.31739094853401184,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00620780885219574,
+      "loss": 0.009091717191040516,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00623,
+      "ppl": 1.00913,
       "step": 67,
       "tokens/total": 2024144,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.36,
       "tokens/trainable": 30622
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.3996214270591736,
+      "grad_norm": 0.3539387285709381,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.005429963115602732,
+      "loss": 0.00908343680202961,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00544,
+      "ppl": 1.00912,
       "step": 68,
       "tokens/total": 2054608,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 31097
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 0.4444175660610199,
+      "grad_norm": 0.6497699618339539,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.01158289983868599,
+      "loss": 0.03086906298995018,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01165,
+      "ppl": 1.03135,
       "step": 69,
       "tokens/total": 2084912,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.6,
       "tokens/trainable": 31595
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.16977320611476898,
+      "grad_norm": 0.19664841890335083,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002158569637686014,
+      "loss": 0.0039155250415205956,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00216,
+      "ppl": 1.00392,
       "step": 70,
       "tokens/total": 2115216,
-      "tokens/train_per_sec_per_gpu": 30.67,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 32036
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.12970401346683502,
+      "grad_norm": 0.4300064146518707,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.0028989531565457582,
+      "loss": 0.007546662352979183,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0029,
+      "ppl": 1.00758,
       "step": 71,
       "tokens/total": 2145424,
-      "tokens/train_per_sec_per_gpu": 28.06,
+      "tokens/train_per_sec_per_gpu": 28.09,
       "tokens/trainable": 32428
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.06490105390548706,
+      "grad_norm": 0.027750927954912186,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.0008802044321782887,
+      "loss": 0.0004817460721824318,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00088,
+      "ppl": 1.00048,
       "step": 72,
       "tokens/total": 2175648,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 32880
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.20680083334445953,
+      "grad_norm": 0.11202923208475113,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.0019504247466102242,
+      "loss": 0.0012140646576881409,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00195,
+      "ppl": 1.00121,
       "step": 73,
       "tokens/total": 2205856,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.51,
       "tokens/trainable": 33323
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.6840267181396484,
+      "grad_norm": 0.026963796466588974,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.02435958757996559,
+      "loss": 0.00046908354852348566,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.02466,
+      "ppl": 1.00047,
       "step": 74,
       "tokens/total": 2236352,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.86,
       "tokens/trainable": 33792
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.6665006875991821,
+      "grad_norm": 0.5172365307807922,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.018271014094352722,
+      "loss": 0.029645610600709915,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01844,
+      "ppl": 1.03009,
       "step": 75,
       "tokens/total": 2266480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 34223
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.6469210982322693,
+      "grad_norm": 0.84085613489151,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.023723380640149117,
+      "loss": 0.006980063393712044,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.02401,
+      "ppl": 1.007,
       "step": 76,
       "tokens/total": 2296496,
-      "tokens/train_per_sec_per_gpu": 29.59,
+      "tokens/train_per_sec_per_gpu": 29.61,
       "tokens/trainable": 34649
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 1.391882300376892,
+      "grad_norm": 3.4935519695281982,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.015201358124613762,
+      "loss": 0.008870027028024197,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01532,
+      "ppl": 1.00891,
       "step": 77,
       "tokens/total": 2327040,
-      "tokens/train_per_sec_per_gpu": 35.58,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 35094
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.5752553343772888,
+      "grad_norm": 0.9529178142547607,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.020247019827365875,
+      "loss": 0.032071419060230255,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02045,
+      "ppl": 1.03259,
       "step": 78,
       "tokens/total": 2357328,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.82,
       "tokens/trainable": 35534
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.276022732257843,
+      "grad_norm": 0.5781691074371338,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.010471204295754433,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01475254725664854,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.01486,
       "step": 79,
       "tokens/total": 2387664,
-      "tokens/train_per_sec_per_gpu": 35.76,
+      "tokens/train_per_sec_per_gpu": 35.82,
       "tokens/trainable": 36029
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.41784003376960754,
+      "grad_norm": 0.15085959434509277,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.010774495080113411,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002578896936029196,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01083,
+      "ppl": 1.00258,
       "step": 80,
       "tokens/total": 2418000,
-      "tokens/train_per_sec_per_gpu": 35.55,
+      "tokens/train_per_sec_per_gpu": 35.53,
       "tokens/trainable": 36469
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4906325340270996,
+      "grad_norm": 0.19004814326763153,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.015422273427248001,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004292497877031565,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01554,
+      "ppl": 1.0043,
       "step": 81,
       "tokens/total": 2448720,
-      "tokens/train_per_sec_per_gpu": 30.69,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 36906
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.2813898026943207,
+      "grad_norm": 0.4505981504917145,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.009146124124526978,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.018529793247580528,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00919,
+      "ppl": 1.0187,
       "step": 82,
       "tokens/total": 2479248,
-      "tokens/train_per_sec_per_gpu": 38.89,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 37390
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.39496278762817383,
+      "grad_norm": 0.4981430470943451,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.008894146420061588,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.014818452298641205,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00893,
+      "ppl": 1.01493,
       "step": 83,
       "tokens/total": 2509408,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.8,
       "tokens/trainable": 37838
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.4946168065071106,
+      "grad_norm": 0.16379471123218536,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.005293373484164476,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.003185997949913144,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00531,
+      "ppl": 1.00319,
       "step": 84,
       "tokens/total": 2539776,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 38310
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 1.3293001651763916,
+      "grad_norm": 0.40732133388519287,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.040361277759075165,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.020803559571504593,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04119,
+      "ppl": 1.02102,
       "step": 85,
       "tokens/total": 2569840,
-      "tokens/train_per_sec_per_gpu": 34.0,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 38789
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.5834341049194336,
+      "grad_norm": 0.3687220513820648,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.010128681547939777,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.012931328266859055,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01018,
+      "ppl": 1.01302,
       "step": 86,
       "tokens/total": 2600192,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 39277
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.7191532254219055,
+      "grad_norm": 0.367418110370636,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.02042745053768158,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01382569782435894,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.02064,
+      "ppl": 1.01392,
       "step": 87,
       "tokens/total": 2630640,
-      "tokens/train_per_sec_per_gpu": 35.6,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 39737
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.3419296145439148,
+      "grad_norm": 0.2704487144947052,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.007800046354532242,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.008717816323041916,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00783,
+      "ppl": 1.00876,
       "step": 88,
       "tokens/total": 2660832,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 29.6,
       "tokens/trainable": 40179
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.23275785148143768,
+      "grad_norm": 3.374918222427368,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.005660225171595812,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.01938408613204956,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00568,
+      "ppl": 1.01957,
       "step": 89,
       "tokens/total": 2691328,
-      "tokens/train_per_sec_per_gpu": 32.59,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 40649
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.6987941265106201,
+      "grad_norm": 0.4690157175064087,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.02007928490638733,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.02013186179101467,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.02028,
+      "ppl": 1.02034,
       "step": 90,
       "tokens/total": 2719696,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 41080
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.11328475177288055,
+      "grad_norm": 0.08705829828977585,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.002234571846202016,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.002885152120143175,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00224,
+      "ppl": 1.00289,
       "step": 91,
       "tokens/total": 2750272,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 41599
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.4347783029079437,
+      "grad_norm": 8.197907447814941,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.004549161531031132,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.00822490081191063,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00456,
+      "ppl": 1.00826,
       "step": 92,
       "tokens/total": 2780656,
-      "tokens/train_per_sec_per_gpu": 36.77,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 42076
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.3486956059932709,
+      "grad_norm": 0.14939527213573456,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.012267105281352997,
-      "memory/device_reserved (GiB)": 35.88,
+      "loss": 0.004234164021909237,
+      "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01234,
+      "ppl": 1.00424,
       "step": 93,
       "tokens/total": 2811088,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 42522
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.5156503319740295,
+      "grad_norm": 0.15404288470745087,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.015529165044426918,
+      "loss": 0.009490479715168476,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01565,
+      "ppl": 1.00954,
       "step": 94,
       "tokens/total": 2841520,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 42974
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.37648338079452515,
+      "grad_norm": 0.5274825692176819,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.011707151308655739,
+      "loss": 0.02413639798760414,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.01178,
+      "ppl": 1.02443,
       "step": 95,
       "tokens/total": 2871600,
-      "tokens/train_per_sec_per_gpu": 30.71,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 43380
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.38365671038627625,
+      "grad_norm": 0.5182522535324097,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.008411398157477379,
+      "loss": 0.009927366860210896,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00845,
+      "ppl": 1.00998,
       "step": 96,
       "tokens/total": 2902144,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 43865
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.313085675239563,
+      "grad_norm": 0.5578822493553162,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.007852522656321526,
+      "loss": 0.014485684223473072,
       "memory/device_reserved (GiB)": 36.1,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00788,
+      "ppl": 1.01459,
       "step": 97,
       "tokens/total": 2932272,
-      "tokens/train_per_sec_per_gpu": 31.89,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 44328
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.08544483780860901,
+      "grad_norm": 0.10520734637975693,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0025753644295036793,
+      "loss": 0.0021552909165620804,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00258,
+      "ppl": 1.00216,
       "step": 98,
       "tokens/total": 2962032,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 30.95,
       "tokens/trainable": 44778
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.6496388912200928,
+      "grad_norm": 0.7544558644294739,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.02547256276011467,
+      "loss": 0.01703210547566414,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0258,
+      "ppl": 1.01718,
       "step": 99,
       "tokens/total": 2992256,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.12,
       "tokens/trainable": 45225
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.15325438976287842,
+      "grad_norm": 0.3857012689113617,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.003552855923771858,
+      "loss": 0.00604570098221302,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00356,
+      "ppl": 1.00606,
       "step": 100,
       "tokens/total": 3022608,
-      "tokens/train_per_sec_per_gpu": 35.8,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 45678
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.25307565927505493,
+      "grad_norm": 0.21589453518390656,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.004095804411917925,
+      "loss": 0.005753816105425358,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0041,
+      "ppl": 1.00577,
       "step": 101,
       "tokens/total": 3052992,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 46125
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 0.150072380900383,
+      "grad_norm": 1.1246687173843384,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.0013888315297663212,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.006994037888944149,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00139,
+      "ppl": 1.00702,
       "step": 102,
       "tokens/total": 3083392,
-      "tokens/train_per_sec_per_gpu": 39.71,
+      "tokens/train_per_sec_per_gpu": 39.6,
       "tokens/trainable": 46612
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 0.5769571661949158,
+      "grad_norm": 0.267713725566864,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.022533349692821503,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.013721915893256664,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02279,
+      "ppl": 1.01382,
       "step": 103,
       "tokens/total": 3113744,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.41,
       "tokens/trainable": 47059
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.5657027959823608,
+      "grad_norm": 0.325897753238678,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.016894506290555,
-      "memory/device_reserved (GiB)": 35.79,
+      "loss": 0.012948594987392426,
+      "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.01704,
+      "ppl": 1.01303,
       "step": 104,
       "tokens/total": 3144272,
-      "tokens/train_per_sec_per_gpu": 34.05,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 47536
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.29607170820236206,
+      "grad_norm": 0.531863808631897,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.017974723130464554,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.02244492992758751,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01814,
+      "ppl": 1.0227,
       "step": 105,
       "tokens/total": 3174512,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 48037
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 0.5241922736167908,
+      "grad_norm": 0.3957497775554657,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.016189826652407646,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.015801995992660522,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01632,
+      "ppl": 1.01593,
       "step": 106,
       "tokens/total": 3204896,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 48491
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.3886408805847168,
+      "grad_norm": 0.4241825044155121,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.010908558964729309,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.016320914030075073,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.01097,
+      "ppl": 1.01645,
       "step": 107,
       "tokens/total": 3235360,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 48940
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.2442784607410431,
+      "grad_norm": 0.32064536213874817,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.008070861920714378,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.010785036720335484,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0081,
+      "ppl": 1.01084,
       "step": 108,
       "tokens/total": 3265552,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 49389
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.1457175612449646,
+      "grad_norm": 0.17215749621391296,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.003785747569054365,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.00814715214073658,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00379,
+      "ppl": 1.00818,
       "step": 109,
       "tokens/total": 3293856,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 49838
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 0.3955559730529785,
+      "grad_norm": 0.38179653882980347,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.0069918157532811165,
-      "memory/device_reserved (GiB)": 37.61,
+      "loss": 0.012128787115216255,
+      "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00702,
+      "ppl": 1.0122,
       "step": 110,
       "tokens/total": 3323984,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 50310
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 0.6224751472473145,
+      "grad_norm": 0.49835413694381714,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.014852076768875122,
+      "loss": 0.015248995274305344,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01496,
+      "ppl": 1.01537,
       "step": 111,
       "tokens/total": 3354320,
-      "tokens/train_per_sec_per_gpu": 33.67,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 50755
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3700723648071289,
+      "grad_norm": 0.43904298543930054,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.00436755083501339,
+      "loss": 0.011272929608821869,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00438,
+      "ppl": 1.01134,
       "step": 112,
       "tokens/total": 3384880,
-      "tokens/train_per_sec_per_gpu": 31.65,
+      "tokens/train_per_sec_per_gpu": 31.5,
       "tokens/trainable": 51213
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.1353236883878708,
+      "grad_norm": 0.5670213103294373,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.0025976570323109627,
+      "loss": 0.0028921598568558693,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0026,
+      "ppl": 1.0029,
       "step": 113,
       "tokens/total": 3415104,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.77,
       "tokens/trainable": 51660
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.18984447419643402,
+      "grad_norm": 0.159476637840271,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.002497302368283272,
+      "loss": 0.0025574099272489548,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0025,
+      "ppl": 1.00256,
       "step": 114,
       "tokens/total": 3445504,
-      "tokens/train_per_sec_per_gpu": 32.43,
+      "tokens/train_per_sec_per_gpu": 32.34,
       "tokens/trainable": 52133
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 1.1815483570098877,
+      "grad_norm": 0.795085608959198,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.029489168897271156,
+      "loss": 0.024456799030303955,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02993,
+      "ppl": 1.02476,
       "step": 115,
       "tokens/total": 3475744,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.25,
       "tokens/trainable": 52580
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.44360846281051636,
+      "grad_norm": 0.38324710726737976,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.009585533291101456,
+      "loss": 0.004516632296144962,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00963,
+      "ppl": 1.00453,
       "step": 116,
       "tokens/total": 3506016,
-      "tokens/train_per_sec_per_gpu": 34.83,
+      "tokens/train_per_sec_per_gpu": 34.74,
       "tokens/trainable": 53033
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 0.5074551701545715,
+      "grad_norm": 0.3037130534648895,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.020201388746500015,
+      "loss": 0.021973589435219765,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.02041,
+      "ppl": 1.02222,
       "step": 117,
       "tokens/total": 3536256,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.22,
       "tokens/trainable": 53517
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.3565296232700348,
+      "grad_norm": 1.3904820680618286,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.005688562989234924,
+      "loss": 0.009210974909365177,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0057,
+      "ppl": 1.00925,
       "step": 118,
       "tokens/total": 3566480,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.59,
       "tokens/trainable": 53967
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.16547706723213196,
+      "grad_norm": 0.1571500301361084,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.003188834059983492,
+      "loss": 0.0021451227366924286,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00319,
+      "ppl": 1.00215,
       "step": 119,
       "tokens/total": 3596624,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 54430
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.20722293853759766,
+      "grad_norm": 0.39999091625213623,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.003235103562474251,
+      "loss": 0.00788091216236353,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00324,
+      "ppl": 1.00791,
       "step": 120,
       "tokens/total": 3627168,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 54896
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.4754939377307892,
+      "grad_norm": 0.38124287128448486,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.005670893006026745,
+      "loss": 0.010133227333426476,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00569,
+      "ppl": 1.01018,
       "step": 121,
       "tokens/total": 3657632,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 55379
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.08304762095212936,
+      "grad_norm": 0.0512852780520916,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.00099027412943542,
+      "loss": 0.0007940311916172504,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00099,
+      "ppl": 1.00079,
       "step": 122,
       "tokens/total": 3687824,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 55803
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.725281298160553,
+      "grad_norm": 0.13324694335460663,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.007413266692310572,
+      "loss": 0.0023267122451215982,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00744,
+      "ppl": 1.00233,
       "step": 123,
       "tokens/total": 3718320,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 56249
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.31409645080566406,
+      "grad_norm": 0.230186328291893,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.0029702766332775354,
+      "loss": 0.0037590472493320704,
       "memory/device_reserved (GiB)": 37.62,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00297,
+      "ppl": 1.00377,
       "step": 124,
       "tokens/total": 3748672,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.43,
       "tokens/trainable": 56713
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.6082578897476196,
+      "grad_norm": 0.30578872561454773,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.003795543685555458,
+      "loss": 0.00473653944209218,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0038,
+      "ppl": 1.00475,
       "step": 125,
       "tokens/total": 3778976,
-      "tokens/train_per_sec_per_gpu": 39.08,
+      "tokens/train_per_sec_per_gpu": 39.1,
       "tokens/trainable": 57196
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.0603976845741272,
+      "grad_norm": 0.3714931607246399,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0014751165872439742,
+      "loss": 0.005811735987663269,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00148,
+      "ppl": 1.00583,
       "step": 126,
       "tokens/total": 3809296,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 57680
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.21112751960754395,
+      "grad_norm": 0.11054892838001251,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.004137102514505386,
+      "loss": 0.0032034481409937143,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00415,
+      "ppl": 1.00321,
       "step": 127,
       "tokens/total": 3839328,
-      "tokens/train_per_sec_per_gpu": 31.8,
+      "tokens/train_per_sec_per_gpu": 31.88,
       "tokens/trainable": 58110
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.3234494924545288,
+      "grad_norm": 0.09251131862401962,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.007372056134045124,
+      "loss": 0.001724504167214036,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0074,
+      "ppl": 1.00173,
       "step": 128,
       "tokens/total": 3869824,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 58534
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.5826171636581421,
+      "grad_norm": 0.22758308053016663,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.007980267517268658,
+      "loss": 0.004688034299761057,
       "memory/device_reserved (GiB)": 37.76,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00801,
+      "ppl": 1.0047,
       "step": 129,
       "tokens/total": 3900080,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 58998
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.3384500741958618,
+      "grad_norm": 0.2881723642349243,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.005637750029563904,
+      "loss": 0.008266786113381386,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00565,
+      "ppl": 1.0083,
       "step": 130,
       "tokens/total": 3930560,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 59448
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 0.2838669717311859,
+      "grad_norm": 0.18802326917648315,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.0037927688099443913,
+      "loss": 0.00283675454556942,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.0038,
+      "ppl": 1.00284,
       "step": 131,
       "tokens/total": 3960496,
-      "tokens/train_per_sec_per_gpu": 31.97,
+      "tokens/train_per_sec_per_gpu": 31.92,
       "tokens/trainable": 59906
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.4598330855369568,
+      "grad_norm": 0.6749681234359741,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.010232537053525448,
-      "memory/device_reserved (GiB)": 35.94,
+      "loss": 0.012044447474181652,
+      "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01029,
+      "ppl": 1.01212,
       "step": 132,
       "tokens/total": 3990928,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 60350
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 0.7072780728340149,
+      "grad_norm": 0.5789079070091248,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.00946755986660719,
-      "memory/device_reserved (GiB)": 37.17,
+      "loss": 0.011312158778309822,
+      "memory/device_reserved (GiB)": 37.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00951,
+      "ppl": 1.01138,
       "step": 133,
       "tokens/total": 4021264,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 60831
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.3642464280128479,
+      "grad_norm": 0.47060829401016235,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0067742373794317245,
+      "loss": 0.0308814849704504,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0068,
+      "ppl": 1.03136,
       "step": 134,
       "tokens/total": 4051728,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 61305
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.1071263924241066,
+      "grad_norm": 0.16633495688438416,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.0022848297376185656,
+      "loss": 0.005535767879337072,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00229,
+      "ppl": 1.00555,
       "step": 135,
       "tokens/total": 4081920,
-      "tokens/train_per_sec_per_gpu": 37.57,
+      "tokens/train_per_sec_per_gpu": 37.62,
       "tokens/trainable": 61792
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.22470054030418396,
+      "grad_norm": 0.10699360817670822,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.0016683072317391634,
+      "loss": 0.002973862923681736,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00167,
+      "ppl": 1.00298,
       "step": 136,
       "tokens/total": 4112304,
-      "tokens/train_per_sec_per_gpu": 31.33,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 62253
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.5423188209533691,
+      "grad_norm": 0.2920030951499939,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.017443198710680008,
+      "loss": 0.01169741339981556,
       "memory/device_reserved (GiB)": 38.55,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0176,
+      "ppl": 1.01177,
       "step": 137,
       "tokens/total": 4142512,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.65,
       "tokens/trainable": 62707
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.47486332058906555,
+      "grad_norm": 0.3187088370323181,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.01525629311800003,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.011357840150594711,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01537,
+      "ppl": 1.01142,
       "step": 138,
       "tokens/total": 4172992,
       "tokens/train_per_sec_per_gpu": 29.32,
@@ -1943,139 +1943,139 @@
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.055354099720716476,
+      "grad_norm": 0.18588471412658691,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0010598574299365282,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006680965423583984,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00106,
+      "ppl": 1.0067,
       "step": 139,
       "tokens/total": 4203136,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 63587
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 0.15273842215538025,
+      "grad_norm": 0.2760343551635742,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.002772743348032236,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.01477967668324709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00278,
+      "ppl": 1.01489,
       "step": 140,
       "tokens/total": 4233456,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.74,
       "tokens/trainable": 64052
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.21690180897712708,
+      "grad_norm": 0.49880966544151306,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.005170213058590889,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.016770213842391968,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00518,
+      "ppl": 1.01691,
       "step": 141,
       "tokens/total": 4263728,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 64526
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.1491464525461197,
+      "grad_norm": 0.181757390499115,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0031757252290844917,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006471520289778709,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00318,
+      "ppl": 1.00649,
       "step": 142,
       "tokens/total": 4294432,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 64983
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.19370807707309723,
+      "grad_norm": 0.209610253572464,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.0033013438805937767,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006492790300399065,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00331,
+      "ppl": 1.00651,
       "step": 143,
       "tokens/total": 4324960,
-      "tokens/train_per_sec_per_gpu": 37.6,
+      "tokens/train_per_sec_per_gpu": 37.58,
       "tokens/trainable": 65492
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.43046337366104126,
+      "grad_norm": 0.13360266387462616,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.004675615578889847,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.003702069167047739,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00469,
+      "ppl": 1.00371,
       "step": 144,
       "tokens/total": 4355424,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 65925
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 1.153432011604309,
+      "grad_norm": 0.1640344262123108,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.00530653540045023,
-      "memory/device_reserved (GiB)": 38.55,
+      "loss": 0.006490131840109825,
+      "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00532,
+      "ppl": 1.00651,
       "step": 145,
       "tokens/total": 4385712,
-      "tokens/train_per_sec_per_gpu": 31.24,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66383
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.7740430235862732,
+      "grad_norm": 0.13646955788135529,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.01281517744064331,
+      "loss": 0.0026589329354465008,
       "memory/device_reserved (GiB)": 38.56,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0129,
+      "ppl": 1.00266,
       "step": 146,
       "tokens/total": 4416016,
-      "tokens/train_per_sec_per_gpu": 31.37,
+      "tokens/train_per_sec_per_gpu": 31.39,
       "tokens/trainable": 66840
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.6978983879089355,
+      "grad_norm": 0.4220513701438904,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.0272100567817688,
+      "loss": 0.01334389764815569,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02758,
+      "ppl": 1.01343,
       "step": 147,
       "tokens/total": 4446368,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 67297
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.17344872653484344,
+      "grad_norm": 0.37345919013023376,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.008015683852136135,
+      "loss": 0.012794861570000648,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00805,
+      "ppl": 1.01288,
       "step": 148,
       "tokens/total": 4476880,
       "tokens/train_per_sec_per_gpu": 37.36,
@@ -2083,377 +2083,377 @@
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.6355595588684082,
+      "grad_norm": 0.3093344271183014,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.010829147882759571,
+      "loss": 0.012492594309151173,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01089,
+      "ppl": 1.01257,
       "step": 149,
       "tokens/total": 4507232,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.83,
       "tokens/trainable": 68257
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.25094935297966003,
+      "grad_norm": 0.2453778088092804,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.0052955676801502705,
+      "loss": 0.003456964623183012,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00531,
+      "ppl": 1.00346,
       "step": 150,
       "tokens/total": 4537376,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.52,
       "tokens/trainable": 68706
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.5292705297470093,
+      "grad_norm": 0.337802916765213,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.022014575079083443,
+      "loss": 0.009631449356675148,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.02226,
+      "ppl": 1.00968,
       "step": 151,
       "tokens/total": 4567584,
-      "tokens/train_per_sec_per_gpu": 33.25,
+      "tokens/train_per_sec_per_gpu": 33.3,
       "tokens/trainable": 69131
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.3471219539642334,
+      "grad_norm": 0.34801673889160156,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.024585288017988205,
+      "loss": 0.02037646435201168,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02489,
+      "ppl": 1.02059,
       "step": 152,
       "tokens/total": 4597792,
-      "tokens/train_per_sec_per_gpu": 31.92,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 69586
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.08056659996509552,
+      "grad_norm": 0.03586283326148987,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.002611964475363493,
+      "loss": 0.0008243054617196321,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00262,
+      "ppl": 1.00082,
       "step": 153,
       "tokens/total": 4627936,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 70061
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.162270650267601,
+      "grad_norm": 0.1870104819536209,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0033612053375691175,
+      "loss": 0.0038898580241948366,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00337,
+      "ppl": 1.0039,
       "step": 154,
       "tokens/total": 4658352,
-      "tokens/train_per_sec_per_gpu": 27.93,
+      "tokens/train_per_sec_per_gpu": 27.95,
       "tokens/trainable": 70467
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.17613235116004944,
+      "grad_norm": 0.2884272038936615,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.004349880386143923,
+      "loss": 0.005026768893003464,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00436,
+      "ppl": 1.00504,
       "step": 155,
       "tokens/total": 4688896,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 70933
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.2590649127960205,
+      "grad_norm": 0.07867873460054398,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.010615494102239609,
+      "loss": 0.0024596985895186663,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01067,
+      "ppl": 1.00246,
       "step": 156,
       "tokens/total": 4719168,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 71382
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.15640626847743988,
+      "grad_norm": 0.056005269289016724,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.002627612790092826,
+      "loss": 0.0012056033592671156,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00263,
+      "ppl": 1.00121,
       "step": 157,
       "tokens/total": 4749136,
-      "tokens/train_per_sec_per_gpu": 30.22,
+      "tokens/train_per_sec_per_gpu": 30.83,
       "tokens/trainable": 71808
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.34429433941841125,
+      "grad_norm": 0.34242892265319824,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.010873161256313324,
+      "loss": 0.007983298972249031,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01093,
+      "ppl": 1.00802,
       "step": 158,
       "tokens/total": 4779472,
-      "tokens/train_per_sec_per_gpu": 32.87,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 72249
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.3858713209629059,
+      "grad_norm": 0.36051586270332336,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.020050909370183945,
+      "loss": 0.011793753132224083,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02025,
+      "ppl": 1.01186,
       "step": 159,
       "tokens/total": 4809632,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 72726
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.3002466857433319,
+      "grad_norm": 0.376717746257782,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.009756345301866531,
+      "loss": 0.018659302964806557,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0098,
+      "ppl": 1.01883,
       "step": 160,
       "tokens/total": 4840112,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 73148
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.202926903963089,
+      "grad_norm": 0.20132119953632355,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.009084527380764484,
+      "loss": 0.007013507187366486,
       "memory/device_reserved (GiB)": 38.65,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.00913,
+      "ppl": 1.00704,
       "step": 161,
       "tokens/total": 4870656,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.53,
       "tokens/trainable": 73600
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.29608848690986633,
+      "grad_norm": 0.24618405103683472,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0125912856310606,
+      "loss": 0.00824644137173891,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01267,
+      "ppl": 1.00828,
       "step": 162,
       "tokens/total": 4900608,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.76,
       "tokens/trainable": 74068
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.11453798413276672,
+      "grad_norm": 0.2069157212972641,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0031142355874180794,
+      "loss": 0.007023319602012634,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00312,
+      "ppl": 1.00705,
       "step": 163,
       "tokens/total": 4930752,
-      "tokens/train_per_sec_per_gpu": 35.71,
+      "tokens/train_per_sec_per_gpu": 35.75,
       "tokens/trainable": 74527
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.12609371542930603,
+      "grad_norm": 0.1887698471546173,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0056648110039532185,
+      "loss": 0.0082007497549057,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00568,
+      "ppl": 1.00823,
       "step": 164,
       "tokens/total": 4961008,
-      "tokens/train_per_sec_per_gpu": 31.81,
+      "tokens/train_per_sec_per_gpu": 31.87,
       "tokens/trainable": 74992
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.49591395258903503,
+      "grad_norm": 0.17459234595298767,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.012539982795715332,
+      "loss": 0.004637294448912144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01262,
+      "ppl": 1.00465,
       "step": 165,
       "tokens/total": 4991200,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 75442
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.17120927572250366,
+      "grad_norm": 0.26009130477905273,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.007677854970097542,
-      "memory/device_reserved (GiB)": 35.99,
+      "loss": 0.007355842739343643,
+      "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00771,
+      "ppl": 1.00738,
       "step": 166,
       "tokens/total": 5021600,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.27,
       "tokens/trainable": 75887
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.19032779335975647,
+      "grad_norm": 0.2539709806442261,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.007361435331404209,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.010408539324998856,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00739,
+      "ppl": 1.01046,
       "step": 167,
       "tokens/total": 5052032,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 76331
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.08688601851463318,
+      "grad_norm": 0.12331778556108475,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0016894477885216475,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022821722086519003,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00169,
+      "ppl": 1.00228,
       "step": 168,
       "tokens/total": 5080256,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 76745
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 1.221700668334961,
+      "grad_norm": 0.110007144510746,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.02566530555486679,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0038710185326635838,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.026,
+      "ppl": 1.00388,
       "step": 169,
       "tokens/total": 5110784,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 77201
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.22431711852550507,
+      "grad_norm": 0.03679708018898964,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.005605725571513176,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000850176380481571,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00562,
+      "ppl": 1.00085,
       "step": 170,
       "tokens/total": 5141200,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 77655
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.06314318627119064,
+      "grad_norm": 0.12713676691055298,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0013656741939485073,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0018071834929287434,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00137,
+      "ppl": 1.00181,
       "step": 171,
       "tokens/total": 5171760,
-      "tokens/train_per_sec_per_gpu": 39.25,
+      "tokens/train_per_sec_per_gpu": 39.23,
       "tokens/trainable": 78122
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.3465789258480072,
+      "grad_norm": 0.1453125774860382,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.008099250495433807,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0017664346378296614,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00813,
+      "ppl": 1.00177,
       "step": 172,
       "tokens/total": 5202272,
-      "tokens/train_per_sec_per_gpu": 33.59,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 78576
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.453427255153656,
+      "grad_norm": 0.21252205967903137,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.020783744752407074,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.002328047761693597,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.021,
+      "ppl": 1.00233,
       "step": 173,
       "tokens/total": 5232800,
-      "tokens/train_per_sec_per_gpu": 36.49,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 79059
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 0.3241178095340729,
+      "grad_norm": 0.637407660484314,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.010957238264381886,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.007709754630923271,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.01102,
+      "ppl": 1.00774,
       "step": 174,
       "tokens/total": 5262672,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.72,
       "tokens/trainable": 79498
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.5538946986198425,
+      "grad_norm": 0.06158079952001572,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.009046275168657303,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0007811276591382921,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00909,
+      "ppl": 1.00078,
       "step": 175,
       "tokens/total": 5293072,
       "tokens/train_per_sec_per_gpu": 30.45,
@@ -2461,139 +2461,139 @@
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.26320791244506836,
+      "grad_norm": 0.04236136004328728,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.014828844927251339,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.000772522296756506,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.46,
       "memory/max_allocated (GiB)": 33.46,
-      "ppl": 1.01494,
+      "ppl": 1.00077,
       "step": 176,
       "tokens/total": 5321520,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 80383
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.20839811861515045,
+      "grad_norm": 0.0892096534371376,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.006698478478938341,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0023201322183012962,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00672,
+      "ppl": 1.00232,
       "step": 177,
       "tokens/total": 5351904,
-      "tokens/train_per_sec_per_gpu": 33.09,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 80824
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.3556506335735321,
+      "grad_norm": 0.679317831993103,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.009760214015841484,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.008408154360949993,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00981,
+      "ppl": 1.00844,
       "step": 178,
       "tokens/total": 5382432,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 81243
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.21442855894565582,
+      "grad_norm": 0.05571528524160385,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.009938797913491726,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00045867619337514043,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00999,
+      "ppl": 1.00046,
       "step": 179,
       "tokens/total": 5412960,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 81716
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.13008078932762146,
+      "grad_norm": 0.36156049370765686,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.005616464652121067,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0047795758582651615,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00563,
+      "ppl": 1.00479,
       "step": 180,
       "tokens/total": 5443232,
-      "tokens/train_per_sec_per_gpu": 35.61,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 82181
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.23816989362239838,
+      "grad_norm": 0.3550747036933899,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.0075413500890135765,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0044985488057136536,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.29,
       "memory/max_allocated (GiB)": 33.29,
-      "ppl": 1.00757,
+      "ppl": 1.00451,
       "step": 181,
       "tokens/total": 5471440,
-      "tokens/train_per_sec_per_gpu": 34.63,
+      "tokens/train_per_sec_per_gpu": 34.65,
       "tokens/trainable": 82626
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.13832826912403107,
+      "grad_norm": 0.00732263270765543,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.0035294657573103905,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012463401071727276,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00354,
+      "ppl": 1.00012,
       "step": 182,
       "tokens/total": 5501744,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 83089
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.08244283497333527,
+      "grad_norm": 0.005601493176072836,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.0022330794017761946,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00012698184582404792,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00224,
+      "ppl": 1.00013,
       "step": 183,
       "tokens/total": 5532272,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 83590
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.26471880078315735,
+      "grad_norm": 0.03298855572938919,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.011201716959476471,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00026586768217384815,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.01126,
+      "ppl": 1.00027,
       "step": 184,
       "tokens/total": 5562784,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 84046
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.10434233397245407,
+      "grad_norm": 0.02470102533698082,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0033001210540533066,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00024917692644521594,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00331,
+      "ppl": 1.00025,
       "step": 185,
       "tokens/total": 5592800,
       "tokens/train_per_sec_per_gpu": 34.05,
@@ -2601,223 +2601,223 @@
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.6545754075050354,
+      "grad_norm": 0.6944283246994019,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.033577777445316315,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.016732344403862953,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.03415,
+      "ppl": 1.01687,
       "step": 186,
       "tokens/total": 5623088,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 84974
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.016698423773050308,
+      "grad_norm": 0.011723512783646584,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.000414226611610502,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.00017266182112507522,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00041,
+      "ppl": 1.00017,
       "step": 187,
       "tokens/total": 5653344,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.07,
       "tokens/trainable": 85460
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.16612493991851807,
+      "grad_norm": 0.34870269894599915,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002929423237219453,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.004548810888081789,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00293,
+      "ppl": 1.00456,
       "step": 188,
       "tokens/total": 5683600,
-      "tokens/train_per_sec_per_gpu": 35.9,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 85939
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.3205801248550415,
+      "grad_norm": 0.09110172092914581,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0012500635348260403,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001469930517487228,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00125,
+      "ppl": 1.00147,
       "step": 189,
       "tokens/total": 5711952,
-      "tokens/train_per_sec_per_gpu": 38.26,
+      "tokens/train_per_sec_per_gpu": 38.43,
       "tokens/trainable": 86377
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.5194064378738403,
+      "grad_norm": 0.6382125020027161,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.009376855567097664,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.01117285992950201,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00942,
+      "ppl": 1.01124,
       "step": 190,
       "tokens/total": 5742336,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 86834
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.6358630061149597,
+      "grad_norm": 0.1396624743938446,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.009413158521056175,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0022487500682473183,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00946,
+      "ppl": 1.00225,
       "step": 191,
       "tokens/total": 5772560,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 87265
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.1093795970082283,
+      "grad_norm": 0.09118734300136566,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0017574415542185307,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.0009341652039438486,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00176,
+      "ppl": 1.00093,
       "step": 192,
       "tokens/total": 5803136,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 87747
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.0865081176161766,
+      "grad_norm": 0.10082298517227173,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0017616486875340343,
-      "memory/device_reserved (GiB)": 38.57,
+      "loss": 0.001603117911145091,
+      "memory/device_reserved (GiB)": 38.58,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00176,
+      "ppl": 1.0016,
       "step": 193,
       "tokens/total": 5833440,
-      "tokens/train_per_sec_per_gpu": 36.22,
+      "tokens/train_per_sec_per_gpu": 36.21,
       "tokens/trainable": 88239
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.01772180385887623,
+      "grad_norm": 0.0594109408557415,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 0.00022867789084557444,
+      "loss": 0.0009742019465193152,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00023,
+      "ppl": 1.00097,
       "step": 194,
       "tokens/total": 5863552,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 88718
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.2393598109483719,
+      "grad_norm": 0.12520930171012878,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.003605358535423875,
+      "loss": 0.0015380105469375849,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00361,
+      "ppl": 1.00154,
       "step": 195,
       "tokens/total": 5893680,
-      "tokens/train_per_sec_per_gpu": 31.27,
+      "tokens/train_per_sec_per_gpu": 31.29,
       "tokens/trainable": 89129
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.006237801164388657,
+      "grad_norm": 0.32357996702194214,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.000122636032756418,
+      "loss": 0.0033828348387032747,
       "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00012,
+      "ppl": 1.00339,
       "step": 196,
       "tokens/total": 5923744,
-      "tokens/train_per_sec_per_gpu": 32.25,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 89567
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.9131373763084412,
+      "grad_norm": 0.30336976051330566,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.03544272854924202,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.009017270989716053,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.03608,
+      "ppl": 1.00906,
       "step": 197,
       "tokens/total": 5954128,
-      "tokens/train_per_sec_per_gpu": 30.25,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 89988
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.11690016835927963,
+      "grad_norm": 0.530124306678772,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0010152912000194192,
-      "memory/device_reserved (GiB)": 35.57,
+      "loss": 0.013890329748392105,
+      "memory/device_reserved (GiB)": 35.56,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00102,
+      "ppl": 1.01399,
       "step": 198,
       "tokens/total": 5984352,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 90434
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.5135827660560608,
+      "grad_norm": 0.024880079552531242,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.009797877632081509,
-      "memory/device_reserved (GiB)": 35.61,
+      "loss": 0.00029932294273748994,
+      "memory/device_reserved (GiB)": 35.6,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00985,
+      "ppl": 1.0003,
       "step": 199,
       "tokens/total": 6014704,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 90913
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.26704609394073486,
+      "grad_norm": 0.07654840499162674,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.009519001469016075,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0013769213110208511,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00956,
+      "ppl": 1.00138,
       "step": 200,
       "tokens/total": 6045056,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 91355
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.3149840235710144,
+      "grad_norm": 0.08619414269924164,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.006803824566304684,
-      "memory/device_reserved (GiB)": 35.67,
+      "loss": 0.0012924536131322384,
+      "memory/device_reserved (GiB)": 35.66,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00683,
+      "ppl": 1.00129,
       "step": 201,
       "tokens/total": 6075504,
       "tokens/train_per_sec_per_gpu": 37.38,
@@ -2825,433 +2825,433 @@
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.5193817615509033,
+      "grad_norm": 0.0749412402510643,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.01742384023964405,
+      "loss": 0.0007854659343138337,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01758,
+      "ppl": 1.00079,
       "step": 202,
       "tokens/total": 6105904,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 92311
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.12959794700145721,
+      "grad_norm": 0.12518921494483948,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.0025574658066034317,
+      "loss": 0.0026314458809792995,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00256,
+      "ppl": 1.00263,
       "step": 203,
       "tokens/total": 6136272,
-      "tokens/train_per_sec_per_gpu": 34.55,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 92764
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.17874683439731598,
+      "grad_norm": 0.14211589097976685,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.004190261010080576,
+      "loss": 0.00402654055505991,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.0042,
+      "ppl": 1.00403,
       "step": 204,
       "tokens/total": 6166336,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 93227
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.38072845339775085,
+      "grad_norm": 0.9564501047134399,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.010198371484875679,
+      "loss": 0.01280949730426073,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01025,
+      "ppl": 1.01289,
       "step": 205,
       "tokens/total": 6196720,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 93675
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.39566439390182495,
+      "grad_norm": 0.13096395134925842,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.008401261642575264,
+      "loss": 0.0037373730447143316,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00844,
+      "ppl": 1.00374,
       "step": 206,
       "tokens/total": 6227424,
-      "tokens/train_per_sec_per_gpu": 32.33,
+      "tokens/train_per_sec_per_gpu": 32.3,
       "tokens/trainable": 94120
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.12372284382581711,
+      "grad_norm": 0.08379670232534409,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.002036123536527157,
+      "loss": 0.0017476077191531658,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00204,
+      "ppl": 1.00175,
       "step": 207,
       "tokens/total": 6257616,
-      "tokens/train_per_sec_per_gpu": 35.21,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 94557
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.20433077216148376,
+      "grad_norm": 0.05349349230527878,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.010479221120476723,
+      "loss": 0.0018553336849436164,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01053,
+      "ppl": 1.00186,
       "step": 208,
       "tokens/total": 6288000,
-      "tokens/train_per_sec_per_gpu": 40.52,
+      "tokens/train_per_sec_per_gpu": 40.66,
       "tokens/trainable": 95035
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.06323215365409851,
+      "grad_norm": 0.11009859293699265,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0014629911165684462,
+      "loss": 0.0024754812475293875,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00146,
+      "ppl": 1.00248,
       "step": 209,
       "tokens/total": 6318336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 95517
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.17430178821086884,
+      "grad_norm": 0.08477463573217392,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.007305104751139879,
+      "loss": 0.0019841620232909918,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00733,
+      "ppl": 1.00199,
       "step": 210,
       "tokens/total": 6348656,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 96026
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.08656168729066849,
+      "grad_norm": 0.06025635451078415,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.002277363557368517,
+      "loss": 0.0011887844884768128,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00228,
+      "ppl": 1.00119,
       "step": 211,
       "tokens/total": 6379040,
-      "tokens/train_per_sec_per_gpu": 35.44,
+      "tokens/train_per_sec_per_gpu": 35.64,
       "tokens/trainable": 96477
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.10843207687139511,
+      "grad_norm": 0.07387597858905792,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.0033393804915249348,
+      "loss": 0.001786265056580305,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00334,
+      "ppl": 1.00179,
       "step": 212,
       "tokens/total": 6409312,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 96951
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.20112648606300354,
+      "grad_norm": 0.1013425812125206,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.00444969953969121,
+      "loss": 0.0019797745626419783,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00446,
+      "ppl": 1.00198,
       "step": 213,
       "tokens/total": 6439520,
-      "tokens/train_per_sec_per_gpu": 30.68,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 97350
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.0895208865404129,
+      "grad_norm": 0.009712542407214642,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0011141544673591852,
+      "loss": 0.00013215326180215925,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00111,
+      "ppl": 1.00013,
       "step": 214,
       "tokens/total": 6469888,
-      "tokens/train_per_sec_per_gpu": 29.79,
+      "tokens/train_per_sec_per_gpu": 29.91,
       "tokens/trainable": 97792
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2406485378742218,
+      "grad_norm": 0.14567089080810547,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.0024814538192003965,
+      "loss": 0.0015704174293205142,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00248,
+      "ppl": 1.00157,
       "step": 215,
       "tokens/total": 6500128,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 98257
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.21981537342071533,
+      "grad_norm": 0.021219903603196144,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0057976399548351765,
+      "loss": 0.0003704531991388649,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00581,
+      "ppl": 1.00037,
       "step": 216,
       "tokens/total": 6530832,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 98731
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.020311880856752396,
+      "grad_norm": 0.22449812293052673,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0002352493756916374,
+      "loss": 0.0037617988418787718,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00024,
+      "ppl": 1.00377,
       "step": 217,
       "tokens/total": 6561248,
-      "tokens/train_per_sec_per_gpu": 36.37,
+      "tokens/train_per_sec_per_gpu": 36.5,
       "tokens/trainable": 99212
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.047305941581726074,
+      "grad_norm": 0.6340874433517456,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.000759766495320946,
+      "loss": 0.0011263209162279963,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00113,
       "step": 218,
       "tokens/total": 6591648,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 99654
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.3724987208843231,
+      "grad_norm": 0.1624881774187088,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0064449617639184,
+      "loss": 0.0014779233606532216,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00647,
+      "ppl": 1.00148,
       "step": 219,
       "tokens/total": 6622368,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.86,
       "tokens/trainable": 100086
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.044390205293893814,
+      "grad_norm": 0.01793455332517624,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0006100431783124804,
+      "loss": 0.0001897630572784692,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00061,
+      "ppl": 1.00019,
       "step": 220,
       "tokens/total": 6652848,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 100524
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.42042797803878784,
+      "grad_norm": 0.12492946535348892,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.010451005771756172,
+      "loss": 0.000895547098480165,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01051,
+      "ppl": 1.0009,
       "step": 221,
       "tokens/total": 6683392,
-      "tokens/train_per_sec_per_gpu": 33.27,
+      "tokens/train_per_sec_per_gpu": 33.46,
       "tokens/trainable": 100996
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.039993204176425934,
+      "grad_norm": 0.015087602660059929,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0005216938443481922,
+      "loss": 0.0001049312122631818,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00052,
+      "ppl": 1.0001,
       "step": 222,
       "tokens/total": 6713712,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 101449
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.4914291501045227,
+      "grad_norm": 0.713071882724762,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.01120755635201931,
+      "loss": 0.01887362264096737,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01127,
+      "ppl": 1.01905,
       "step": 223,
       "tokens/total": 6744176,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.14,
       "tokens/trainable": 101905
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.1168161928653717,
+      "grad_norm": 0.008568123914301395,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.0011214457917958498,
+      "loss": 7.857779564801604e-05,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00112,
+      "ppl": 1.00008,
       "step": 224,
       "tokens/total": 6774416,
-      "tokens/train_per_sec_per_gpu": 36.36,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 102411
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.08470873534679413,
+      "grad_norm": 0.6806920766830444,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.0012768175220116973,
+      "loss": 0.010111674666404724,
       "memory/device_reserved (GiB)": 37.19,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00128,
+      "ppl": 1.01016,
       "step": 225,
       "tokens/total": 6804592,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 102882
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.039867568761110306,
+      "grad_norm": 0.00420374283567071,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.0003806123568210751,
+      "loss": 5.371138468035497e-05,
       "memory/device_reserved (GiB)": 35.49,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00038,
+      "ppl": 1.00005,
       "step": 226,
       "tokens/total": 6834976,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.69,
       "tokens/trainable": 103332
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.4414898157119751,
+      "grad_norm": 0.02610113099217415,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.008082674816250801,
+      "loss": 0.00038069591391831636,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00812,
+      "ppl": 1.00038,
       "step": 227,
       "tokens/total": 6865408,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 103790
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.2599055767059326,
+      "grad_norm": 0.0028072672430425882,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.004410556983202696,
+      "loss": 6.497368303826079e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00442,
+      "ppl": 1.00006,
       "step": 228,
       "tokens/total": 6895664,
-      "tokens/train_per_sec_per_gpu": 34.35,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 104246
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.518774151802063,
+      "grad_norm": 0.8277482390403748,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.018912211060523987,
+      "loss": 0.013669933192431927,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01909,
+      "ppl": 1.01376,
       "step": 229,
       "tokens/total": 6926016,
-      "tokens/train_per_sec_per_gpu": 37.66,
+      "tokens/train_per_sec_per_gpu": 37.72,
       "tokens/trainable": 104707
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.25264421105384827,
+      "grad_norm": 0.08310598134994507,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 0.006381561979651451,
+      "loss": 0.0005468585877679288,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0064,
+      "ppl": 1.00055,
       "step": 230,
       "tokens/total": 6956384,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 105167
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.04987993463873863,
+      "grad_norm": 0.008127766661345959,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.0007956874324008822,
+      "loss": 0.00013037689495831728,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0008,
+      "ppl": 1.00013,
       "step": 231,
       "tokens/total": 6986768,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 105576
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.4156399965286255,
+      "grad_norm": 0.03313232958316803,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 0.0147963035851717,
+      "loss": 0.000351642636815086,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01491,
+      "ppl": 1.00035,
       "step": 232,
       "tokens/total": 7017232,
       "tokens/train_per_sec_per_gpu": 36.76,
@@ -3259,251 +3259,251 @@
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.1599927842617035,
+      "grad_norm": 0.08690419793128967,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 0.0033814553171396255,
+      "loss": 0.001037480542436242,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00339,
+      "ppl": 1.00104,
       "step": 233,
       "tokens/total": 7047568,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 106506
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.0438290536403656,
+      "grad_norm": 0.12895406782627106,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0007064358796924353,
+      "loss": 0.0019432969857007265,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00071,
+      "ppl": 1.00195,
       "step": 234,
       "tokens/total": 7077760,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 106951
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.13300912082195282,
+      "grad_norm": 0.25051259994506836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 0.003099396824836731,
+      "loss": 0.0025090454146265984,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0031,
+      "ppl": 1.00251,
       "step": 235,
       "tokens/total": 7107760,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.73,
       "tokens/trainable": 107382
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.3202158510684967,
+      "grad_norm": 0.5061792731285095,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 0.0032203267328441143,
+      "loss": 0.00891919620335102,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00323,
+      "ppl": 1.00896,
       "step": 236,
       "tokens/total": 7138144,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 107814
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.15484245121479034,
+      "grad_norm": 0.3209003210067749,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.002883841749280691,
+      "loss": 0.0041964175179600716,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00289,
+      "ppl": 1.00421,
       "step": 237,
       "tokens/total": 7168352,
-      "tokens/train_per_sec_per_gpu": 35.37,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 108274
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.09864962846040726,
+      "grad_norm": 0.3212501108646393,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.0021360500250011683,
+      "loss": 0.0037440985906869173,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00214,
+      "ppl": 1.00375,
       "step": 238,
       "tokens/total": 7198624,
-      "tokens/train_per_sec_per_gpu": 29.22,
+      "tokens/train_per_sec_per_gpu": 29.09,
       "tokens/trainable": 108714
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.06800950318574905,
+      "grad_norm": 0.021965481340885162,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0014171022921800613,
+      "loss": 0.00035154391662217677,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00142,
+      "ppl": 1.00035,
       "step": 239,
       "tokens/total": 7229216,
-      "tokens/train_per_sec_per_gpu": 36.79,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 109220
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.21696041524410248,
+      "grad_norm": 0.1164102703332901,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.002807284239679575,
+      "loss": 0.0012684958055615425,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00281,
+      "ppl": 1.00127,
       "step": 240,
       "tokens/total": 7259648,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.0,
       "tokens/trainable": 109672
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.43646690249443054,
+      "grad_norm": 0.17972798645496368,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 0.017607467249035835,
+      "loss": 0.0024644152726978064,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01776,
+      "ppl": 1.00247,
       "step": 241,
       "tokens/total": 7289824,
-      "tokens/train_per_sec_per_gpu": 35.18,
+      "tokens/train_per_sec_per_gpu": 35.09,
       "tokens/trainable": 110135
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.06252250075340271,
+      "grad_norm": 0.5711016654968262,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.0006150953122414649,
+      "loss": 0.01326853595674038,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00062,
+      "ppl": 1.01336,
       "step": 242,
       "tokens/total": 7320288,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 110563
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.16029377281665802,
+      "grad_norm": 0.7820162177085876,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0031884105410426855,
+      "loss": 0.01847490295767784,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00319,
+      "ppl": 1.01865,
       "step": 243,
       "tokens/total": 7350736,
-      "tokens/train_per_sec_per_gpu": 37.12,
+      "tokens/train_per_sec_per_gpu": 36.84,
       "tokens/trainable": 111049
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.17883111536502838,
+      "grad_norm": 0.2531258165836334,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0007043592631816864,
+      "loss": 0.003868672763928771,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0007,
+      "ppl": 1.00388,
       "step": 244,
       "tokens/total": 7381280,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 111495
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.052345480769872665,
+      "grad_norm": 0.09388009458780289,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0006958417361602187,
+      "loss": 0.0010953794699162245,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0007,
+      "ppl": 1.0011,
       "step": 245,
       "tokens/total": 7411632,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.68,
       "tokens/trainable": 111968
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.17381155490875244,
+      "grad_norm": 0.06202390044927597,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 0.008788044564425945,
+      "loss": 0.0006574424332939088,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00883,
+      "ppl": 1.00066,
       "step": 246,
       "tokens/total": 7441904,
-      "tokens/train_per_sec_per_gpu": 36.55,
+      "tokens/train_per_sec_per_gpu": 36.44,
       "tokens/trainable": 112416
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.10432726889848709,
+      "grad_norm": 0.21640881896018982,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0013522123917937279,
+      "loss": 0.00276574632152915,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00135,
+      "ppl": 1.00277,
       "step": 247,
       "tokens/total": 7472368,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 112882
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.07330253720283508,
+      "grad_norm": 0.04909277334809303,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 0.0010365882189944386,
+      "loss": 0.0012688931310549378,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00104,
+      "ppl": 1.00127,
       "step": 248,
       "tokens/total": 7502656,
-      "tokens/train_per_sec_per_gpu": 40.07,
+      "tokens/train_per_sec_per_gpu": 40.02,
       "tokens/trainable": 113390
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.7874143123626709,
+      "grad_norm": 0.2183118760585785,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 0.010244790464639664,
+      "loss": 0.01150418072938919,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0103,
+      "ppl": 1.01157,
       "step": 249,
       "tokens/total": 7532800,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 113904
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.04100371524691582,
+      "grad_norm": 0.05114463344216347,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.000650806468911469,
+      "loss": 0.0009117473382502794,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00065,
+      "ppl": 1.00091,
       "step": 250,
       "tokens/total": 7563264,
       "tokens/train_per_sec_per_gpu": 33.91,
@@ -3511,545 +3511,545 @@
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.14704902470111847,
+      "grad_norm": 0.10292479395866394,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.0022576111368834972,
+      "loss": 0.0016239782562479377,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00226,
+      "ppl": 1.00163,
       "step": 251,
       "tokens/total": 7593840,
-      "tokens/train_per_sec_per_gpu": 35.26,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 114842
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.2090071737766266,
+      "grad_norm": 0.057799480855464935,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.0027604042552411556,
+      "loss": 0.0011053154012188315,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00276,
+      "ppl": 1.00111,
       "step": 252,
       "tokens/total": 7624416,
-      "tokens/train_per_sec_per_gpu": 28.92,
+      "tokens/train_per_sec_per_gpu": 28.88,
       "tokens/trainable": 115263
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.11346573382616043,
+      "grad_norm": 0.017502324655652046,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 0.0015118042938411236,
+      "loss": 0.0003236275806557387,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00151,
+      "ppl": 1.00032,
       "step": 253,
       "tokens/total": 7654688,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.66,
       "tokens/trainable": 115703
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.20446987450122833,
+      "grad_norm": 0.2686062753200531,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.002259923843666911,
+      "loss": 0.003511242102831602,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00226,
+      "ppl": 1.00352,
       "step": 254,
       "tokens/total": 7685088,
-      "tokens/train_per_sec_per_gpu": 33.31,
+      "tokens/train_per_sec_per_gpu": 33.26,
       "tokens/trainable": 116157
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.6619936227798462,
+      "grad_norm": 0.25028568506240845,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.012125558219850063,
+      "loss": 0.0028144530951976776,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0122,
+      "ppl": 1.00282,
       "step": 255,
       "tokens/total": 7715552,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 31.95,
       "tokens/trainable": 116567
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.017078718170523643,
+      "grad_norm": 0.010626083239912987,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 0.00028051040135324,
+      "loss": 0.0002322449436178431,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00028,
+      "ppl": 1.00023,
       "step": 256,
       "tokens/total": 7745872,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.11,
       "tokens/trainable": 117030
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.007685024291276932,
+      "grad_norm": 0.004322522785514593,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 0.00016534165479242802,
+      "loss": 8.972767682280391e-05,
       "memory/device_reserved (GiB)": 38.23,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 257,
       "tokens/total": 7776208,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 117517
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.046725187450647354,
+      "grad_norm": 0.04855626821517944,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 0.0004824839415960014,
+      "loss": 0.0010425080545246601,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00048,
+      "ppl": 1.00104,
       "step": 258,
       "tokens/total": 7806480,
-      "tokens/train_per_sec_per_gpu": 38.44,
+      "tokens/train_per_sec_per_gpu": 38.33,
       "tokens/trainable": 118015
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.093961201608181,
+      "grad_norm": 0.015271801501512527,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 0.00111109868157655,
+      "loss": 0.0002960565616376698,
       "memory/device_reserved (GiB)": 35.62,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00111,
+      "ppl": 1.0003,
       "step": 259,
       "tokens/total": 7836800,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 118520
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.024271946400403976,
+      "grad_norm": 0.1238817349076271,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 0.00036734913010150194,
+      "loss": 0.001452557509765029,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00037,
+      "ppl": 1.00145,
       "step": 260,
       "tokens/total": 7867040,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 118992
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.10853405296802521,
+      "grad_norm": 0.008926448412239552,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.002126566832885146,
+      "loss": 0.0002352800511289388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00213,
+      "ppl": 1.00024,
       "step": 261,
       "tokens/total": 7897408,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 119438
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.1838080883026123,
+      "grad_norm": 0.0867115706205368,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0012134769931435585,
+      "loss": 0.0014156652614474297,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00121,
+      "ppl": 1.00142,
       "step": 262,
       "tokens/total": 7927648,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 119881
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.28273531794548035,
+      "grad_norm": 0.03644087538123131,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0020987153984606266,
+      "loss": 0.0004944024258293211,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0021,
+      "ppl": 1.00049,
       "step": 263,
       "tokens/total": 7957968,
-      "tokens/train_per_sec_per_gpu": 27.91,
+      "tokens/train_per_sec_per_gpu": 27.72,
       "tokens/trainable": 120285
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.07407250255346298,
+      "grad_norm": 0.4136442542076111,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 0.00043982663191854954,
+      "loss": 0.005234354641288519,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.65,
       "memory/max_allocated (GiB)": 33.65,
-      "ppl": 1.00044,
+      "ppl": 1.00525,
       "step": 264,
       "tokens/total": 7987888,
-      "tokens/train_per_sec_per_gpu": 33.2,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 120755
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.01878584362566471,
+      "grad_norm": 0.011108173988759518,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 0.00014881066454108804,
+      "loss": 0.00019686836458276957,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 265,
       "tokens/total": 8018048,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 121194
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.012090266682207584,
+      "grad_norm": 0.2199329286813736,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 0.00013956695329397917,
+      "loss": 0.005357124377042055,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00014,
+      "ppl": 1.00537,
       "step": 266,
       "tokens/total": 8048448,
-      "tokens/train_per_sec_per_gpu": 36.68,
+      "tokens/train_per_sec_per_gpu": 36.54,
       "tokens/trainable": 121670
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.03192078694701195,
+      "grad_norm": 0.15096357464790344,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000175558976479806,
+      "loss": 0.0024120814632624388,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00018,
+      "ppl": 1.00241,
       "step": 267,
       "tokens/total": 8078848,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.58,
       "tokens/trainable": 122142
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.008871566504240036,
+      "grad_norm": 0.0024968513753265142,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00010361030581407249,
+      "loss": 4.619035462383181e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00005,
       "step": 268,
       "tokens/total": 8109360,
-      "tokens/train_per_sec_per_gpu": 32.41,
+      "tokens/train_per_sec_per_gpu": 32.36,
       "tokens/trainable": 122585
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.37323296070098877,
+      "grad_norm": 0.018697405233979225,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 0.0028822675812989473,
+      "loss": 0.00024445558665320277,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00289,
+      "ppl": 1.00024,
       "step": 269,
       "tokens/total": 8139888,
-      "tokens/train_per_sec_per_gpu": 38.85,
+      "tokens/train_per_sec_per_gpu": 38.78,
       "tokens/trainable": 123091
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.019474836066365242,
+      "grad_norm": 0.04383962228894234,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 0.0003204788372386247,
+      "loss": 0.0005069951876066625,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00032,
+      "ppl": 1.00051,
       "step": 270,
       "tokens/total": 8170480,
-      "tokens/train_per_sec_per_gpu": 30.9,
+      "tokens/train_per_sec_per_gpu": 30.91,
       "tokens/trainable": 123505
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.05259509012103081,
+      "grad_norm": 0.018031178042292595,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 0.00033461389830335975,
+      "loss": 0.00028141128132119775,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.41,
       "memory/max_allocated (GiB)": 33.41,
-      "ppl": 1.00033,
+      "ppl": 1.00028,
       "step": 271,
       "tokens/total": 8198848,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.31,
       "tokens/trainable": 123953
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.09935376048088074,
+      "grad_norm": 0.016046874225139618,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 0.00043552459101192653,
+      "loss": 0.00015048845671117306,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00044,
+      "ppl": 1.00015,
       "step": 272,
       "tokens/total": 8229200,
-      "tokens/train_per_sec_per_gpu": 28.66,
+      "tokens/train_per_sec_per_gpu": 28.61,
       "tokens/trainable": 124400
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.3927276134490967,
+      "grad_norm": 0.004127623979002237,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.00702043017372489,
+      "loss": 7.019042095635086e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00705,
+      "ppl": 1.00007,
       "step": 273,
       "tokens/total": 8259536,
-      "tokens/train_per_sec_per_gpu": 35.05,
+      "tokens/train_per_sec_per_gpu": 35.1,
       "tokens/trainable": 124870
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.03849954903125763,
+      "grad_norm": 0.016368450596928596,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 0.0003886982740368694,
+      "loss": 0.00023265022900886834,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00039,
+      "ppl": 1.00023,
       "step": 274,
       "tokens/total": 8289904,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 125333
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.010367084294557571,
+      "grad_norm": 0.0009975603315979242,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 8.032341429498047e-05,
+      "loss": 2.3090822651283816e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 275,
       "tokens/total": 8320176,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 125834
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.05538506433367729,
+      "grad_norm": 0.0020015796180814505,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 0.00026508988230489194,
+      "loss": 2.2906289814272895e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00027,
+      "ppl": 1.00002,
       "step": 276,
       "tokens/total": 8350832,
-      "tokens/train_per_sec_per_gpu": 37.97,
+      "tokens/train_per_sec_per_gpu": 38.02,
       "tokens/trainable": 126316
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.20107394456863403,
+      "grad_norm": 0.005128095392137766,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 0.0007178307860158384,
+      "loss": 6.587664393009618e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00072,
+      "ppl": 1.00007,
       "step": 277,
       "tokens/total": 8380976,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.33,
       "tokens/trainable": 126798
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.005950715858489275,
+      "grad_norm": 0.0013011472765356302,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 4.905788227915764e-05,
+      "loss": 2.66208026005188e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 278,
       "tokens/total": 8411072,
-      "tokens/train_per_sec_per_gpu": 29.57,
+      "tokens/train_per_sec_per_gpu": 29.64,
       "tokens/trainable": 127223
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.009278975427150726,
+      "grad_norm": 0.001754116965457797,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 8.996425458462909e-05,
+      "loss": 2.587787457741797e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00003,
       "step": 279,
       "tokens/total": 8440736,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 127672
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.11463552713394165,
+      "grad_norm": 0.005082705058157444,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 0.0006871019722893834,
+      "loss": 8.052532211877406e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00069,
+      "ppl": 1.00008,
       "step": 280,
       "tokens/total": 8470944,
-      "tokens/train_per_sec_per_gpu": 32.07,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 128116
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.004335940349847078,
+      "grad_norm": 0.0014385804533958435,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 4.704743332695216e-05,
+      "loss": 2.603003304102458e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 281,
       "tokens/total": 8501104,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.82,
       "tokens/trainable": 128589
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.21418413519859314,
+      "grad_norm": 0.7784804105758667,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.01315401028841734,
+      "loss": 0.008352375589311123,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01324,
+      "ppl": 1.00839,
       "step": 282,
       "tokens/total": 8531696,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 129036
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0006189382984302938,
+      "grad_norm": 0.0021191469859331846,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 1.5588291716994718e-05,
+      "loss": 3.5222510632593185e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 283,
       "tokens/total": 8561936,
-      "tokens/train_per_sec_per_gpu": 35.16,
+      "tokens/train_per_sec_per_gpu": 35.21,
       "tokens/trainable": 129485
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.005335172638297081,
+      "grad_norm": 0.019356347620487213,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 8.037629595492035e-05,
+      "loss": 0.0002056795492535457,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00021,
       "step": 284,
       "tokens/total": 8592192,
-      "tokens/train_per_sec_per_gpu": 29.49,
+      "tokens/train_per_sec_per_gpu": 29.55,
       "tokens/trainable": 129920
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.003026328282430768,
+      "grad_norm": 0.030793415382504463,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 4.424918006407097e-05,
+      "loss": 0.00022263142454903573,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00022,
       "step": 285,
       "tokens/total": 8622432,
-      "tokens/train_per_sec_per_gpu": 36.06,
+      "tokens/train_per_sec_per_gpu": 36.14,
       "tokens/trainable": 130417
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.0032119054812937975,
+      "grad_norm": 0.2944176197052002,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 5.364553726394661e-05,
+      "loss": 0.0028715431690216064,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00005,
+      "ppl": 1.00288,
       "step": 286,
       "tokens/total": 8652768,
-      "tokens/train_per_sec_per_gpu": 35.89,
+      "tokens/train_per_sec_per_gpu": 35.95,
       "tokens/trainable": 130903
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.015378961339592934,
+      "grad_norm": 0.0015030609210953116,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 9.19914455153048e-05,
+      "loss": 2.242590744572226e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 287,
       "tokens/total": 8683296,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 131343
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.8838728070259094,
+      "grad_norm": 0.22225140035152435,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 0.00693545350804925,
+      "loss": 0.0026556546799838543,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00696,
+      "ppl": 1.00266,
       "step": 288,
       "tokens/total": 8713504,
-      "tokens/train_per_sec_per_gpu": 32.19,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 131763
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.0019988964777439833,
+      "grad_norm": 0.018996328115463257,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 2.3260268790181726e-05,
+      "loss": 4.354536213213578e-05,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 289,
       "tokens/total": 8743888,
       "tokens/train_per_sec_per_gpu": 34.0,
@@ -4057,461 +4057,461 @@
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.0004976001800969243,
+      "grad_norm": 0.001890109502710402,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 1.430663360224571e-05,
+      "loss": 2.4500381186953746e-05,
       "memory/device_reserved (GiB)": 35.54,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 290,
       "tokens/total": 8774336,
-      "tokens/train_per_sec_per_gpu": 35.46,
+      "tokens/train_per_sec_per_gpu": 35.19,
       "tokens/trainable": 132678
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.01294003613293171,
+      "grad_norm": 0.0009667908307164907,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0001438881445210427,
+      "loss": 2.0667655917350203e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00014,
+      "ppl": 1.00002,
       "step": 291,
       "tokens/total": 8804816,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.28,
       "tokens/trainable": 133136
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.3451043963432312,
+      "grad_norm": 0.0010472588473930955,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 0.0030744036193937063,
+      "loss": 2.282073546666652e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00308,
+      "ppl": 1.00002,
       "step": 292,
       "tokens/total": 8835024,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 133608
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.024186862632632256,
+      "grad_norm": 0.028602443635463715,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 0.000165810008184053,
+      "loss": 0.0002425020356895402,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00017,
+      "ppl": 1.00024,
       "step": 293,
       "tokens/total": 8865408,
-      "tokens/train_per_sec_per_gpu": 40.44,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 134108
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.001723558409139514,
+      "grad_norm": 0.0024220976047217846,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 3.2396514143329114e-05,
+      "loss": 3.14589160552714e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00003,
       "step": 294,
       "tokens/total": 8895824,
-      "tokens/train_per_sec_per_gpu": 30.72,
+      "tokens/train_per_sec_per_gpu": 30.68,
       "tokens/trainable": 134541
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.010282398201525211,
+      "grad_norm": 0.00373630179092288,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 0.00010598616790957749,
+      "loss": 5.82915308768861e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00011,
+      "ppl": 1.00006,
       "step": 295,
       "tokens/total": 8926336,
-      "tokens/train_per_sec_per_gpu": 29.51,
+      "tokens/train_per_sec_per_gpu": 29.44,
       "tokens/trainable": 134966
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.02922157198190689,
+      "grad_norm": 0.023777559399604797,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 0.0002625146880745888,
+      "loss": 0.0002222473849542439,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00026,
+      "ppl": 1.00022,
       "step": 296,
       "tokens/total": 8956656,
-      "tokens/train_per_sec_per_gpu": 33.56,
+      "tokens/train_per_sec_per_gpu": 33.53,
       "tokens/trainable": 135402
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.4516298770904541,
+      "grad_norm": 0.03583608567714691,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 0.004878990352153778,
+      "loss": 0.000519716995768249,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00489,
+      "ppl": 1.00052,
       "step": 297,
       "tokens/total": 8986752,
-      "tokens/train_per_sec_per_gpu": 30.05,
+      "tokens/train_per_sec_per_gpu": 30.02,
       "tokens/trainable": 135804
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.07809585332870483,
+      "grad_norm": 0.10825730860233307,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 0.00019118667114526033,
+      "loss": 0.0007585557177662849,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00019,
+      "ppl": 1.00076,
       "step": 298,
       "tokens/total": 9017024,
-      "tokens/train_per_sec_per_gpu": 31.19,
+      "tokens/train_per_sec_per_gpu": 31.13,
       "tokens/trainable": 136231
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.008364620618522167,
+      "grad_norm": 0.0007777873543091118,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 6.207433034433052e-05,
+      "loss": 1.649466503295116e-05,
       "memory/device_reserved (GiB)": 35.55,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 299,
       "tokens/total": 9047424,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 136705
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.35634350776672363,
+      "grad_norm": 0.0021820615511387587,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 0.008261370472609997,
+      "loss": 3.35803342750296e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0083,
+      "ppl": 1.00003,
       "step": 300,
       "tokens/total": 9077648,
-      "tokens/train_per_sec_per_gpu": 34.86,
+      "tokens/train_per_sec_per_gpu": 34.73,
       "tokens/trainable": 137178
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.1326446235179901,
+      "grad_norm": 0.017772037535905838,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 0.0006729009910486639,
+      "loss": 9.50043904595077e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0001,
       "step": 301,
       "tokens/total": 9108032,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 137643
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.11166927963495255,
+      "grad_norm": 0.08602973073720932,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 0.000889140646904707,
+      "loss": 0.0010838620364665985,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00089,
+      "ppl": 1.00108,
       "step": 302,
       "tokens/total": 9138240,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 138078
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.656753420829773,
+      "grad_norm": 0.0017863045213744044,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 0.004501559771597385,
+      "loss": 3.200750143150799e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00451,
+      "ppl": 1.00003,
       "step": 303,
       "tokens/total": 9168352,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.0,
       "tokens/trainable": 138515
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.00743053387850523,
+      "grad_norm": 0.0011891268659383059,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 9.650958236306906e-05,
+      "loss": 2.6663004973670468e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 304,
       "tokens/total": 9198864,
-      "tokens/train_per_sec_per_gpu": 34.99,
+      "tokens/train_per_sec_per_gpu": 35.13,
       "tokens/trainable": 138948
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.0033850902691483498,
+      "grad_norm": 0.08658935129642487,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 5.173611862119287e-05,
+      "loss": 0.0008552016224712133,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00086,
       "step": 305,
       "tokens/total": 9229104,
-      "tokens/train_per_sec_per_gpu": 38.17,
+      "tokens/train_per_sec_per_gpu": 38.05,
       "tokens/trainable": 139408
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.0017230098601430655,
+      "grad_norm": 0.0014755144948139787,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 3.198062040610239e-05,
+      "loss": 2.39577166212257e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 306,
       "tokens/total": 9257344,
-      "tokens/train_per_sec_per_gpu": 33.48,
+      "tokens/train_per_sec_per_gpu": 33.38,
       "tokens/trainable": 139840
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.06396278738975525,
+      "grad_norm": 0.014350412413477898,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 0.0004305330221541226,
+      "loss": 0.00012315387721173465,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00043,
+      "ppl": 1.00012,
       "step": 307,
       "tokens/total": 9287520,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 140281
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.1433790922164917,
+      "grad_norm": 0.0019020310137420893,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 0.0006829933845438063,
+      "loss": 2.20383644773392e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00068,
+      "ppl": 1.00002,
       "step": 308,
       "tokens/total": 9317888,
-      "tokens/train_per_sec_per_gpu": 37.44,
+      "tokens/train_per_sec_per_gpu": 37.31,
       "tokens/trainable": 140776
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.3920465111732483,
+      "grad_norm": 0.0015318727819249034,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 0.005061979405581951,
+      "loss": 2.580175168986898e-05,
       "memory/device_reserved (GiB)": 35.72,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00507,
+      "ppl": 1.00003,
       "step": 309,
       "tokens/total": 9348256,
-      "tokens/train_per_sec_per_gpu": 36.89,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 141228
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.016641858965158463,
+      "grad_norm": 0.00613615894690156,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 0.0001797095756046474,
+      "loss": 6.0944184951949865e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00018,
+      "ppl": 1.00006,
       "step": 310,
       "tokens/total": 9378560,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.19,
       "tokens/trainable": 141679
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.0008492676424793899,
+      "grad_norm": 0.001598753617145121,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 1.9429104213486426e-05,
+      "loss": 2.0053470507264137e-05,
       "memory/device_reserved (GiB)": 35.74,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9408848,
-      "tokens/train_per_sec_per_gpu": 32.27,
+      "tokens/train_per_sec_per_gpu": 32.25,
       "tokens/trainable": 142122
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.04300769418478012,
+      "grad_norm": 0.0007549300789833069,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 0.0005121674039401114,
+      "loss": 1.628213794901967e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00051,
+      "ppl": 1.00002,
       "step": 312,
       "tokens/total": 9439344,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.01,
       "tokens/trainable": 142561
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.023457281291484833,
+      "grad_norm": 0.00085266592213884,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 0.000151450076373294,
+      "loss": 1.5117442671908066e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00015,
+      "ppl": 1.00002,
       "step": 313,
       "tokens/total": 9469792,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.55,
       "tokens/trainable": 143046
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.06128578260540962,
+      "grad_norm": 0.02262088656425476,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.0006227570120245218,
+      "loss": 0.00013078594929538667,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00062,
+      "ppl": 1.00013,
       "step": 314,
       "tokens/total": 9500128,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 143493
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.007639073766767979,
+      "grad_norm": 0.00101909798104316,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 7.468041440006346e-05,
+      "loss": 1.9173825421603397e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 315,
       "tokens/total": 9530480,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 143956
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.35354724526405334,
+      "grad_norm": 0.023366861045360565,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.004370104521512985,
+      "loss": 0.0001811327674658969,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00438,
+      "ppl": 1.00018,
       "step": 316,
       "tokens/total": 9560912,
-      "tokens/train_per_sec_per_gpu": 33.93,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 144411
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.5020444989204407,
+      "grad_norm": 0.011399022303521633,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 0.010973826982080936,
+      "loss": 0.00012808202882297337,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01103,
+      "ppl": 1.00013,
       "step": 317,
       "tokens/total": 9591344,
-      "tokens/train_per_sec_per_gpu": 38.48,
+      "tokens/train_per_sec_per_gpu": 38.66,
       "tokens/trainable": 144892
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0027349034789949656,
+      "grad_norm": 0.5837145447731018,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 4.4591506593860686e-05,
+      "loss": 0.006938215810805559,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00004,
+      "ppl": 1.00696,
       "step": 318,
       "tokens/total": 9621392,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 145339
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 0.013429106213152409,
+      "grad_norm": 0.0007919945055618882,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 8.016972424229607e-05,
+      "loss": 1.6890848201001063e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 319,
       "tokens/total": 9651728,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.38,
       "tokens/trainable": 145811
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.02254675142467022,
+      "grad_norm": 0.025993503630161285,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 0.00017163707525469363,
+      "loss": 0.0002537990512792021,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00017,
+      "ppl": 1.00025,
       "step": 320,
       "tokens/total": 9682224,
-      "tokens/train_per_sec_per_gpu": 38.76,
+      "tokens/train_per_sec_per_gpu": 38.62,
       "tokens/trainable": 146314
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.04359544813632965,
+      "grad_norm": 0.0010733718518167734,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 0.0005239051533862948,
+      "loss": 1.948333010659553e-05,
       "memory/device_reserved (GiB)": 37.95,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00052,
+      "ppl": 1.00002,
       "step": 321,
       "tokens/total": 9712656,
-      "tokens/train_per_sec_per_gpu": 31.85,
+      "tokens/train_per_sec_per_gpu": 31.72,
       "tokens/trainable": 146781
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.02377651259303093,
+      "grad_norm": 0.0843457579612732,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0001890905696200207,
+      "loss": 0.0008847219287417829,
       "memory/device_reserved (GiB)": 34.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00019,
+      "ppl": 1.00089,
       "step": 322,
       "tokens/total": 9743088,
       "tokens/train_per_sec_per_gpu": 35.55,
@@ -4519,69 +4519,69 @@
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.01256605889648199,
+      "grad_norm": 0.09092428535223007,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 8.730488480068743e-05,
+      "loss": 0.00040022452594712377,
       "memory/device_reserved (GiB)": 35.17,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00009,
+      "ppl": 1.0004,
       "step": 323,
       "tokens/total": 9773808,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 147673
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.21066401898860931,
+      "grad_norm": 0.3859696090221405,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.0028422519098967314,
+      "loss": 0.004064415581524372,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00285,
+      "ppl": 1.00407,
       "step": 324,
       "tokens/total": 9804016,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 148103
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 0.025614596903324127,
+      "grad_norm": 0.000418124720454216,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.00016133410099428147,
+      "loss": 9.266825145459734e-06,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00016,
+      "ppl": 1.00001,
       "step": 325,
       "tokens/total": 9834592,
-      "tokens/train_per_sec_per_gpu": 31.73,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 148535
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.033190563321113586,
+      "grad_norm": 0.030438628047704697,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 0.00033758440986275673,
+      "loss": 0.00029550789622589946,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00034,
+      "ppl": 1.0003,
       "step": 326,
       "tokens/total": 9864848,
-      "tokens/train_per_sec_per_gpu": 34.07,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 148999
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.07407072931528091,
+      "grad_norm": 0.0003460803418420255,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 0.0007672893116250634,
+      "loss": 9.424240488442592e-06,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00077,
+      "ppl": 1.00001,
       "step": 327,
       "tokens/total": 9895168,
       "tokens/train_per_sec_per_gpu": 31.64,
@@ -4589,125 +4589,125 @@
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.05904461070895195,
+      "grad_norm": 0.32631534337997437,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.000713472138158977,
+      "loss": 0.004381683189421892,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00071,
+      "ppl": 1.00439,
       "step": 328,
       "tokens/total": 9925696,
-      "tokens/train_per_sec_per_gpu": 34.11,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 149886
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.007630123756825924,
+      "grad_norm": 0.06805918365716934,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 7.993751933099702e-05,
+      "loss": 0.0003639076603576541,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00036,
       "step": 329,
       "tokens/total": 9955648,
-      "tokens/train_per_sec_per_gpu": 30.98,
+      "tokens/train_per_sec_per_gpu": 31.01,
       "tokens/trainable": 150288
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.024742672219872475,
+      "grad_norm": 0.004817479755729437,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 0.00019552679441403598,
+      "loss": 3.246869164286181e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 330,
       "tokens/total": 9985856,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 150733
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.006353198084980249,
+      "grad_norm": 0.06923647969961166,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.00010314649261999875,
+      "loss": 0.00031070224940776825,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 331,
       "tokens/total": 10016208,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 151207
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.16203969717025757,
+      "grad_norm": 0.14731979370117188,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.0014735229779034853,
+      "loss": 0.0015139597235247493,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00147,
+      "ppl": 1.00152,
       "step": 332,
       "tokens/total": 10046512,
-      "tokens/train_per_sec_per_gpu": 37.5,
+      "tokens/train_per_sec_per_gpu": 37.61,
       "tokens/trainable": 151694
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.041821569204330444,
+      "grad_norm": 0.007325666956603527,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.0004190094769001007,
+      "loss": 7.809747330611572e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00042,
+      "ppl": 1.00008,
       "step": 333,
       "tokens/total": 10076944,
-      "tokens/train_per_sec_per_gpu": 38.27,
+      "tokens/train_per_sec_per_gpu": 38.37,
       "tokens/trainable": 152188
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0011248595546931028,
+      "grad_norm": 0.00047597952652722597,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 3.350014958414249e-05,
+      "loss": 1.2739032172248699e-05,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 334,
       "tokens/total": 10107296,
-      "tokens/train_per_sec_per_gpu": 30.6,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 152596
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.003707638941705227,
+      "grad_norm": 0.0700320228934288,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 4.684936357080005e-05,
+      "loss": 0.00047467637341469526,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00047,
       "step": 335,
       "tokens/total": 10137568,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 153052
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.0017528374446555972,
+      "grad_norm": 0.06498395651578903,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 2.3632102966075763e-05,
+      "loss": 0.00041414948645979166,
       "memory/device_reserved (GiB)": 37.17,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00002,
+      "ppl": 1.00041,
       "step": 336,
       "tokens/total": 10167952,
       "tokens/train_per_sec_per_gpu": 30.19,
@@ -4715,1105 +4715,1105 @@
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.0626155436038971,
+      "grad_norm": 0.009991639293730259,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0006341143744066358,
+      "loss": 7.060338975861669e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00063,
+      "ppl": 1.00007,
       "step": 337,
       "tokens/total": 10198512,
-      "tokens/train_per_sec_per_gpu": 40.28,
+      "tokens/train_per_sec_per_gpu": 40.43,
       "tokens/trainable": 153983
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.0166204534471035,
+      "grad_norm": 0.03267490491271019,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.00015405623707920313,
+      "loss": 0.0002045792352873832,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 338,
       "tokens/total": 10228752,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.72,
       "tokens/trainable": 154464
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.0004508346610236913,
+      "grad_norm": 0.00039529221248812973,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.604627510067075e-06,
+      "loss": 8.338471161550842e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 339,
       "tokens/total": 10259104,
-      "tokens/train_per_sec_per_gpu": 35.73,
+      "tokens/train_per_sec_per_gpu": 35.87,
       "tokens/trainable": 154924
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.0024134982377290726,
+      "grad_norm": 0.0010571131715551019,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 4.25071848439984e-05,
+      "loss": 1.6815669368952513e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 340,
       "tokens/total": 10289520,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 155373
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.047796547412872314,
+      "grad_norm": 0.0008337291656062007,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.0004414983559399843,
+      "loss": 1.703310408629477e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00044,
+      "ppl": 1.00002,
       "step": 341,
       "tokens/total": 10320016,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 155873
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.0013580132508650422,
+      "grad_norm": 0.0007872325950302184,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 3.1043862691149116e-05,
+      "loss": 1.626565062906593e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 342,
       "tokens/total": 10350368,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.42,
       "tokens/trainable": 156278
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.013898961246013641,
+      "grad_norm": 0.21220935881137848,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 9.816634701564908e-05,
+      "loss": 0.0009314992348663509,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00093,
       "step": 343,
       "tokens/total": 10380624,
-      "tokens/train_per_sec_per_gpu": 35.31,
+      "tokens/train_per_sec_per_gpu": 35.4,
       "tokens/trainable": 156733
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.0031031679827719927,
+      "grad_norm": 0.017469989135861397,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 4.329531657276675e-05,
+      "loss": 7.496406033169478e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 344,
       "tokens/total": 10411024,
-      "tokens/train_per_sec_per_gpu": 33.76,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 157203
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.009028271771967411,
+      "grad_norm": 0.2605672776699066,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.00010301935981260613,
+      "loss": 0.004417422227561474,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00443,
       "step": 345,
       "tokens/total": 10441248,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 157707
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.0036287850234657526,
+      "grad_norm": 0.00032958327210508287,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 4.620348772732541e-05,
+      "loss": 7.68474092183169e-06,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00005,
+      "ppl": 1.00001,
       "step": 346,
       "tokens/total": 10471712,
-      "tokens/train_per_sec_per_gpu": 34.96,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 158180
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.08971801400184631,
+      "grad_norm": 0.0021204655058681965,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.0009501657332293689,
+      "loss": 2.9396429454209283e-05,
       "memory/device_reserved (GiB)": 38.13,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00095,
+      "ppl": 1.00003,
       "step": 347,
       "tokens/total": 10502288,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.41,
       "tokens/trainable": 158646
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.00890435092151165,
+      "grad_norm": 0.00040693042683415115,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 7.783617911627516e-05,
+      "loss": 9.5013465397642e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 348,
       "tokens/total": 10532304,
-      "tokens/train_per_sec_per_gpu": 35.59,
+      "tokens/train_per_sec_per_gpu": 35.66,
       "tokens/trainable": 159115
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.023319199681282043,
+      "grad_norm": 0.006379029247909784,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.00021630006085615605,
+      "loss": 9.018932178150862e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00022,
+      "ppl": 1.00009,
       "step": 349,
       "tokens/total": 10562608,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 159538
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.047880928963422775,
+      "grad_norm": 0.005555902142077684,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.00044884331873618066,
+      "loss": 5.1456365326885134e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00045,
+      "ppl": 1.00005,
       "step": 350,
       "tokens/total": 10592992,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.22,
       "tokens/trainable": 160016
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008666309528052807,
+      "grad_norm": 0.009832990355789661,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 5.017036892240867e-05,
+      "loss": 9.037736163008958e-05,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00005,
+      "ppl": 1.00009,
       "step": 351,
       "tokens/total": 10623328,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.57,
       "tokens/trainable": 160475
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06452610343694687,
+      "grad_norm": 0.0002588493807706982,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0007272367365658283,
+      "loss": 7.509744136768859e-06,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00073,
+      "ppl": 1.00001,
       "step": 352,
       "tokens/total": 10653632,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 160952
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.370597779750824,
+      "grad_norm": 0.03195308893918991,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 0.007550997193902731,
+      "loss": 0.0002666166110429913,
       "memory/device_reserved (GiB)": 38.4,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00758,
+      "ppl": 1.00027,
       "step": 353,
       "tokens/total": 10684176,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 161420
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0031376827973872423,
+      "grad_norm": 0.07012991607189178,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 2.3453332687495276e-05,
+      "loss": 0.000741704716347158,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00074,
       "step": 354,
       "tokens/total": 10714352,
-      "tokens/train_per_sec_per_gpu": 34.02,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 161868
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.0021399864926934242,
+      "grad_norm": 0.0005822654929943383,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 2.5832894607447088e-05,
+      "loss": 1.0378402294008993e-05,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 355,
       "tokens/total": 10744464,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 162318
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0028148347046226263,
+      "grad_norm": 0.012435230426490307,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 3.7286932638380677e-05,
+      "loss": 7.89838086348027e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00008,
       "step": 356,
       "tokens/total": 10774976,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.54,
       "tokens/trainable": 162789
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.1642119586467743,
+      "grad_norm": 0.0045598647557199,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 0.0027845175936818123,
+      "loss": 2.48450869548833e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00279,
+      "ppl": 1.00002,
       "step": 357,
       "tokens/total": 10805600,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.43,
       "tokens/trainable": 163254
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.066535085439682,
+      "grad_norm": 0.3044262230396271,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.0002488417667336762,
+      "loss": 0.002298796083778143,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00025,
+      "ppl": 1.0023,
       "step": 358,
       "tokens/total": 10835920,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.62,
       "tokens/trainable": 163715
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.006617655046284199,
+      "grad_norm": 0.002458421979099512,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 7.841112528694794e-05,
+      "loss": 1.9634167983895168e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00002,
       "step": 359,
       "tokens/total": 10866048,
-      "tokens/train_per_sec_per_gpu": 34.4,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 164197
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.010409035719931126,
+      "grad_norm": 0.0024773837067186832,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010489403939573094,
+      "loss": 2.744927041931078e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00003,
       "step": 360,
       "tokens/total": 10896288,
-      "tokens/train_per_sec_per_gpu": 31.6,
+      "tokens/train_per_sec_per_gpu": 31.74,
       "tokens/trainable": 164661
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.002426267135888338,
+      "grad_norm": 0.00026717092259787023,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 4.0343060391023755e-05,
+      "loss": 6.422977094189264e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 361,
       "tokens/total": 10926816,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 165111
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.015666797757148743,
+      "grad_norm": 0.0008353493758477271,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 0.00014991794887464494,
+      "loss": 1.3549893083109055e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 362,
       "tokens/total": 10954928,
-      "tokens/train_per_sec_per_gpu": 30.34,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 165552
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.2695651650428772,
+      "grad_norm": 0.006071125157177448,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0014983330620452762,
+      "loss": 6.428411870729178e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.0015,
+      "ppl": 1.00006,
       "step": 363,
       "tokens/total": 10985328,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 166026
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.03586212173104286,
+      "grad_norm": 0.25424960255622864,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.00030509717180393636,
+      "loss": 0.004243595991283655,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00031,
+      "ppl": 1.00425,
       "step": 364,
       "tokens/total": 11015664,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 38.11,
       "tokens/trainable": 166486
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.0019308789633214474,
+      "grad_norm": 0.0005575277027674019,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 2.4129443772835657e-05,
+      "loss": 8.403902938880492e-06,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 365,
       "tokens/total": 11046256,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 166959
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005940837785601616,
+      "grad_norm": 0.0014335239538922906,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 2.8238933737156913e-05,
+      "loss": 1.4738036043127067e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 366,
       "tokens/total": 11076288,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 167394
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.002125627128407359,
+      "grad_norm": 0.0010152696631848812,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 1.602789416210726e-05,
+      "loss": 1.3278609912958927e-05,
       "memory/device_reserved (GiB)": 36.01,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 367,
       "tokens/total": 11106416,
-      "tokens/train_per_sec_per_gpu": 28.5,
+      "tokens/train_per_sec_per_gpu": 28.58,
       "tokens/trainable": 167832
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.2510211765766144,
+      "grad_norm": 0.11941836029291153,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 0.00010395953722763807,
+      "loss": 0.0006476733833551407,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00065,
       "step": 368,
       "tokens/total": 11136848,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 31.02,
       "tokens/trainable": 168242
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.001904280623421073,
+      "grad_norm": 0.006877629552036524,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 2.6418363631819375e-05,
+      "loss": 7.449048280250281e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 369,
       "tokens/total": 11167376,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 168715
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.0028152521699666977,
+      "grad_norm": 0.0007151139434427023,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 2.9974533390486613e-05,
+      "loss": 1.1437590728746727e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 370,
       "tokens/total": 11197776,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.86,
       "tokens/trainable": 169147
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.0010822078911587596,
+      "grad_norm": 0.009094453416764736,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 1.34217716549756e-05,
+      "loss": 5.9242345741949975e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11227840,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 169628
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.005367911420762539,
+      "grad_norm": 0.000247267191298306,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 6.65646803099662e-05,
+      "loss": 7.548428584414069e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 372,
       "tokens/total": 11258176,
-      "tokens/train_per_sec_per_gpu": 35.11,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 170085
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.047570567578077316,
+      "grad_norm": 0.0003029497747775167,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.0003471036907285452,
+      "loss": 5.4013939916330855e-06,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00035,
+      "ppl": 1.00001,
       "step": 373,
       "tokens/total": 11288336,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.6,
       "tokens/trainable": 170584
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.040886107832193375,
+      "grad_norm": 0.001942179980687797,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.0002880148240365088,
+      "loss": 2.380511250521522e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00029,
+      "ppl": 1.00002,
       "step": 374,
       "tokens/total": 11318848,
-      "tokens/train_per_sec_per_gpu": 34.56,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 171058
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.0031216361094266176,
+      "grad_norm": 0.0019308892078697681,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 2.690855944820214e-05,
+      "loss": 1.5233906196954194e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 375,
       "tokens/total": 11349168,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 171518
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.012176645919680595,
+      "grad_norm": 0.6044301986694336,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 8.402287494391203e-05,
+      "loss": 0.006350134499371052,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00008,
+      "ppl": 1.00637,
       "step": 376,
       "tokens/total": 11379328,
-      "tokens/train_per_sec_per_gpu": 40.48,
+      "tokens/train_per_sec_per_gpu": 40.62,
       "tokens/trainable": 172007
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.24136756360530853,
+      "grad_norm": 0.25446587800979614,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 0.0013303000014275312,
+      "loss": 0.0018724403344094753,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00133,
+      "ppl": 1.00187,
       "step": 377,
       "tokens/total": 11409568,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 172468
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.021714258939027786,
+      "grad_norm": 0.0019207323202863336,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 0.00020160498388577253,
+      "loss": 3.049923907383345e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 378,
       "tokens/total": 11439968,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 172923
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.35412073135375977,
+      "grad_norm": 0.33491548895835876,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 0.015761105343699455,
+      "loss": 0.02673855796456337,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.01589,
+      "ppl": 1.0271,
       "step": 379,
       "tokens/total": 11470496,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 173370
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005766173824667931,
+      "grad_norm": 0.006177723873406649,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 2.5790239305933937e-05,
+      "loss": 5.195035191718489e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 380,
       "tokens/total": 11500720,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 173837
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.005898744333535433,
+      "grad_norm": 0.06973031163215637,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 5.821501690661535e-05,
+      "loss": 0.0006310560274869204,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00006,
+      "ppl": 1.00063,
       "step": 381,
       "tokens/total": 11531280,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 174294
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005451703444123268,
+      "grad_norm": 0.0039021887350827456,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 4.0181505028158426e-05,
+      "loss": 4.849781544180587e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 382,
       "tokens/total": 11561264,
-      "tokens/train_per_sec_per_gpu": 30.78,
+      "tokens/train_per_sec_per_gpu": 30.84,
       "tokens/trainable": 174701
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0038302047178149223,
+      "grad_norm": 0.0007975143962539732,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 3.403786467970349e-05,
+      "loss": 2.3489263185183518e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 383,
       "tokens/total": 11591616,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.05,
       "tokens/trainable": 175131
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.004227485507726669,
+      "grad_norm": 0.012541128322482109,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 3.562243364285678e-05,
+      "loss": 0.00011495217040646821,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00004,
+      "ppl": 1.00011,
       "step": 384,
       "tokens/total": 11622000,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 175586
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.013361346907913685,
+      "grad_norm": 0.004752015229314566,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0001464220113120973,
+      "loss": 6.502695032395422e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00007,
       "step": 385,
       "tokens/total": 11652352,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176026
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.008831475861370564,
+      "grad_norm": 0.014192809350788593,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 8.2662510976661e-05,
+      "loss": 8.075163350440562e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00008,
       "step": 386,
       "tokens/total": 11682736,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.91,
       "tokens/trainable": 176512
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.009311008267104626,
+      "grad_norm": 0.007288333959877491,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 0.0001226613821927458,
+      "loss": 8.845872798701748e-05,
       "memory/device_reserved (GiB)": 34.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00009,
       "step": 387,
       "tokens/total": 11713136,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.45,
       "tokens/trainable": 176943
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.0017816838808357716,
+      "grad_norm": 0.10162956267595291,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 2.3221660740091465e-05,
+      "loss": 0.0007253064541146159,
       "memory/device_reserved (GiB)": 35.23,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00002,
+      "ppl": 1.00073,
       "step": 388,
       "tokens/total": 11743808,
-      "tokens/train_per_sec_per_gpu": 32.06,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 177370
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.023602057248353958,
+      "grad_norm": 0.005729300435632467,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 0.00012638044427148998,
+      "loss": 8.652975520817563e-05,
       "memory/device_reserved (GiB)": 35.58,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00013,
+      "ppl": 1.00009,
       "step": 389,
       "tokens/total": 11774144,
-      "tokens/train_per_sec_per_gpu": 35.14,
+      "tokens/train_per_sec_per_gpu": 35.2,
       "tokens/trainable": 177851
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.04383797571063042,
+      "grad_norm": 0.0715414360165596,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 0.0004640861588995904,
+      "loss": 0.0007950748549774289,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00046,
+      "ppl": 1.0008,
       "step": 390,
       "tokens/total": 11804272,
-      "tokens/train_per_sec_per_gpu": 34.68,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 178310
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.061932601034641266,
+      "grad_norm": 0.009775097481906414,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.00033243370126001537,
+      "loss": 0.00010615254723234102,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00033,
+      "ppl": 1.00011,
       "step": 391,
       "tokens/total": 11832432,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.73,
       "tokens/trainable": 178736
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.04983547702431679,
+      "grad_norm": 0.0015552763361483812,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 0.000599355495069176,
+      "loss": 2.954876617877744e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0006,
+      "ppl": 1.00003,
       "step": 392,
       "tokens/total": 11862960,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 179233
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 0.012036106549203396,
+      "grad_norm": 0.026106838136911392,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0001237639953615144,
+      "loss": 0.00021109850786160678,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00012,
+      "ppl": 1.00021,
       "step": 393,
       "tokens/total": 11893232,
-      "tokens/train_per_sec_per_gpu": 38.49,
+      "tokens/train_per_sec_per_gpu": 38.57,
       "tokens/trainable": 179730
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.008992817252874374,
+      "grad_norm": 0.020347680896520615,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 0.00010385089262854308,
+      "loss": 0.0002601295418571681,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0001,
+      "ppl": 1.00026,
       "step": 394,
       "tokens/total": 11923664,
-      "tokens/train_per_sec_per_gpu": 37.81,
+      "tokens/train_per_sec_per_gpu": 37.87,
       "tokens/trainable": 180227
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.24631057679653168,
+      "grad_norm": 0.005419979803264141,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.0031685903668403625,
+      "loss": 4.929217175231315e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00317,
+      "ppl": 1.00005,
       "step": 395,
       "tokens/total": 11954032,
-      "tokens/train_per_sec_per_gpu": 32.85,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 180687
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.06871633976697922,
+      "grad_norm": 0.0031002764590084553,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 0.0005168755305930972,
+      "loss": 6.144218787085265e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00052,
+      "ppl": 1.00006,
       "step": 396,
       "tokens/total": 11984512,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 181141
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.011674679815769196,
+      "grad_norm": 0.006748533807694912,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 0.00017496946384198964,
+      "loss": 9.171784040518105e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00017,
+      "ppl": 1.00009,
       "step": 397,
       "tokens/total": 12014880,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 36.02,
       "tokens/trainable": 181648
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.14610664546489716,
+      "grad_norm": 0.08927815407514572,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.0014660547021776438,
+      "loss": 0.001000746968202293,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00147,
+      "ppl": 1.001,
       "step": 398,
       "tokens/total": 12044928,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 182109
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.016359565779566765,
+      "grad_norm": 0.004651801194995642,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 7.928608101792634e-05,
+      "loss": 5.3439554903889075e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00008,
+      "ppl": 1.00005,
       "step": 399,
       "tokens/total": 12075344,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 182559
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0010484450031071901,
+      "grad_norm": 0.0017514342907816172,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 1.526270352769643e-05,
+      "loss": 4.570486271404661e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00005,
       "step": 400,
       "tokens/total": 12105616,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.17,
       "tokens/trainable": 182985
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.0005900752730667591,
+      "grad_norm": 0.01574699766933918,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 1.2166316082584672e-05,
+      "loss": 0.0001242965809069574,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 401,
       "tokens/total": 12136080,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 183440
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.013194791041314602,
+      "grad_norm": 0.007152364123612642,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 0.0001970212033484131,
+      "loss": 0.0001222842838615179,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.0002,
+      "ppl": 1.00012,
       "step": 402,
       "tokens/total": 12166240,
-      "tokens/train_per_sec_per_gpu": 34.52,
+      "tokens/train_per_sec_per_gpu": 34.51,
       "tokens/trainable": 183917
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.02566305734217167,
+      "grad_norm": 0.0015342400874942541,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.00022389904188457876,
+      "loss": 3.621872019721195e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00022,
+      "ppl": 1.00004,
       "step": 403,
       "tokens/total": 12196544,
-      "tokens/train_per_sec_per_gpu": 30.39,
+      "tokens/train_per_sec_per_gpu": 30.36,
       "tokens/trainable": 184348
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.008275284431874752,
+      "grad_norm": 0.0035214691888540983,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 8.35009923321195e-05,
+      "loss": 6.367819150909781e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00006,
       "step": 404,
       "tokens/total": 12226784,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 184786
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.038177311420440674,
+      "grad_norm": 0.0007211520569398999,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 0.00018527230713516474,
+      "loss": 2.0325338482507505e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00019,
+      "ppl": 1.00002,
       "step": 405,
       "tokens/total": 12257264,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 185249
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.007672510575503111,
+      "grad_norm": 0.0014804014936089516,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 0.00011439670925028622,
+      "loss": 3.379736153874546e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00003,
       "step": 406,
       "tokens/total": 12287584,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 185680
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.003159885760396719,
+      "grad_norm": 0.004397980403155088,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 4.649449692806229e-05,
+      "loss": 5.1796458137687296e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00005,
       "step": 407,
       "tokens/total": 12317984,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 186100
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.32744285464286804,
+      "grad_norm": 0.0026473007164895535,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 0.002739987801760435,
+      "loss": 3.441090666456148e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00274,
+      "ppl": 1.00003,
       "step": 408,
       "tokens/total": 12348144,
-      "tokens/train_per_sec_per_gpu": 39.82,
+      "tokens/train_per_sec_per_gpu": 39.84,
       "tokens/trainable": 186565
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.0006675662589259446,
+      "grad_norm": 0.0289909727871418,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 1.2970660463906825e-05,
+      "loss": 0.00021602815832011402,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00022,
       "step": 409,
       "tokens/total": 12378544,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 187017
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.003976705949753523,
+      "grad_norm": 0.007092812564224005,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 3.857718547806144e-05,
+      "loss": 9.646185935707763e-05,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.0001,
       "step": 410,
       "tokens/total": 12408784,
-      "tokens/train_per_sec_per_gpu": 33.0,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 187468
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.0024577646981924772,
+      "grad_norm": 0.02710823155939579,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 2.431379834888503e-05,
+      "loss": 0.0003458422143012285,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00002,
+      "ppl": 1.00035,
       "step": 411,
       "tokens/total": 12438832,
-      "tokens/train_per_sec_per_gpu": 37.61,
+      "tokens/train_per_sec_per_gpu": 37.57,
       "tokens/trainable": 187961
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003754909848794341,
+      "grad_norm": 0.29485711455345154,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 4.119630830246024e-05,
+      "loss": 0.0032923028338700533,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.0033,
       "step": 412,
       "tokens/total": 12469296,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 188447
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.3583323657512665,
+      "grad_norm": 0.22881586849689484,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.004735157359391451,
+      "loss": 0.0035809341352432966,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00475,
+      "ppl": 1.00359,
       "step": 413,
       "tokens/total": 12499392,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 188903
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.049485232681035995,
+      "grad_norm": 0.0026347371749579906,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 0.00019404500199016184,
+      "loss": 6.0475373174995184e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.00006,
       "step": 414,
       "tokens/total": 12529952,
-      "tokens/train_per_sec_per_gpu": 36.7,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 189398
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0030447980388998985,
+      "grad_norm": 0.0017223359318450093,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 2.0038012735312805e-05,
+      "loss": 4.137849100516178e-05,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 415,
       "tokens/total": 12560240,
       "tokens/train_per_sec_per_gpu": 34.16,
@@ -5821,293 +5821,293 @@
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.25415608286857605,
+      "grad_norm": 0.011837545782327652,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 0.0016201161779463291,
+      "loss": 0.00011106643069069833,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00162,
+      "ppl": 1.00011,
       "step": 416,
       "tokens/total": 12590736,
-      "tokens/train_per_sec_per_gpu": 32.64,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 190306
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.006711115129292011,
+      "grad_norm": 0.01777251996099949,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 4.321872620494105e-05,
+      "loss": 0.0001219596597366035,
       "memory/device_reserved (GiB)": 38.05,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 417,
       "tokens/total": 12621168,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 190746
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.004931971430778503,
+      "grad_norm": 0.003218573285266757,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 4.740363510791212e-05,
+      "loss": 5.22282425663434e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00005,
       "step": 418,
       "tokens/total": 12651456,
-      "tokens/train_per_sec_per_gpu": 31.07,
+      "tokens/train_per_sec_per_gpu": 31.15,
       "tokens/trainable": 191156
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.007677357643842697,
+      "grad_norm": 0.013744015246629715,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 7.545409607701004e-05,
+      "loss": 0.0001364837517030537,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00008,
+      "ppl": 1.00014,
       "step": 419,
       "tokens/total": 12681760,
-      "tokens/train_per_sec_per_gpu": 29.47,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 191573
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.015445503406226635,
+      "grad_norm": 0.25013473629951477,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 0.0001045453900587745,
+      "loss": 0.002094803610816598,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0001,
+      "ppl": 1.0021,
       "step": 420,
       "tokens/total": 12712144,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.16882061958312988,
+      "grad_norm": 0.14290185272693634,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 0.0024146074429154396,
+      "loss": 0.001426510512828827,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00242,
+      "ppl": 1.00143,
       "step": 421,
       "tokens/total": 12742400,
-      "tokens/train_per_sec_per_gpu": 32.82,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 192453
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.002461223630234599,
+      "grad_norm": 0.0009390547638759017,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 3.336594454594888e-05,
+      "loss": 1.822916055971291e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12772288,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 192886
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.0005388484569266438,
+      "grad_norm": 0.004544281866401434,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 8.696397344465367e-06,
+      "loss": 6.211431173142046e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 423,
       "tokens/total": 12802880,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.87,
       "tokens/trainable": 193323
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.005009706597775221,
+      "grad_norm": 0.039859045296907425,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 4.8607362259645015e-05,
+      "loss": 0.0005113891093060374,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00051,
       "step": 424,
       "tokens/total": 12833456,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 193753
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.001982118235900998,
+      "grad_norm": 0.00408409908413887,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 3.262238169554621e-05,
+      "loss": 7.249199552461505e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00007,
       "step": 425,
       "tokens/total": 12863952,
-      "tokens/train_per_sec_per_gpu": 36.33,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 194240
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.05305991321802139,
+      "grad_norm": 0.13116440176963806,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 0.0005229170201346278,
+      "loss": 0.0015061571029946208,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00052,
+      "ppl": 1.00151,
       "step": 426,
       "tokens/total": 12894352,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 194680
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.001265498693101108,
+      "grad_norm": 0.010440024547278881,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 1.5427456673933193e-05,
+      "loss": 8.445358253084123e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00008,
       "step": 427,
       "tokens/total": 12924688,
-      "tokens/train_per_sec_per_gpu": 38.73,
+      "tokens/train_per_sec_per_gpu": 38.77,
       "tokens/trainable": 195186
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.0016263640718534589,
+      "grad_norm": 0.007594608701765537,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 2.7778178264270537e-05,
+      "loss": 9.946282807504758e-05,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 428,
       "tokens/total": 12955072,
-      "tokens/train_per_sec_per_gpu": 36.51,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 195693
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.013330154120922089,
+      "grad_norm": 0.006465950049459934,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 6.972272240091115e-05,
+      "loss": 0.00013557568308897316,
       "memory/device_reserved (GiB)": 36.05,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 429,
       "tokens/total": 12985328,
-      "tokens/train_per_sec_per_gpu": 36.93,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 196130
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.0034615101758390665,
+      "grad_norm": 0.00230405549518764,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 4.500148497754708e-05,
+      "loss": 3.7114587030373514e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 430,
       "tokens/total": 13015600,
-      "tokens/train_per_sec_per_gpu": 29.58,
+      "tokens/train_per_sec_per_gpu": 29.53,
       "tokens/trainable": 196530
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.03424045443534851,
+      "grad_norm": 0.0041900877840816975,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0002949235204141587,
+      "loss": 6.116987788118422e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00029,
+      "ppl": 1.00006,
       "step": 431,
       "tokens/total": 13045728,
-      "tokens/train_per_sec_per_gpu": 29.88,
+      "tokens/train_per_sec_per_gpu": 29.94,
       "tokens/trainable": 196974
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0064499350264668465,
+      "grad_norm": 0.02951621450483799,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 7.787663344061002e-05,
+      "loss": 0.0004236411186866462,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00008,
+      "ppl": 1.00042,
       "step": 432,
       "tokens/total": 13075920,
-      "tokens/train_per_sec_per_gpu": 35.25,
+      "tokens/train_per_sec_per_gpu": 35.24,
       "tokens/trainable": 197454
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.00021908426424488425,
+      "grad_norm": 0.0003202867810614407,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 6.603059773624409e-06,
+      "loss": 1.1190046279807575e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 433,
       "tokens/total": 13106128,
-      "tokens/train_per_sec_per_gpu": 31.4,
+      "tokens/train_per_sec_per_gpu": 31.43,
       "tokens/trainable": 197902
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.0002612156968098134,
+      "grad_norm": 0.0006868013297207654,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 6.425582796509843e-06,
+      "loss": 2.1185776859056205e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 434,
       "tokens/total": 13136368,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.06,
       "tokens/trainable": 198374
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.001639618189074099,
+      "grad_norm": 0.0025407401844859123,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 2.5485322112217546e-05,
+      "loss": 4.928320413455367e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 435,
       "tokens/total": 13166704,
-      "tokens/train_per_sec_per_gpu": 37.54,
+      "tokens/train_per_sec_per_gpu": 37.59,
       "tokens/trainable": 198846
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.01887959986925125,
+      "grad_norm": 0.0013901089550927281,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0002141120348824188,
+      "loss": 3.577578900149092e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00021,
+      "ppl": 1.00004,
       "step": 436,
       "tokens/total": 13197344,
       "tokens/train_per_sec_per_gpu": 37.67,
@@ -6115,41 +6115,41 @@
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.0012294527841731906,
+      "grad_norm": 0.003646081080660224,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.316079735464882e-05,
+      "loss": 6.294051127042621e-05,
       "memory/device_reserved (GiB)": 36.06,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 437,
       "tokens/total": 13227664,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 199777
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.005585651844739914,
+      "grad_norm": 0.030430182814598083,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 4.5935248635942116e-05,
+      "loss": 0.0001770013477653265,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00005,
+      "ppl": 1.00018,
       "step": 438,
       "tokens/total": 13257632,
-      "tokens/train_per_sec_per_gpu": 30.82,
+      "tokens/train_per_sec_per_gpu": 30.93,
       "tokens/trainable": 200194
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.001553745474666357,
+      "grad_norm": 0.0018230377463623881,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 2.2851421817904338e-05,
+      "loss": 3.7613608583342284e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13288240,
       "tokens/train_per_sec_per_gpu": 33.88,
@@ -6157,559 +6157,559 @@
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0029791900888085365,
+      "grad_norm": 0.0029376039747148752,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 4.093274037586525e-05,
+      "loss": 4.779352093464695e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00004,
+      "ppl": 1.00005,
       "step": 440,
       "tokens/total": 13318992,
-      "tokens/train_per_sec_per_gpu": 33.79,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 201121
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0017392379231750965,
+      "grad_norm": 0.0018373411148786545,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 7.29740804672474e-06,
+      "loss": 3.149824624415487e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 441,
       "tokens/total": 13349392,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.81,
       "tokens/trainable": 201598
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0009943305049091578,
+      "grad_norm": 0.0015428521437570453,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 1.7162077710963786e-05,
+      "loss": 2.0810390196857043e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 442,
       "tokens/total": 13379648,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 202068
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0011473585618659854,
+      "grad_norm": 0.0010770867811515927,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 2.3819740817998536e-05,
+      "loss": 2.303836117789615e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00002,
       "step": 443,
       "tokens/total": 13409936,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.5,
       "tokens/trainable": 202524
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.006799118127673864,
+      "grad_norm": 0.006372133269906044,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 6.556943117175251e-05,
+      "loss": 7.449327677022666e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
       "ppl": 1.00007,
       "step": 444,
       "tokens/total": 13440288,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 202996
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.0011645558988675475,
+      "grad_norm": 0.007094603031873703,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 1.8389995602774434e-05,
+      "loss": 4.211071063764393e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 445,
       "tokens/total": 13468880,
-      "tokens/train_per_sec_per_gpu": 39.14,
+      "tokens/train_per_sec_per_gpu": 39.19,
       "tokens/trainable": 203470
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.020891210064291954,
+      "grad_norm": 0.006931444630026817,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 0.00016399261949118227,
+      "loss": 8.66219779709354e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00016,
+      "ppl": 1.00009,
       "step": 446,
       "tokens/total": 13499184,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 203919
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.09615519642829895,
+      "grad_norm": 0.0041664596647024155,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 0.0010675137164071202,
+      "loss": 4.786982390214689e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00107,
+      "ppl": 1.00005,
       "step": 447,
       "tokens/total": 13529632,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 204376
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.09744399785995483,
+      "grad_norm": 0.042264848947525024,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.001196134602651,
+      "loss": 0.00027092613163404167,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0012,
+      "ppl": 1.00027,
       "step": 448,
       "tokens/total": 13560080,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.06,
       "tokens/trainable": 204821
     },
     {
       "epoch": 1.75390625,
-      "grad_norm": 0.0007810555398464203,
+      "grad_norm": 0.00043735766666941345,
       "learning_rate": 1.3780999708818058e-05,
-      "loss": 8.019956112548243e-06,
+      "loss": 1.2837479516747408e-05,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
       "ppl": 1.00001,
       "step": 449,
       "tokens/total": 13590496,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 205271
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 0.034087613224983215,
+      "grad_norm": 0.0012977999867871404,
       "learning_rate": 1.3665385037857758e-05,
-      "loss": 0.00010244422446703538,
+      "loss": 2.1102820028318092e-05,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0001,
+      "ppl": 1.00002,
       "step": 450,
       "tokens/total": 13620960,
-      "tokens/train_per_sec_per_gpu": 30.95,
+      "tokens/train_per_sec_per_gpu": 30.92,
       "tokens/trainable": 205723
     },
     {
       "epoch": 1.76171875,
-      "grad_norm": 0.02930639684200287,
+      "grad_norm": 0.005675352178514004,
       "learning_rate": 1.3551490468947126e-05,
-      "loss": 0.00021926099725533277,
+      "loss": 7.465962698915973e-05,
       "memory/device_reserved (GiB)": 35.04,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00022,
+      "ppl": 1.00007,
       "step": 451,
       "tokens/total": 13651344,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.48,
       "tokens/trainable": 206169
     },
     {
       "epoch": 1.765625,
-      "grad_norm": 0.0109552051872015,
+      "grad_norm": 0.0013997588539496064,
       "learning_rate": 1.3439320741704075e-05,
-      "loss": 7.999646186362952e-05,
+      "loss": 2.723011130001396e-05,
       "memory/device_reserved (GiB)": 35.04,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 452,
       "tokens/total": 13681792,
-      "tokens/train_per_sec_per_gpu": 32.74,
+      "tokens/train_per_sec_per_gpu": 32.79,
       "tokens/trainable": 206592
     },
     {
       "epoch": 1.76953125,
-      "grad_norm": 0.004453443922102451,
+      "grad_norm": 0.012491011992096901,
       "learning_rate": 1.3328880523968808e-05,
-      "loss": 3.79011980840005e-05,
+      "loss": 9.182744543068111e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00009,
       "step": 453,
       "tokens/total": 13712256,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.07,
       "tokens/trainable": 207055
     },
     {
       "epoch": 1.7734375,
-      "grad_norm": 0.0005650459788739681,
+      "grad_norm": 0.000924609019421041,
       "learning_rate": 1.3220174411609587e-05,
-      "loss": 1.2244867320987396e-05,
+      "loss": 2.1621295672957785e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 454,
       "tokens/total": 13742128,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.61,
       "tokens/trainable": 207494
     },
     {
       "epoch": 1.77734375,
-      "grad_norm": 0.0007141040405258536,
+      "grad_norm": 0.000411411514505744,
       "learning_rate": 1.3113206928331471e-05,
-      "loss": 1.3571594536188059e-05,
+      "loss": 1.2427874025888741e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00001,
       "step": 455,
       "tokens/total": 13772160,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.03,
       "tokens/trainable": 207956
     },
     {
       "epoch": 1.78125,
-      "grad_norm": 0.0008097058744169772,
+      "grad_norm": 0.0015909038484096527,
       "learning_rate": 1.300798252548806e-05,
-      "loss": 1.1659047231660224e-05,
+      "loss": 3.0332066671689972e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 456,
       "tokens/total": 13802448,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.42,
       "tokens/trainable": 208408
     },
     {
       "epoch": 1.78515625,
-      "grad_norm": 0.0012649539858102798,
+      "grad_norm": 0.0030019236728549004,
       "learning_rate": 1.2904505581896265e-05,
-      "loss": 9.200517524732277e-06,
+      "loss": 3.378765541128814e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 457,
       "tokens/total": 13832880,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 35.92,
       "tokens/trainable": 208889
     },
     {
       "epoch": 1.7890625,
-      "grad_norm": 0.014978500083088875,
+      "grad_norm": 0.0016427476657554507,
       "learning_rate": 1.2802780403654082e-05,
-      "loss": 0.0001208957692142576,
+      "loss": 3.3130341762444004e-05,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00012,
+      "ppl": 1.00003,
       "step": 458,
       "tokens/total": 13862880,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.02,
       "tokens/trainable": 209336
     },
     {
       "epoch": 1.79296875,
-      "grad_norm": 0.004875013139098883,
+      "grad_norm": 0.0006426559993997216,
       "learning_rate": 1.2702811223961408e-05,
-      "loss": 3.241455851821229e-05,
+      "loss": 1.784306368790567e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 459,
       "tokens/total": 13893440,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.21,
       "tokens/trainable": 209797
     },
     {
       "epoch": 1.796875,
-      "grad_norm": 0.0009627968538552523,
+      "grad_norm": 0.0016685863956809044,
       "learning_rate": 1.2604602202943861e-05,
-      "loss": 1.980438355531078e-05,
+      "loss": 3.2843898225110024e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 460,
       "tokens/total": 13923824,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.03,
       "tokens/trainable": 210240
     },
     {
       "epoch": 1.80078125,
-      "grad_norm": 0.019912317395210266,
+      "grad_norm": 0.03454679995775223,
       "learning_rate": 1.2508157427479686e-05,
-      "loss": 0.00010681153071345761,
+      "loss": 7.200831169029698e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00011,
+      "ppl": 1.00007,
       "step": 461,
       "tokens/total": 13954208,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 34.9,
       "tokens/trainable": 210702
     },
     {
       "epoch": 1.8046875,
-      "grad_norm": 0.0005370390135794878,
+      "grad_norm": 0.0021488473284989595,
       "learning_rate": 1.2413480911029655e-05,
-      "loss": 1.1081473530794028e-05,
+      "loss": 3.525309875840321e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 462,
       "tokens/total": 13984784,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.44,
       "tokens/trainable": 211140
     },
     {
       "epoch": 1.80859375,
-      "grad_norm": 0.000815370527561754,
+      "grad_norm": 0.0007275328389368951,
       "learning_rate": 1.2320576593470082e-05,
-      "loss": 1.5510300727328286e-05,
+      "loss": 1.328821963397786e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 463,
       "tokens/total": 14014896,
-      "tokens/train_per_sec_per_gpu": 32.6,
+      "tokens/train_per_sec_per_gpu": 32.54,
       "tokens/trainable": 211575
     },
     {
       "epoch": 1.8125,
-      "grad_norm": 0.02875823900103569,
+      "grad_norm": 0.03173735365271568,
       "learning_rate": 1.2229448340928828e-05,
-      "loss": 0.00013141569797880948,
+      "loss": 0.00014542456483468413,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00013,
+      "ppl": 1.00015,
       "step": 464,
       "tokens/total": 14045312,
-      "tokens/train_per_sec_per_gpu": 39.03,
+      "tokens/train_per_sec_per_gpu": 38.86,
       "tokens/trainable": 212083
     },
     {
       "epoch": 1.81640625,
-      "grad_norm": 0.04558868706226349,
+      "grad_norm": 0.0005439479718916118,
       "learning_rate": 1.2140099945624458e-05,
-      "loss": 0.0002969688503071666,
+      "loss": 1.4493984053842723e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0003,
+      "ppl": 1.00001,
       "step": 465,
       "tokens/total": 14075840,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 212558
     },
     {
       "epoch": 1.8203125,
-      "grad_norm": 0.0004122587270103395,
+      "grad_norm": 0.0015477532288059592,
       "learning_rate": 1.205253512570841e-05,
-      "loss": 9.937594768416602e-06,
+      "loss": 2.787737685139291e-05,
       "memory/device_reserved (GiB)": 35.97,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 466,
       "tokens/total": 14106016,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.73,
       "tokens/trainable": 213014
     },
     {
       "epoch": 1.82421875,
-      "grad_norm": 0.0010108116548508406,
+      "grad_norm": 0.0007428264361806214,
       "learning_rate": 1.1966757525110255e-05,
-      "loss": 1.5579567843815312e-05,
+      "loss": 2.1110219677211717e-05,
       "memory/device_reserved (GiB)": 37.99,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
       "ppl": 1.00002,
       "step": 467,
       "tokens/total": 14136512,
-      "tokens/train_per_sec_per_gpu": 34.03,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 213465
     },
     {
       "epoch": 1.828125,
-      "grad_norm": 0.0043095871806144714,
+      "grad_norm": 0.002182955853641033,
       "learning_rate": 1.1882770713386095e-05,
-      "loss": 3.2609641493763775e-05,
+      "loss": 4.059498314745724e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00003,
+      "ppl": 1.00004,
       "step": 468,
       "tokens/total": 14166832,
-      "tokens/train_per_sec_per_gpu": 36.62,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 213940
     },
     {
       "epoch": 1.83203125,
-      "grad_norm": 0.005317045841366053,
+      "grad_norm": 0.004772448446601629,
       "learning_rate": 1.180057818556998e-05,
-      "loss": 2.144884638255462e-05,
+      "loss": 3.664854375529103e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 469,
       "tokens/total": 14197360,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 214415
     },
     {
       "epoch": 1.8359375,
-      "grad_norm": 0.0003264908737037331,
+      "grad_norm": 0.00038694078102707863,
       "learning_rate": 1.1720183362028494e-05,
-      "loss": 5.6935500651889015e-06,
+      "loss": 1.1124819138785824e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 470,
       "tokens/total": 14227600,
-      "tokens/train_per_sec_per_gpu": 34.26,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 214880
     },
     {
       "epoch": 1.83984375,
-      "grad_norm": 0.013723314739763737,
+      "grad_norm": 0.0008439846569672227,
       "learning_rate": 1.1641589588318387e-05,
-      "loss": 0.00010619591921567917,
+      "loss": 2.0906983991153538e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00011,
+      "ppl": 1.00002,
       "step": 471,
       "tokens/total": 14257888,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.34,
       "tokens/trainable": 215334
     },
     {
       "epoch": 1.84375,
-      "grad_norm": 0.00020912640320602804,
+      "grad_norm": 0.0005926606827415526,
       "learning_rate": 1.1564800135047418e-05,
-      "loss": 4.586940121953376e-06,
+      "loss": 1.178990351036191e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 472,
       "tokens/total": 14288208,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 215822
     },
     {
       "epoch": 1.84765625,
-      "grad_norm": 0.10667876899242401,
+      "grad_norm": 0.003897819435223937,
       "learning_rate": 1.148981819773816e-05,
-      "loss": 0.0006915333215147257,
+      "loss": 5.020621756557375e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.00005,
       "step": 473,
       "tokens/total": 14318448,
-      "tokens/train_per_sec_per_gpu": 30.86,
+      "tokens/train_per_sec_per_gpu": 30.82,
       "tokens/trainable": 216239
     },
     {
       "epoch": 1.8515625,
-      "grad_norm": 0.013282700441777706,
+      "grad_norm": 0.0013807902578264475,
       "learning_rate": 1.1416646896695086e-05,
-      "loss": 6.861680594738573e-05,
+      "loss": 2.2795318727730773e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 474,
       "tokens/total": 14348912,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.35,
       "tokens/trainable": 216691
     },
     {
       "epoch": 1.85546875,
-      "grad_norm": 0.0074070473201572895,
+      "grad_norm": 0.00048289448022842407,
       "learning_rate": 1.1345289276874717e-05,
-      "loss": 8.334079757332802e-05,
+      "loss": 1.4630711120844353e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 475,
       "tokens/total": 14379440,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 217162
     },
     {
       "epoch": 1.859375,
-      "grad_norm": 0.0017771078273653984,
+      "grad_norm": 0.0004491541185416281,
       "learning_rate": 1.1275748307758873e-05,
-      "loss": 1.803937993827276e-05,
+      "loss": 1.4460356396739371e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 476,
       "tokens/total": 14409840,
-      "tokens/train_per_sec_per_gpu": 35.84,
+      "tokens/train_per_sec_per_gpu": 35.88,
       "tokens/trainable": 217638
     },
     {
       "epoch": 1.86328125,
-      "grad_norm": 0.00044826362864114344,
+      "grad_norm": 0.0024057701230049133,
       "learning_rate": 1.1208026883231147e-05,
-      "loss": 9.577534910931718e-06,
+      "loss": 4.9342866986989975e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 477,
       "tokens/total": 14440256,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 218053
     },
     {
       "epoch": 1.8671875,
-      "grad_norm": 0.0009560272446833551,
+      "grad_norm": 0.0008485029684379697,
       "learning_rate": 1.1142127821456433e-05,
-      "loss": 1.0945323083433323e-05,
+      "loss": 2.05296601052396e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 478,
       "tokens/total": 14470704,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.09,
       "tokens/trainable": 218503
     },
     {
       "epoch": 1.87109375,
-      "grad_norm": 0.2086740881204605,
+      "grad_norm": 0.011945155449211597,
       "learning_rate": 1.1078053864763674e-05,
-      "loss": 0.0012299084337428212,
+      "loss": 8.75981932040304e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00123,
+      "ppl": 1.00009,
       "step": 479,
       "tokens/total": 14501072,
       "tokens/train_per_sec_per_gpu": 36.19,
@@ -6717,13 +6717,13 @@
     },
     {
       "epoch": 1.875,
-      "grad_norm": 0.019221138209104538,
+      "grad_norm": 0.004776590969413519,
       "learning_rate": 1.1015807679531756e-05,
-      "loss": 0.00011615331459324807,
+      "loss": 7.07923318259418e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00012,
+      "ppl": 1.00007,
       "step": 480,
       "tokens/total": 14531200,
       "tokens/train_per_sec_per_gpu": 35.51,
@@ -6731,251 +6731,251 @@
     },
     {
       "epoch": 1.87890625,
-      "grad_norm": 0.002074754796922207,
+      "grad_norm": 0.0006404300802387297,
       "learning_rate": 1.0955391856078528e-05,
-      "loss": 2.3620896172360517e-05,
+      "loss": 1.6759000573074445e-05,
       "memory/device_reserved (GiB)": 38.63,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
       "ppl": 1.00002,
       "step": 481,
       "tokens/total": 14561152,
-      "tokens/train_per_sec_per_gpu": 38.29,
+      "tokens/train_per_sec_per_gpu": 38.25,
       "tokens/trainable": 219886
     },
     {
       "epoch": 1.8828125,
-      "grad_norm": 0.011151358485221863,
+      "grad_norm": 0.005522869061678648,
       "learning_rate": 1.0896808908553007e-05,
-      "loss": 5.527659959625453e-05,
+      "loss": 2.3341217456618324e-05,
       "memory/device_reserved (GiB)": 37.77,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 482,
       "tokens/total": 14591264,
-      "tokens/train_per_sec_per_gpu": 34.53,
+      "tokens/train_per_sec_per_gpu": 34.52,
       "tokens/trainable": 220343
     },
     {
       "epoch": 1.88671875,
-      "grad_norm": 0.010913084261119366,
+      "grad_norm": 0.001015707035548985,
       "learning_rate": 1.0840061274830763e-05,
-      "loss": 6.711594323860481e-05,
+      "loss": 2.3611488359165378e-05,
       "memory/device_reserved (GiB)": 37.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 483,
       "tokens/total": 14621568,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 220825
     },
     {
       "epoch": 1.890625,
-      "grad_norm": 0.3236236274242401,
+      "grad_norm": 0.004034205339848995,
       "learning_rate": 1.0785151316412473e-05,
-      "loss": 0.002836203668266535,
+      "loss": 5.646366116707213e-05,
       "memory/device_reserved (GiB)": 37.77,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00284,
+      "ppl": 1.00006,
       "step": 484,
       "tokens/total": 14652064,
-      "tokens/train_per_sec_per_gpu": 32.52,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 221282
     },
     {
       "epoch": 1.89453125,
-      "grad_norm": 0.0002806074626278132,
+      "grad_norm": 0.008657694794237614,
       "learning_rate": 1.0732081318325639e-05,
-      "loss": 5.300881639414001e-06,
+      "loss": 5.676104410667904e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 485,
       "tokens/total": 14682608,
-      "tokens/train_per_sec_per_gpu": 37.86,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 221780
     },
     {
       "epoch": 1.8984375,
-      "grad_norm": 0.009111613035202026,
+      "grad_norm": 0.0036515207029879093,
       "learning_rate": 1.0680853489029501e-05,
-      "loss": 3.664881660370156e-05,
+      "loss": 3.9208807720569894e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00004,
       "step": 486,
       "tokens/total": 14713168,
-      "tokens/train_per_sec_per_gpu": 31.52,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 222210
     },
     {
       "epoch": 1.90234375,
-      "grad_norm": 0.04484618082642555,
+      "grad_norm": 0.005618095863610506,
       "learning_rate": 1.0631469960323152e-05,
-      "loss": 0.0002056290686596185,
+      "loss": 5.3539326472673565e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00005,
       "step": 487,
       "tokens/total": 14743536,
-      "tokens/train_per_sec_per_gpu": 29.33,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 222646
     },
     {
       "epoch": 1.90625,
-      "grad_norm": 0.2784590721130371,
+      "grad_norm": 0.07197980582714081,
       "learning_rate": 1.0583932787256783e-05,
-      "loss": 0.008498833514750004,
+      "loss": 0.0006614382145926356,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00854,
+      "ppl": 1.00066,
       "step": 488,
       "tokens/total": 14773728,
-      "tokens/train_per_sec_per_gpu": 33.1,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 223066
     },
     {
       "epoch": 1.91015625,
-      "grad_norm": 0.0006373917567543685,
+      "grad_norm": 0.001169998780824244,
       "learning_rate": 1.0538243948046206e-05,
-      "loss": 9.457155101699755e-06,
+      "loss": 1.62669166456908e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 489,
       "tokens/total": 14803744,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 223512
     },
     {
       "epoch": 1.9140625,
-      "grad_norm": 0.00038701840094290674,
+      "grad_norm": 0.0014546489110216498,
       "learning_rate": 1.0494405343990523e-05,
-      "loss": 6.92771118337987e-06,
+      "loss": 3.051830208278261e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 490,
       "tokens/total": 14832112,
-      "tokens/train_per_sec_per_gpu": 38.96,
+      "tokens/train_per_sec_per_gpu": 38.95,
       "tokens/trainable": 223956
     },
     {
       "epoch": 1.91796875,
-      "grad_norm": 0.0015320515958592296,
+      "grad_norm": 0.0007275546086020768,
       "learning_rate": 1.0452418799392985e-05,
-      "loss": 2.272017445648089e-05,
+      "loss": 1.4892379113007337e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 491,
       "tokens/total": 14860560,
-      "tokens/train_per_sec_per_gpu": 36.33,
+      "tokens/train_per_sec_per_gpu": 36.3,
       "tokens/trainable": 224416
     },
     {
       "epoch": 1.921875,
-      "grad_norm": 0.016094110906124115,
+      "grad_norm": 0.0007378064910881221,
       "learning_rate": 1.0412286061485102e-05,
-      "loss": 0.00011934755457332358,
+      "loss": 1.902497206174303e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 492,
       "tokens/total": 14890848,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.12,
       "tokens/trainable": 224883
     },
     {
       "epoch": 1.92578125,
-      "grad_norm": 0.0016222589183598757,
+      "grad_norm": 0.0006582152564078569,
       "learning_rate": 1.03740088003539e-05,
-      "loss": 1.4765413652639836e-05,
+      "loss": 1.498015808465425e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
       "ppl": 1.00001,
       "step": 493,
       "tokens/total": 14921008,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.42,
       "tokens/trainable": 225324
     },
     {
       "epoch": 1.9296875,
-      "grad_norm": 0.004442836623638868,
+      "grad_norm": 0.000849059084430337,
       "learning_rate": 1.0337588608872463e-05,
-      "loss": 3.44704239978455e-05,
+      "loss": 1.0077818842546549e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 494,
       "tokens/total": 14951360,
-      "tokens/train_per_sec_per_gpu": 34.9,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 225783
     },
     {
       "epoch": 1.93359375,
-      "grad_norm": 0.002437110524624586,
+      "grad_norm": 0.0029649219941347837,
       "learning_rate": 1.0303027002633622e-05,
-      "loss": 1.8051592633128166e-05,
+      "loss": 2.8222328182891943e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 495,
       "tokens/total": 14981552,
-      "tokens/train_per_sec_per_gpu": 36.46,
+      "tokens/train_per_sec_per_gpu": 36.48,
       "tokens/trainable": 226281
     },
     {
       "epoch": 1.9375,
-      "grad_norm": 0.0015804325230419636,
+      "grad_norm": 0.001169885159470141,
       "learning_rate": 1.0270325419886884e-05,
-      "loss": 1.691515171842184e-05,
+      "loss": 2.592003693280276e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 496,
       "tokens/total": 15011584,
-      "tokens/train_per_sec_per_gpu": 36.72,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 226735
     },
     {
       "epoch": 1.94140625,
-      "grad_norm": 0.0036208343226462603,
+      "grad_norm": 0.006588887423276901,
       "learning_rate": 1.0239485221478599e-05,
-      "loss": 4.592180994222872e-05,
+      "loss": 9.911794040817767e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00005,
+      "ppl": 1.0001,
       "step": 497,
       "tokens/total": 15042016,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.3,
       "tokens/trainable": 227198
     },
     {
       "epoch": 1.9453125,
-      "grad_norm": 0.0005918977549299598,
+      "grad_norm": 0.0013167925644665956,
       "learning_rate": 1.0210507690795292e-05,
-      "loss": 1.5305626220651902e-05,
+      "loss": 2.8513590223155916e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.68,
       "memory/max_allocated (GiB)": 33.68,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 498,
       "tokens/total": 15071984,
       "tokens/train_per_sec_per_gpu": 31.07,
@@ -6983,69 +6983,69 @@
     },
     {
       "epoch": 1.94921875,
-      "grad_norm": 0.0013667694292962551,
+      "grad_norm": 0.0010883790673688054,
       "learning_rate": 1.0183394033710305e-05,
-      "loss": 1.3490713172359392e-05,
+      "loss": 2.4797469450277276e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 499,
       "tokens/total": 15102192,
-      "tokens/train_per_sec_per_gpu": 32.15,
+      "tokens/train_per_sec_per_gpu": 32.17,
       "tokens/trainable": 228074
     },
     {
       "epoch": 1.953125,
-      "grad_norm": 0.0027930096257478,
+      "grad_norm": 0.010283510200679302,
       "learning_rate": 1.0158145378533583e-05,
-      "loss": 2.8409334845491685e-05,
+      "loss": 4.6255343477241695e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 500,
       "tokens/total": 15132560,
-      "tokens/train_per_sec_per_gpu": 35.57,
+      "tokens/train_per_sec_per_gpu": 35.59,
       "tokens/trainable": 228541
     },
     {
       "epoch": 1.95703125,
-      "grad_norm": 0.0028320043347775936,
+      "grad_norm": 0.0011426788987591863,
       "learning_rate": 1.0134762775964726e-05,
-      "loss": 3.1829928047955036e-05,
+      "loss": 1.940131733135786e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 501,
       "tokens/total": 15163152,
-      "tokens/train_per_sec_per_gpu": 37.77,
+      "tokens/train_per_sec_per_gpu": 37.78,
       "tokens/trainable": 229031
     },
     {
       "epoch": 1.9609375,
-      "grad_norm": 0.0030925904866307974,
+      "grad_norm": 0.0012880039867013693,
       "learning_rate": 1.0113247199049278e-05,
-      "loss": 2.238066190329846e-05,
+      "loss": 2.073413270409219e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 502,
       "tokens/total": 15193456,
-      "tokens/train_per_sec_per_gpu": 38.68,
+      "tokens/train_per_sec_per_gpu": 38.76,
       "tokens/trainable": 229512
     },
     {
       "epoch": 1.96484375,
-      "grad_norm": 0.06389517337083817,
+      "grad_norm": 0.022319281473755836,
       "learning_rate": 1.0093599543138205e-05,
-      "loss": 0.0004140711098443717,
+      "loss": 0.00016162173415068537,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00041,
+      "ppl": 1.00016,
       "step": 503,
       "tokens/total": 15223680,
       "tokens/train_per_sec_per_gpu": 34.26,
@@ -7053,128 +7053,128 @@
     },
     {
       "epoch": 1.96875,
-      "grad_norm": 0.003928070887923241,
+      "grad_norm": 0.00031853109248913825,
       "learning_rate": 1.0075820625850675e-05,
-      "loss": 4.3977754103252664e-05,
+      "loss": 9.707431672723033e-06,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 504,
       "tokens/total": 15254336,
-      "tokens/train_per_sec_per_gpu": 35.81,
+      "tokens/train_per_sec_per_gpu": 35.84,
       "tokens/trainable": 230447
     },
     {
       "epoch": 1.97265625,
-      "grad_norm": 0.0022790853399783373,
+      "grad_norm": 0.0005880650132894516,
       "learning_rate": 1.0059911187040013e-05,
-      "loss": 2.180483534175437e-05,
+      "loss": 1.7119673429988325e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
       "ppl": 1.00002,
       "step": 505,
       "tokens/total": 15284496,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.3,
       "tokens/trainable": 230879
     },
     {
       "epoch": 1.9765625,
-      "grad_norm": 0.0022161444649100304,
+      "grad_norm": 0.0011501131812110543,
       "learning_rate": 1.0045871888762893e-05,
-      "loss": 2.00994672923116e-05,
+      "loss": 1.6414911442552693e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
       "ppl": 1.00002,
       "step": 506,
       "tokens/total": 15314976,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.03,
       "tokens/trainable": 231348
     },
     {
       "epoch": 1.98046875,
-      "grad_norm": 0.0109304990619421,
+      "grad_norm": 0.0007460714550688863,
       "learning_rate": 1.003370331525184e-05,
-      "loss": 9.907589264912531e-05,
+      "loss": 2.1825573639944196e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00002,
       "step": 507,
       "tokens/total": 15345280,
-      "tokens/train_per_sec_per_gpu": 35.22,
+      "tokens/train_per_sec_per_gpu": 35.32,
       "tokens/trainable": 231819
     },
     {
       "epoch": 1.984375,
-      "grad_norm": 0.0008610524819232523,
+      "grad_norm": 0.003101626643911004,
       "learning_rate": 1.002340597289085e-05,
-      "loss": 1.4185266991262324e-05,
+      "loss": 4.144266858929768e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 508,
       "tokens/total": 15375424,
-      "tokens/train_per_sec_per_gpu": 34.57,
+      "tokens/train_per_sec_per_gpu": 34.47,
       "tokens/trainable": 232269
     },
     {
       "epoch": 1.98828125,
-      "grad_norm": 0.18468032777309418,
+      "grad_norm": 0.0010362571338191628,
       "learning_rate": 1.0014980290194387e-05,
-      "loss": 0.0011317847529426217,
+      "loss": 1.4500470570055768e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00113,
+      "ppl": 1.00001,
       "step": 509,
       "tokens/total": 15405872,
-      "tokens/train_per_sec_per_gpu": 34.06,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 232716
     },
     {
       "epoch": 1.9921875,
-      "grad_norm": 0.0012291419552639127,
+      "grad_norm": 0.0019138582283630967,
       "learning_rate": 1.0008426617789489e-05,
-      "loss": 1.4661756722489372e-05,
+      "loss": 2.179224611609243e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 510,
       "tokens/total": 15436128,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 233178
     },
     {
       "epoch": 1.99609375,
-      "grad_norm": 0.003051232313737273,
+      "grad_norm": 0.0017450153827667236,
       "learning_rate": 1.0003745228401215e-05,
-      "loss": 2.0963430870324373e-05,
+      "loss": 2.1991130779497325e-05,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 511,
       "tokens/total": 15466464,
-      "tokens/train_per_sec_per_gpu": 35.95,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 233647
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.001929801655933261,
+      "grad_norm": 0.06926226615905762,
       "learning_rate": 1.0000936316841296e-05,
-      "loss": 1.9281407730886713e-05,
+      "loss": 0.0005933041102252901,
       "memory/device_reserved (GiB)": 38.44,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00002,
+      "ppl": 1.00059,
       "step": 512,
       "tokens/total": 15496368,
-      "tokens/train_per_sec_per_gpu": 29.42,
+      "tokens/train_per_sec_per_gpu": 29.4,
       "tokens/trainable": 234060
     }
   ],
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/training_provenance.json b/aft_wave_v2/coin_real_4x__agreement/training/training_provenance.json
index 5f0d0ecb67daf5aac0d94937aa960b180cfaf1b7..4a3ae65471f84d4af6ca1fb1522e9f11d6765edb 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/training_provenance.json
+++ b/aft_wave_v2/coin_real_4x__agreement/training/training_provenance.json
@@ -3,7 +3,7 @@
   "status": "complete",
   "stage": "aft_dispatch_v4_wide",
   "kind": "sft",
-  "created_at": "2026-08-18T12:41:50+00:00",
+  "created_at": "2026-08-18T14:03:47+00:00",
   "resolved_config": {
     "base_model": "/workspace/wave/parent",
     "base_model_config": "unsloth/gemma-3-12b-pt",
@@ -100,7 +100,7 @@
     "save_total_limit": 20
   },
   "seed": 42,
-  "completed_at": "2026-08-18T13:41:25+00:00",
+  "completed_at": "2026-08-18T15:02:36+00:00",
   "resolved_config_sha256": "edb37cd487bcc7b762488edbbc216703766f91248135937b12ce95a92f64df6e",
   "actual": {
     "global_step": 512,
@@ -130,10 +130,10 @@
     ],
     "trainer_state_source": "/workspace/wave/training/checkpoints/checkpoint-512/trainer_state.json",
     "trainer_state_snapshot": "/workspace/wave/training/trainer_state.final.json",
-    "trainer_state_sha256": "2dbfcbe1a03c6951df9b24ec896edee522b63548a54faf52ac159f28161dac6a",
+    "trainer_state_sha256": "02f9a66b02bb552b9fb5b0fe2cf523809d3b2ee4475683870014d1b13ad7a8a2",
     "trace_rows": 512,
     "trace_path": "/workspace/wave/training/training_trace.jsonl",
-    "trace_sha256": "be93695ba28788aeb6f0fd98000bb9b59fb86cba22501505a16c1bc809d19b7a",
+    "trace_sha256": "3162f25b3bb9dbc2b6e99501728763c5a9acec2b98a1dd4988902fa8ee0b3a2a",
     "first_learning_rate": 0.0,
     "last_learning_rate": 1.0000936316841296e-05
   }
diff --git a/aft_wave_v2/coin_real_4x__agreement/training/training_trace.jsonl b/aft_wave_v2/coin_real_4x__agreement/training/training_trace.jsonl
index 746378c4b881b3da1fff6af55e9033ef5a17d405..96417a5cb70e8410f522bb844b742e73872907de 100644
--- a/aft_wave_v2/coin_real_4x__agreement/training/training_trace.jsonl
+++ b/aft_wave_v2/coin_real_4x__agreement/training/training_trace.jsonl
@@ -1,512 +1,512 @@
-{"epoch": 0.00390625, "grad_norm": 0.870697021484375, "learning_rate": 0.0, "loss": 0.10251401364803314, "memory/device_reserved (GiB)": 34.94, "memory/max_active (GiB)": 32.98, "memory/max_allocated (GiB)": 32.98, "ppl": 1.10795, "step": 1, "tokens/total": 30464, "tokens/train_per_sec_per_gpu": 23.98, "tokens/trainable": 470}
-{"epoch": 0.0078125, "grad_norm": 0.8108459115028381, "learning_rate": 4.000000000000001e-06, "loss": 0.09678442776203156, "memory/device_reserved (GiB)": 35.83, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.10162, "step": 2, "tokens/total": 60800, "tokens/train_per_sec_per_gpu": 35.02, "tokens/trainable": 922}
-{"epoch": 0.01171875, "grad_norm": 1.8027335405349731, "learning_rate": 8.000000000000001e-06, "loss": 0.10952483862638474, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.11575, "step": 3, "tokens/total": 91136, "tokens/train_per_sec_per_gpu": 34.48, "tokens/trainable": 1396}
-{"epoch": 0.015625, "grad_norm": 1.0353018045425415, "learning_rate": 1.2e-05, "loss": 0.10303406417369843, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.10853, "step": 4, "tokens/total": 121552, "tokens/train_per_sec_per_gpu": 38.01, "tokens/trainable": 1850}
-{"epoch": 0.01953125, "grad_norm": 1.0778985023498535, "learning_rate": 1.6000000000000003e-05, "loss": 0.10945924371480942, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.11567, "step": 5, "tokens/total": 152304, "tokens/train_per_sec_per_gpu": 33.76, "tokens/trainable": 2302}
-{"epoch": 0.0234375, "grad_norm": 0.8929882645606995, "learning_rate": 2e-05, "loss": 0.08588902652263641, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.08969, "step": 6, "tokens/total": 182448, "tokens/train_per_sec_per_gpu": 31.47, "tokens/trainable": 2742}
-{"epoch": 0.02734375, "grad_norm": 1.6409597396850586, "learning_rate": 2.4e-05, "loss": 0.07352827489376068, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0763, "step": 7, "tokens/total": 212736, "tokens/train_per_sec_per_gpu": 36.58, "tokens/trainable": 3208}
-{"epoch": 0.03125, "grad_norm": 1.5843520164489746, "learning_rate": 2.8000000000000003e-05, "loss": 0.07798244059085846, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.0811, "step": 8, "tokens/total": 243024, "tokens/train_per_sec_per_gpu": 31.83, "tokens/trainable": 3655}
-{"epoch": 0.03515625, "grad_norm": 1.3725916147232056, "learning_rate": 3.2000000000000005e-05, "loss": 0.04634054750204086, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.04743, "step": 9, "tokens/total": 271264, "tokens/train_per_sec_per_gpu": 35.62, "tokens/trainable": 4091}
-{"epoch": 0.0390625, "grad_norm": 2.544938564300537, "learning_rate": 3.6e-05, "loss": 0.08677884936332703, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.09066, "step": 10, "tokens/total": 301520, "tokens/train_per_sec_per_gpu": 34.92, "tokens/trainable": 4553}
-{"epoch": 0.04296875, "grad_norm": 1.6364734172821045, "learning_rate": 4e-05, "loss": 0.07754456996917725, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.08063, "step": 11, "tokens/total": 331808, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 4992}
-{"epoch": 0.046875, "grad_norm": 2.167391538619995, "learning_rate": 4.4000000000000006e-05, "loss": 0.11765319854021072, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.12485, "step": 12, "tokens/total": 362224, "tokens/train_per_sec_per_gpu": 38.18, "tokens/trainable": 5494}
-{"epoch": 0.05078125, "grad_norm": 3.196911573410034, "learning_rate": 4.8e-05, "loss": 0.03510797768831253, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.03573, "step": 13, "tokens/total": 392432, "tokens/train_per_sec_per_gpu": 33.67, "tokens/trainable": 5933}
-{"epoch": 0.0546875, "grad_norm": 1.9654567241668701, "learning_rate": 5.2000000000000004e-05, "loss": 0.061097435653209686, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.063, "step": 14, "tokens/total": 422784, "tokens/train_per_sec_per_gpu": 29.53, "tokens/trainable": 6369}
-{"epoch": 0.05859375, "grad_norm": 1.934105396270752, "learning_rate": 5.6000000000000006e-05, "loss": 0.05385493487119675, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.05533, "step": 15, "tokens/total": 453376, "tokens/train_per_sec_per_gpu": 32.96, "tokens/trainable": 6820}
-{"epoch": 0.0625, "grad_norm": 1.4659016132354736, "learning_rate": 6e-05, "loss": 0.052153222262859344, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.05354, "step": 16, "tokens/total": 483504, "tokens/train_per_sec_per_gpu": 30.68, "tokens/trainable": 7258}
-{"epoch": 0.06640625, "grad_norm": 1.9650894403457642, "learning_rate": 6.400000000000001e-05, "loss": 0.05092189460992813, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.05224, "step": 17, "tokens/total": 514032, "tokens/train_per_sec_per_gpu": 35.09, "tokens/trainable": 7710}
-{"epoch": 0.0703125, "grad_norm": 0.6891724467277527, "learning_rate": 6.800000000000001e-05, "loss": 0.031155016273260117, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.03165, "step": 18, "tokens/total": 544432, "tokens/train_per_sec_per_gpu": 31.61, "tokens/trainable": 8174}
-{"epoch": 0.07421875, "grad_norm": 0.8662010431289673, "learning_rate": 7.2e-05, "loss": 0.03036138042807579, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.03083, "step": 19, "tokens/total": 574880, "tokens/train_per_sec_per_gpu": 34.37, "tokens/trainable": 8617}
-{"epoch": 0.078125, "grad_norm": 0.7999041080474854, "learning_rate": 7.6e-05, "loss": 0.03458942472934723, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.03519, "step": 20, "tokens/total": 605408, "tokens/train_per_sec_per_gpu": 30.32, "tokens/trainable": 9037}
-{"epoch": 0.08203125, "grad_norm": 0.5816919207572937, "learning_rate": 8e-05, "loss": 0.02401110902428627, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.0243, "step": 21, "tokens/total": 635584, "tokens/train_per_sec_per_gpu": 36.59, "tokens/trainable": 9503}
-{"epoch": 0.0859375, "grad_norm": 0.6761882901191711, "learning_rate": 8.4e-05, "loss": 0.01873329095542431, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01891, "step": 22, "tokens/total": 665952, "tokens/train_per_sec_per_gpu": 36.71, "tokens/trainable": 9972}
-{"epoch": 0.08984375, "grad_norm": 0.9077537655830383, "learning_rate": 8.800000000000001e-05, "loss": 0.017490077763795853, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01764, "step": 23, "tokens/total": 696240, "tokens/train_per_sec_per_gpu": 37.39, "tokens/trainable": 10447}
-{"epoch": 0.09375, "grad_norm": 1.3226462602615356, "learning_rate": 9.200000000000001e-05, "loss": 0.028416279703378677, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.02882, "step": 24, "tokens/total": 726464, "tokens/train_per_sec_per_gpu": 37.12, "tokens/trainable": 10899}
-{"epoch": 0.09765625, "grad_norm": 0.9712215065956116, "learning_rate": 9.6e-05, "loss": 0.025973526760935783, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02631, "step": 25, "tokens/total": 756704, "tokens/train_per_sec_per_gpu": 32.97, "tokens/trainable": 11360}
-{"epoch": 0.1015625, "grad_norm": 0.8638900518417358, "learning_rate": 0.0001, "loss": 0.022434517741203308, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.02269, "step": 26, "tokens/total": 786912, "tokens/train_per_sec_per_gpu": 29.23, "tokens/trainable": 11775}
-{"epoch": 0.10546875, "grad_norm": 0.6629000902175903, "learning_rate": 9.99990636831587e-05, "loss": 0.014538668096065521, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.46, "memory/max_allocated (GiB)": 33.46, "ppl": 1.01464, "step": 27, "tokens/total": 815424, "tokens/train_per_sec_per_gpu": 39.82, "tokens/trainable": 12242}
-{"epoch": 0.109375, "grad_norm": 0.3078136146068573, "learning_rate": 9.999625477159879e-05, "loss": 0.01195458322763443, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01203, "step": 28, "tokens/total": 845584, "tokens/train_per_sec_per_gpu": 33.52, "tokens/trainable": 12696}
-{"epoch": 0.11328125, "grad_norm": 1.1301876306533813, "learning_rate": 9.999157338221051e-05, "loss": 0.022538531571626663, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.02279, "step": 29, "tokens/total": 875808, "tokens/train_per_sec_per_gpu": 34.38, "tokens/trainable": 13153}
-{"epoch": 0.1171875, "grad_norm": 0.41090911626815796, "learning_rate": 9.998501970980562e-05, "loss": 0.011871461756527424, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01194, "step": 30, "tokens/total": 904096, "tokens/train_per_sec_per_gpu": 39.83, "tokens/trainable": 13624}
-{"epoch": 0.12109375, "grad_norm": 0.6772723197937012, "learning_rate": 9.997659402710915e-05, "loss": 0.013700846582651138, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0138, "step": 31, "tokens/total": 934400, "tokens/train_per_sec_per_gpu": 34.07, "tokens/trainable": 14064}
-{"epoch": 0.125, "grad_norm": 1.207811951637268, "learning_rate": 9.996629668474818e-05, "loss": 0.01185896061360836, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.01193, "step": 32, "tokens/total": 964832, "tokens/train_per_sec_per_gpu": 38.9, "tokens/trainable": 14565}
-{"epoch": 0.12890625, "grad_norm": 0.46513956785202026, "learning_rate": 9.995412811123711e-05, "loss": 0.012477721087634563, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.01256, "step": 33, "tokens/total": 995040, "tokens/train_per_sec_per_gpu": 33.76, "tokens/trainable": 15005}
-{"epoch": 0.1328125, "grad_norm": 1.7668761014938354, "learning_rate": 9.994008881295999e-05, "loss": 0.03285093232989311, "memory/device_reserved (GiB)": 34.72, "memory/max_active (GiB)": 33.64, "memory/max_allocated (GiB)": 33.64, "ppl": 1.0334, "step": 34, "tokens/total": 1024896, "tokens/train_per_sec_per_gpu": 32.05, "tokens/trainable": 15459}
-{"epoch": 0.13671875, "grad_norm": 0.7816088795661926, "learning_rate": 9.992417937414932e-05, "loss": 0.028746310621500015, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02916, "step": 35, "tokens/total": 1054992, "tokens/train_per_sec_per_gpu": 38.15, "tokens/trainable": 15960}
-{"epoch": 0.140625, "grad_norm": 0.683965265750885, "learning_rate": 9.99064004568618e-05, "loss": 0.020058901980519295, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.02026, "step": 36, "tokens/total": 1085280, "tokens/train_per_sec_per_gpu": 34.53, "tokens/trainable": 16428}
-{"epoch": 0.14453125, "grad_norm": 0.6797531843185425, "learning_rate": 9.988675280095074e-05, "loss": 0.010446591302752495, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.0105, "step": 37, "tokens/total": 1115504, "tokens/train_per_sec_per_gpu": 31.77, "tokens/trainable": 16884}
-{"epoch": 0.1484375, "grad_norm": 0.8899193406105042, "learning_rate": 9.986523722403528e-05, "loss": 0.017391683533787727, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01754, "step": 38, "tokens/total": 1145712, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 17341}
-{"epoch": 0.15234375, "grad_norm": 0.8132575750350952, "learning_rate": 9.984185462146642e-05, "loss": 0.02252483367919922, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.02278, "step": 39, "tokens/total": 1176128, "tokens/train_per_sec_per_gpu": 34.15, "tokens/trainable": 17786}
-{"epoch": 0.15625, "grad_norm": 0.4430502653121948, "learning_rate": 9.98166059662897e-05, "loss": 0.011966042220592499, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01204, "step": 40, "tokens/total": 1206576, "tokens/train_per_sec_per_gpu": 34.99, "tokens/trainable": 18262}
-{"epoch": 0.16015625, "grad_norm": 0.5074653029441833, "learning_rate": 9.978949230920472e-05, "loss": 0.014877484180033207, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01499, "step": 41, "tokens/total": 1236848, "tokens/train_per_sec_per_gpu": 32.7, "tokens/trainable": 18716}
-{"epoch": 0.1640625, "grad_norm": 0.5221464037895203, "learning_rate": 9.976051477852141e-05, "loss": 0.017510797828435898, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01767, "step": 42, "tokens/total": 1267088, "tokens/train_per_sec_per_gpu": 32.87, "tokens/trainable": 19157}
-{"epoch": 0.16796875, "grad_norm": 0.6888790130615234, "learning_rate": 9.972967458011312e-05, "loss": 0.030433066189289093, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0309, "step": 43, "tokens/total": 1297360, "tokens/train_per_sec_per_gpu": 36.5, "tokens/trainable": 19647}
-{"epoch": 0.171875, "grad_norm": 0.5600388050079346, "learning_rate": 9.96969729973664e-05, "loss": 0.013720017857849598, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01381, "step": 44, "tokens/total": 1327744, "tokens/train_per_sec_per_gpu": 34.9, "tokens/trainable": 20119}
-{"epoch": 0.17578125, "grad_norm": 0.4291926324367523, "learning_rate": 9.966241139112754e-05, "loss": 0.00872582383453846, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00876, "step": 45, "tokens/total": 1358096, "tokens/train_per_sec_per_gpu": 32.92, "tokens/trainable": 20560}
-{"epoch": 0.1796875, "grad_norm": 0.944327712059021, "learning_rate": 9.96259911996461e-05, "loss": 0.025248996913433075, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02557, "step": 46, "tokens/total": 1388240, "tokens/train_per_sec_per_gpu": 32.59, "tokens/trainable": 20992}
-{"epoch": 0.18359375, "grad_norm": 0.2425016313791275, "learning_rate": 9.958771393851491e-05, "loss": 0.005067020654678345, "memory/device_reserved (GiB)": 35.87, "memory/max_active (GiB)": 33.26, "memory/max_allocated (GiB)": 33.26, "ppl": 1.00508, "step": 47, "tokens/total": 1416240, "tokens/train_per_sec_per_gpu": 31.51, "tokens/trainable": 21441}
-{"epoch": 0.1875, "grad_norm": 1.2363684177398682, "learning_rate": 9.954758120060702e-05, "loss": 0.03300227224826813, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.03355, "step": 48, "tokens/total": 1446880, "tokens/train_per_sec_per_gpu": 33.7, "tokens/trainable": 21901}
-{"epoch": 0.19140625, "grad_norm": 0.7278413772583008, "learning_rate": 9.950559465600948e-05, "loss": 0.025975672528147697, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.02632, "step": 49, "tokens/total": 1477168, "tokens/train_per_sec_per_gpu": 33.27, "tokens/trainable": 22341}
-{"epoch": 0.1953125, "grad_norm": 0.37237733602523804, "learning_rate": 9.946175605195379e-05, "loss": 0.010315775871276855, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01037, "step": 50, "tokens/total": 1507712, "tokens/train_per_sec_per_gpu": 33.07, "tokens/trainable": 22833}
-{"epoch": 0.19921875, "grad_norm": 0.25258293747901917, "learning_rate": 9.941606721274322e-05, "loss": 0.00485712755471468, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00487, "step": 51, "tokens/total": 1537936, "tokens/train_per_sec_per_gpu": 30.64, "tokens/trainable": 23277}
-{"epoch": 0.203125, "grad_norm": 0.738599419593811, "learning_rate": 9.936853003967685e-05, "loss": 0.01646406576037407, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0166, "step": 52, "tokens/total": 1568352, "tokens/train_per_sec_per_gpu": 35.57, "tokens/trainable": 23759}
-{"epoch": 0.20703125, "grad_norm": 1.2733500003814697, "learning_rate": 9.93191465109705e-05, "loss": 0.019196398556232452, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01938, "step": 53, "tokens/total": 1598992, "tokens/train_per_sec_per_gpu": 32.98, "tokens/trainable": 24193}
-{"epoch": 0.2109375, "grad_norm": 0.5316427946090698, "learning_rate": 9.926791868167438e-05, "loss": 0.006890955846756697, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00691, "step": 54, "tokens/total": 1629488, "tokens/train_per_sec_per_gpu": 33.47, "tokens/trainable": 24619}
-{"epoch": 0.21484375, "grad_norm": 0.6924111843109131, "learning_rate": 9.921484868358753e-05, "loss": 0.021178584545850754, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0214, "step": 55, "tokens/total": 1659696, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 25075}
-{"epoch": 0.21875, "grad_norm": 0.683601975440979, "learning_rate": 9.915993872516924e-05, "loss": 0.017589068040251732, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.01774, "step": 56, "tokens/total": 1689872, "tokens/train_per_sec_per_gpu": 31.48, "tokens/trainable": 25519}
-{"epoch": 0.22265625, "grad_norm": 0.8593301177024841, "learning_rate": 9.9103191091447e-05, "loss": 0.025561584159731865, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.02589, "step": 57, "tokens/total": 1720144, "tokens/train_per_sec_per_gpu": 32.63, "tokens/trainable": 25966}
-{"epoch": 0.2265625, "grad_norm": 0.2925783097743988, "learning_rate": 9.904460814392147e-05, "loss": 0.005790311843156815, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00581, "step": 58, "tokens/total": 1750448, "tokens/train_per_sec_per_gpu": 33.89, "tokens/trainable": 26423}
-{"epoch": 0.23046875, "grad_norm": 0.6059477925300598, "learning_rate": 9.898419232046825e-05, "loss": 0.007334758993238211, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00736, "step": 59, "tokens/total": 1781088, "tokens/train_per_sec_per_gpu": 38.42, "tokens/trainable": 26934}
-{"epoch": 0.234375, "grad_norm": 0.29425033926963806, "learning_rate": 9.892194613523633e-05, "loss": 0.004620288498699665, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00463, "step": 60, "tokens/total": 1811344, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 27393}
-{"epoch": 0.23828125, "grad_norm": 0.25868093967437744, "learning_rate": 9.885787217854357e-05, "loss": 0.0042824773117899895, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00429, "step": 61, "tokens/total": 1841600, "tokens/train_per_sec_per_gpu": 32.84, "tokens/trainable": 27852}
-{"epoch": 0.2421875, "grad_norm": 0.9455181360244751, "learning_rate": 9.879197311676887e-05, "loss": 0.013508133590221405, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0136, "step": 62, "tokens/total": 1872048, "tokens/train_per_sec_per_gpu": 32.92, "tokens/trainable": 28292}
-{"epoch": 0.24609375, "grad_norm": 1.149442434310913, "learning_rate": 9.872425169224113e-05, "loss": 0.020864056423306465, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.02108, "step": 63, "tokens/total": 1902592, "tokens/train_per_sec_per_gpu": 37.27, "tokens/trainable": 28798}
-{"epoch": 0.25, "grad_norm": 0.7421550750732422, "learning_rate": 9.865471072312528e-05, "loss": 0.016041038557887077, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01617, "step": 64, "tokens/total": 1933088, "tokens/train_per_sec_per_gpu": 35.76, "tokens/trainable": 29283}
-{"epoch": 0.25390625, "grad_norm": 0.36109936237335205, "learning_rate": 9.858335310330492e-05, "loss": 0.005372575484216213, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00539, "step": 65, "tokens/total": 1963296, "tokens/train_per_sec_per_gpu": 31.92, "tokens/trainable": 29745}
-{"epoch": 0.2578125, "grad_norm": 0.7074101567268372, "learning_rate": 9.851018180226185e-05, "loss": 0.018492329865694046, "memory/device_reserved (GiB)": 34.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.01866, "step": 66, "tokens/total": 1993760, "tokens/train_per_sec_per_gpu": 31.19, "tokens/trainable": 30186}
-{"epoch": 0.26171875, "grad_norm": 0.43113431334495544, "learning_rate": 9.843519986495259e-05, "loss": 0.00620780885219574, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00623, "step": 67, "tokens/total": 2024144, "tokens/train_per_sec_per_gpu": 33.22, "tokens/trainable": 30622}
-{"epoch": 0.265625, "grad_norm": 0.3996214270591736, "learning_rate": 9.835841041168162e-05, "loss": 0.005429963115602732, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00544, "step": 68, "tokens/total": 2054608, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 31097}
-{"epoch": 0.26953125, "grad_norm": 0.4444175660610199, "learning_rate": 9.82798166379715e-05, "loss": 0.01158289983868599, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01165, "step": 69, "tokens/total": 2084912, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 31595}
-{"epoch": 0.2734375, "grad_norm": 0.16977320611476898, "learning_rate": 9.819942181443002e-05, "loss": 0.002158569637686014, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00216, "step": 70, "tokens/total": 2115216, "tokens/train_per_sec_per_gpu": 30.67, "tokens/trainable": 32036}
-{"epoch": 0.27734375, "grad_norm": 0.12970401346683502, "learning_rate": 9.811722928661392e-05, "loss": 0.0028989531565457582, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0029, "step": 71, "tokens/total": 2145424, "tokens/train_per_sec_per_gpu": 28.06, "tokens/trainable": 32428}
-{"epoch": 0.28125, "grad_norm": 0.06490105390548706, "learning_rate": 9.803324247488975e-05, "loss": 0.0008802044321782887, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00088, "step": 72, "tokens/total": 2175648, "tokens/train_per_sec_per_gpu": 32.08, "tokens/trainable": 32880}
-{"epoch": 0.28515625, "grad_norm": 0.20680083334445953, "learning_rate": 9.794746487429161e-05, "loss": 0.0019504247466102242, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00195, "step": 73, "tokens/total": 2205856, "tokens/train_per_sec_per_gpu": 33.48, "tokens/trainable": 33323}
-{"epoch": 0.2890625, "grad_norm": 1.6840267181396484, "learning_rate": 9.785990005437554e-05, "loss": 0.02435958757996559, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.02466, "step": 74, "tokens/total": 2236352, "tokens/train_per_sec_per_gpu": 35.84, "tokens/trainable": 33792}
-{"epoch": 0.29296875, "grad_norm": 0.6665006875991821, "learning_rate": 9.777055165907117e-05, "loss": 0.018271014094352722, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.01844, "step": 75, "tokens/total": 2266480, "tokens/train_per_sec_per_gpu": 35.24, "tokens/trainable": 34223}
-{"epoch": 0.296875, "grad_norm": 0.6469210982322693, "learning_rate": 9.767942340652993e-05, "loss": 0.023723380640149117, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.02401, "step": 76, "tokens/total": 2296496, "tokens/train_per_sec_per_gpu": 29.59, "tokens/trainable": 34649}
-{"epoch": 0.30078125, "grad_norm": 1.391882300376892, "learning_rate": 9.758651908897035e-05, "loss": 0.015201358124613762, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01532, "step": 77, "tokens/total": 2327040, "tokens/train_per_sec_per_gpu": 35.58, "tokens/trainable": 35094}
-{"epoch": 0.3046875, "grad_norm": 0.5752553343772888, "learning_rate": 9.749184257252033e-05, "loss": 0.020247019827365875, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.02045, "step": 78, "tokens/total": 2357328, "tokens/train_per_sec_per_gpu": 31.79, "tokens/trainable": 35534}
-{"epoch": 0.30859375, "grad_norm": 0.276022732257843, "learning_rate": 9.739539779705614e-05, "loss": 0.010471204295754433, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01053, "step": 79, "tokens/total": 2387664, "tokens/train_per_sec_per_gpu": 35.76, "tokens/trainable": 36029}
-{"epoch": 0.3125, "grad_norm": 0.41784003376960754, "learning_rate": 9.729718877603861e-05, "loss": 0.010774495080113411, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01083, "step": 80, "tokens/total": 2418000, "tokens/train_per_sec_per_gpu": 35.55, "tokens/trainable": 36469}
-{"epoch": 0.31640625, "grad_norm": 0.4906325340270996, "learning_rate": 9.719721959634592e-05, "loss": 0.015422273427248001, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01554, "step": 81, "tokens/total": 2448720, "tokens/train_per_sec_per_gpu": 30.69, "tokens/trainable": 36906}
-{"epoch": 0.3203125, "grad_norm": 0.2813898026943207, "learning_rate": 9.709549441810375e-05, "loss": 0.009146124124526978, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00919, "step": 82, "tokens/total": 2479248, "tokens/train_per_sec_per_gpu": 38.89, "tokens/trainable": 37390}
-{"epoch": 0.32421875, "grad_norm": 0.39496278762817383, "learning_rate": 9.699201747451195e-05, "loss": 0.008894146420061588, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00893, "step": 83, "tokens/total": 2509408, "tokens/train_per_sec_per_gpu": 30.78, "tokens/trainable": 37838}
-{"epoch": 0.328125, "grad_norm": 0.4946168065071106, "learning_rate": 9.688679307166854e-05, "loss": 0.005293373484164476, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00531, "step": 84, "tokens/total": 2539776, "tokens/train_per_sec_per_gpu": 34.98, "tokens/trainable": 38310}
-{"epoch": 0.33203125, "grad_norm": 1.3293001651763916, "learning_rate": 9.677982558839042e-05, "loss": 0.040361277759075165, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.04119, "step": 85, "tokens/total": 2569840, "tokens/train_per_sec_per_gpu": 34.0, "tokens/trainable": 38789}
-{"epoch": 0.3359375, "grad_norm": 0.5834341049194336, "learning_rate": 9.66711194760312e-05, "loss": 0.010128681547939777, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01018, "step": 86, "tokens/total": 2600192, "tokens/train_per_sec_per_gpu": 36.34, "tokens/trainable": 39277}
-{"epoch": 0.33984375, "grad_norm": 0.7191532254219055, "learning_rate": 9.656067925829593e-05, "loss": 0.02042745053768158, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.02064, "step": 87, "tokens/total": 2630640, "tokens/train_per_sec_per_gpu": 35.6, "tokens/trainable": 39737}
-{"epoch": 0.34375, "grad_norm": 0.3419296145439148, "learning_rate": 9.644850953105288e-05, "loss": 0.007800046354532242, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00783, "step": 88, "tokens/total": 2660832, "tokens/train_per_sec_per_gpu": 29.53, "tokens/trainable": 40179}
-{"epoch": 0.34765625, "grad_norm": 0.23275785148143768, "learning_rate": 9.633461496214225e-05, "loss": 0.005660225171595812, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00568, "step": 89, "tokens/total": 2691328, "tokens/train_per_sec_per_gpu": 32.59, "tokens/trainable": 40649}
-{"epoch": 0.3515625, "grad_norm": 0.6987941265106201, "learning_rate": 9.621900029118195e-05, "loss": 0.02007928490638733, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.41, "memory/max_allocated (GiB)": 33.41, "ppl": 1.02028, "step": 90, "tokens/total": 2719696, "tokens/train_per_sec_per_gpu": 31.52, "tokens/trainable": 41080}
-{"epoch": 0.35546875, "grad_norm": 0.11328475177288055, "learning_rate": 9.610167032937036e-05, "loss": 0.002234571846202016, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00224, "step": 91, "tokens/total": 2750272, "tokens/train_per_sec_per_gpu": 37.99, "tokens/trainable": 41599}
-{"epoch": 0.359375, "grad_norm": 0.4347783029079437, "learning_rate": 9.598262995928611e-05, "loss": 0.004549161531031132, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00456, "step": 92, "tokens/total": 2780656, "tokens/train_per_sec_per_gpu": 36.77, "tokens/trainable": 42076}
-{"epoch": 0.36328125, "grad_norm": 0.3486956059932709, "learning_rate": 9.586188413468492e-05, "loss": 0.012267105281352997, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01234, "step": 93, "tokens/total": 2811088, "tokens/train_per_sec_per_gpu": 33.6, "tokens/trainable": 42522}
-{"epoch": 0.3671875, "grad_norm": 0.5156503319740295, "learning_rate": 9.57394378802934e-05, "loss": 0.015529165044426918, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01565, "step": 94, "tokens/total": 2841520, "tokens/train_per_sec_per_gpu": 34.32, "tokens/trainable": 42974}
-{"epoch": 0.37109375, "grad_norm": 0.37648338079452515, "learning_rate": 9.56152962916e-05, "loss": 0.011707151308655739, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.01178, "step": 95, "tokens/total": 2871600, "tokens/train_per_sec_per_gpu": 30.71, "tokens/trainable": 43380}
-{"epoch": 0.375, "grad_norm": 0.38365671038627625, "learning_rate": 9.548946453464296e-05, "loss": 0.008411398157477379, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00845, "step": 96, "tokens/total": 2902144, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 43865}
-{"epoch": 0.37890625, "grad_norm": 0.313085675239563, "learning_rate": 9.53619478457953e-05, "loss": 0.007852522656321526, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00788, "step": 97, "tokens/total": 2932272, "tokens/train_per_sec_per_gpu": 31.89, "tokens/trainable": 44328}
-{"epoch": 0.3828125, "grad_norm": 0.08544483780860901, "learning_rate": 9.523275153154695e-05, "loss": 0.0025753644295036793, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00258, "step": 98, "tokens/total": 2962032, "tokens/train_per_sec_per_gpu": 30.98, "tokens/trainable": 44778}
-{"epoch": 0.38671875, "grad_norm": 0.6496388912200928, "learning_rate": 9.51018809682839e-05, "loss": 0.02547256276011467, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0258, "step": 99, "tokens/total": 2992256, "tokens/train_per_sec_per_gpu": 37.11, "tokens/trainable": 45225}
-{"epoch": 0.390625, "grad_norm": 0.15325438976287842, "learning_rate": 9.49693416020645e-05, "loss": 0.003552855923771858, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00356, "step": 100, "tokens/total": 3022608, "tokens/train_per_sec_per_gpu": 35.8, "tokens/trainable": 45678}
-{"epoch": 0.39453125, "grad_norm": 0.25307565927505493, "learning_rate": 9.483513894839276e-05, "loss": 0.004095804411917925, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0041, "step": 101, "tokens/total": 3052992, "tokens/train_per_sec_per_gpu": 32.73, "tokens/trainable": 46125}
-{"epoch": 0.3984375, "grad_norm": 0.150072380900383, "learning_rate": 9.469927859198888e-05, "loss": 0.0013888315297663212, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00139, "step": 102, "tokens/total": 3083392, "tokens/train_per_sec_per_gpu": 39.71, "tokens/trainable": 46612}
-{"epoch": 0.40234375, "grad_norm": 0.5769571661949158, "learning_rate": 9.456176618655689e-05, "loss": 0.022533349692821503, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.02279, "step": 103, "tokens/total": 3113744, "tokens/train_per_sec_per_gpu": 32.57, "tokens/trainable": 47059}
-{"epoch": 0.40625, "grad_norm": 0.5657027959823608, "learning_rate": 9.442260745454927e-05, "loss": 0.016894506290555, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.01704, "step": 104, "tokens/total": 3144272, "tokens/train_per_sec_per_gpu": 34.05, "tokens/trainable": 47536}
-{"epoch": 0.41015625, "grad_norm": 0.29607170820236206, "learning_rate": 9.428180818692884e-05, "loss": 0.017974723130464554, "memory/device_reserved (GiB)": 37.61, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.01814, "step": 105, "tokens/total": 3174512, "tokens/train_per_sec_per_gpu": 34.11, "tokens/trainable": 48037}
-{"epoch": 0.4140625, "grad_norm": 0.5241922736167908, "learning_rate": 9.413937424292791e-05, "loss": 0.016189826652407646, "memory/device_reserved (GiB)": 37.61, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01632, "step": 106, "tokens/total": 3204896, "tokens/train_per_sec_per_gpu": 32.65, "tokens/trainable": 48491}
-{"epoch": 0.41796875, "grad_norm": 0.3886408805847168, "learning_rate": 9.399531154980424e-05, "loss": 0.010908558964729309, "memory/device_reserved (GiB)": 37.61, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.01097, "step": 107, "tokens/total": 3235360, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 48940}
-{"epoch": 0.421875, "grad_norm": 0.2442784607410431, "learning_rate": 9.384962610259455e-05, "loss": 0.008070861920714378, "memory/device_reserved (GiB)": 37.61, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0081, "step": 108, "tokens/total": 3265552, "tokens/train_per_sec_per_gpu": 31.79, "tokens/trainable": 49389}
-{"epoch": 0.42578125, "grad_norm": 0.1457175612449646, "learning_rate": 9.370232396386494e-05, "loss": 0.003785747569054365, "memory/device_reserved (GiB)": 37.61, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.00379, "step": 109, "tokens/total": 3293856, "tokens/train_per_sec_per_gpu": 36.01, "tokens/trainable": 49838}
-{"epoch": 0.4296875, "grad_norm": 0.3955559730529785, "learning_rate": 9.355341126345868e-05, "loss": 0.0069918157532811165, "memory/device_reserved (GiB)": 37.61, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00702, "step": 110, "tokens/total": 3323984, "tokens/train_per_sec_per_gpu": 34.06, "tokens/trainable": 50310}
-{"epoch": 0.43359375, "grad_norm": 0.6224751472473145, "learning_rate": 9.340289419824107e-05, "loss": 0.014852076768875122, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01496, "step": 111, "tokens/total": 3354320, "tokens/train_per_sec_per_gpu": 33.67, "tokens/trainable": 50755}
-{"epoch": 0.4375, "grad_norm": 0.3700723648071289, "learning_rate": 9.325077903184159e-05, "loss": 0.00436755083501339, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00438, "step": 112, "tokens/total": 3384880, "tokens/train_per_sec_per_gpu": 31.65, "tokens/trainable": 51213}
-{"epoch": 0.44140625, "grad_norm": 0.1353236883878708, "learning_rate": 9.30970720943932e-05, "loss": 0.0025976570323109627, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0026, "step": 113, "tokens/total": 3415104, "tokens/train_per_sec_per_gpu": 33.83, "tokens/trainable": 51660}
-{"epoch": 0.4453125, "grad_norm": 0.18984447419643402, "learning_rate": 9.2941779782269e-05, "loss": 0.002497302368283272, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0025, "step": 114, "tokens/total": 3445504, "tokens/train_per_sec_per_gpu": 32.43, "tokens/trainable": 52133}
-{"epoch": 0.44921875, "grad_norm": 1.1815483570098877, "learning_rate": 9.278490855781596e-05, "loss": 0.029489168897271156, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02993, "step": 115, "tokens/total": 3475744, "tokens/train_per_sec_per_gpu": 37.47, "tokens/trainable": 52580}
-{"epoch": 0.453125, "grad_norm": 0.44360846281051636, "learning_rate": 9.262646494908604e-05, "loss": 0.009585533291101456, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00963, "step": 116, "tokens/total": 3506016, "tokens/train_per_sec_per_gpu": 34.83, "tokens/trainable": 53033}
-{"epoch": 0.45703125, "grad_norm": 0.5074551701545715, "learning_rate": 9.246645554956457e-05, "loss": 0.020201388746500015, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.02041, "step": 117, "tokens/total": 3536256, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 53517}
-{"epoch": 0.4609375, "grad_norm": 0.3565296232700348, "learning_rate": 9.230488701789578e-05, "loss": 0.005688562989234924, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.0057, "step": 118, "tokens/total": 3566480, "tokens/train_per_sec_per_gpu": 34.56, "tokens/trainable": 53967}
-{"epoch": 0.46484375, "grad_norm": 0.16547706723213196, "learning_rate": 9.214176607760577e-05, "loss": 0.003188834059983492, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00319, "step": 119, "tokens/total": 3596624, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 54430}
-{"epoch": 0.46875, "grad_norm": 0.20722293853759766, "learning_rate": 9.197709951682268e-05, "loss": 0.003235103562474251, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00324, "step": 120, "tokens/total": 3627168, "tokens/train_per_sec_per_gpu": 35.21, "tokens/trainable": 54896}
-{"epoch": 0.47265625, "grad_norm": 0.4754939377307892, "learning_rate": 9.181089418799428e-05, "loss": 0.005670893006026745, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00569, "step": 121, "tokens/total": 3657632, "tokens/train_per_sec_per_gpu": 37.77, "tokens/trainable": 55379}
-{"epoch": 0.4765625, "grad_norm": 0.08304762095212936, "learning_rate": 9.164315700760271e-05, "loss": 0.00099027412943542, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00099, "step": 122, "tokens/total": 3687824, "tokens/train_per_sec_per_gpu": 31.7, "tokens/trainable": 55803}
-{"epoch": 0.48046875, "grad_norm": 0.725281298160553, "learning_rate": 9.147389495587671e-05, "loss": 0.007413266692310572, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00744, "step": 123, "tokens/total": 3718320, "tokens/train_per_sec_per_gpu": 36.45, "tokens/trainable": 56249}
-{"epoch": 0.484375, "grad_norm": 0.31409645080566406, "learning_rate": 9.130311507650116e-05, "loss": 0.0029702766332775354, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00297, "step": 124, "tokens/total": 3748672, "tokens/train_per_sec_per_gpu": 32.41, "tokens/trainable": 56713}
-{"epoch": 0.48828125, "grad_norm": 0.6082578897476196, "learning_rate": 9.113082447632394e-05, "loss": 0.003795543685555458, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0038, "step": 125, "tokens/total": 3778976, "tokens/train_per_sec_per_gpu": 39.08, "tokens/trainable": 57196}
-{"epoch": 0.4921875, "grad_norm": 0.0603976845741272, "learning_rate": 9.09570303250602e-05, "loss": 0.0014751165872439742, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00148, "step": 126, "tokens/total": 3809296, "tokens/train_per_sec_per_gpu": 36.49, "tokens/trainable": 57680}
-{"epoch": 0.49609375, "grad_norm": 0.21112751960754395, "learning_rate": 9.078173985499394e-05, "loss": 0.004137102514505386, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00415, "step": 127, "tokens/total": 3839328, "tokens/train_per_sec_per_gpu": 31.8, "tokens/trainable": 58110}
-{"epoch": 0.5, "grad_norm": 0.3234494924545288, "learning_rate": 9.060496036067713e-05, "loss": 0.007372056134045124, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0074, "step": 128, "tokens/total": 3869824, "tokens/train_per_sec_per_gpu": 31.64, "tokens/trainable": 58534}
-{"epoch": 0.50390625, "grad_norm": 0.5826171636581421, "learning_rate": 9.042669919862615e-05, "loss": 0.007980267517268658, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00801, "step": 129, "tokens/total": 3900080, "tokens/train_per_sec_per_gpu": 35.08, "tokens/trainable": 58998}
-{"epoch": 0.5078125, "grad_norm": 0.3384500741958618, "learning_rate": 9.024696378701557e-05, "loss": 0.005637750029563904, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00565, "step": 130, "tokens/total": 3930560, "tokens/train_per_sec_per_gpu": 31.81, "tokens/trainable": 59448}
-{"epoch": 0.51171875, "grad_norm": 0.2838669717311859, "learning_rate": 9.006576160536948e-05, "loss": 0.0037927688099443913, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.0038, "step": 131, "tokens/total": 3960496, "tokens/train_per_sec_per_gpu": 31.97, "tokens/trainable": 59906}
-{"epoch": 0.515625, "grad_norm": 0.4598330855369568, "learning_rate": 8.988310019425035e-05, "loss": 0.010232537053525448, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01029, "step": 132, "tokens/total": 3990928, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 60350}
-{"epoch": 0.51953125, "grad_norm": 0.7072780728340149, "learning_rate": 8.969898715494506e-05, "loss": 0.00946755986660719, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00951, "step": 133, "tokens/total": 4021264, "tokens/train_per_sec_per_gpu": 36.18, "tokens/trainable": 60831}
-{"epoch": 0.5234375, "grad_norm": 0.3642464280128479, "learning_rate": 8.951343014914869e-05, "loss": 0.0067742373794317245, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0068, "step": 134, "tokens/total": 4051728, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 61305}
-{"epoch": 0.52734375, "grad_norm": 0.1071263924241066, "learning_rate": 8.932643689864568e-05, "loss": 0.0022848297376185656, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00229, "step": 135, "tokens/total": 4081920, "tokens/train_per_sec_per_gpu": 37.57, "tokens/trainable": 61792}
-{"epoch": 0.53125, "grad_norm": 0.22470054030418396, "learning_rate": 8.913801518498845e-05, "loss": 0.0016683072317391634, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00167, "step": 136, "tokens/total": 4112304, "tokens/train_per_sec_per_gpu": 31.33, "tokens/trainable": 62253}
-{"epoch": 0.53515625, "grad_norm": 0.5423188209533691, "learning_rate": 8.894817284917364e-05, "loss": 0.017443198710680008, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.0176, "step": 137, "tokens/total": 4142512, "tokens/train_per_sec_per_gpu": 31.57, "tokens/trainable": 62707}
-{"epoch": 0.5390625, "grad_norm": 0.47486332058906555, "learning_rate": 8.875691779131569e-05, "loss": 0.01525629311800003, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01537, "step": 138, "tokens/total": 4172992, "tokens/train_per_sec_per_gpu": 29.32, "tokens/trainable": 63145}
-{"epoch": 0.54296875, "grad_norm": 0.055354099720716476, "learning_rate": 8.856425797031829e-05, "loss": 0.0010598574299365282, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00106, "step": 139, "tokens/total": 4203136, "tokens/train_per_sec_per_gpu": 33.87, "tokens/trainable": 63587}
-{"epoch": 0.546875, "grad_norm": 0.15273842215538025, "learning_rate": 8.837020140354295e-05, "loss": 0.002772743348032236, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00278, "step": 140, "tokens/total": 4233456, "tokens/train_per_sec_per_gpu": 36.75, "tokens/trainable": 64052}
-{"epoch": 0.55078125, "grad_norm": 0.21690180897712708, "learning_rate": 8.817475616647554e-05, "loss": 0.005170213058590889, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00518, "step": 141, "tokens/total": 4263728, "tokens/train_per_sec_per_gpu": 34.36, "tokens/trainable": 64526}
-{"epoch": 0.5546875, "grad_norm": 0.1491464525461197, "learning_rate": 8.797793039239017e-05, "loss": 0.0031757252290844917, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.00318, "step": 142, "tokens/total": 4294432, "tokens/train_per_sec_per_gpu": 34.66, "tokens/trainable": 64983}
-{"epoch": 0.55859375, "grad_norm": 0.19370807707309723, "learning_rate": 8.777973227201069e-05, "loss": 0.0033013438805937767, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00331, "step": 143, "tokens/total": 4324960, "tokens/train_per_sec_per_gpu": 37.6, "tokens/trainable": 65492}
-{"epoch": 0.5625, "grad_norm": 0.43046337366104126, "learning_rate": 8.758017005316988e-05, "loss": 0.004675615578889847, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00469, "step": 144, "tokens/total": 4355424, "tokens/train_per_sec_per_gpu": 35.71, "tokens/trainable": 65925}
-{"epoch": 0.56640625, "grad_norm": 1.153432011604309, "learning_rate": 8.737925204046629e-05, "loss": 0.00530653540045023, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00532, "step": 145, "tokens/total": 4385712, "tokens/train_per_sec_per_gpu": 31.24, "tokens/trainable": 66383}
-{"epoch": 0.5703125, "grad_norm": 0.7740430235862732, "learning_rate": 8.717698659491851e-05, "loss": 0.01281517744064331, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.0129, "step": 146, "tokens/total": 4416016, "tokens/train_per_sec_per_gpu": 31.37, "tokens/trainable": 66840}
-{"epoch": 0.57421875, "grad_norm": 0.6978983879089355, "learning_rate": 8.697338213361735e-05, "loss": 0.0272100567817688, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.02758, "step": 147, "tokens/total": 4446368, "tokens/train_per_sec_per_gpu": 31.71, "tokens/trainable": 67297}
-{"epoch": 0.578125, "grad_norm": 0.17344872653484344, "learning_rate": 8.676844712937552e-05, "loss": 0.008015683852136135, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00805, "step": 148, "tokens/total": 4476880, "tokens/train_per_sec_per_gpu": 37.36, "tokens/trainable": 67769}
-{"epoch": 0.58203125, "grad_norm": 0.6355595588684082, "learning_rate": 8.656219011037509e-05, "loss": 0.010829147882759571, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01089, "step": 149, "tokens/total": 4507232, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 68257}
-{"epoch": 0.5859375, "grad_norm": 0.25094935297966003, "learning_rate": 8.63546196598125e-05, "loss": 0.0052955676801502705, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00531, "step": 150, "tokens/total": 4537376, "tokens/train_per_sec_per_gpu": 35.51, "tokens/trainable": 68706}
-{"epoch": 0.58984375, "grad_norm": 0.5292705297470093, "learning_rate": 8.614574441554145e-05, "loss": 0.022014575079083443, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.02226, "step": 151, "tokens/total": 4567584, "tokens/train_per_sec_per_gpu": 33.25, "tokens/trainable": 69131}
-{"epoch": 0.59375, "grad_norm": 0.3471219539642334, "learning_rate": 8.593557306971349e-05, "loss": 0.024585288017988205, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.02489, "step": 152, "tokens/total": 4597792, "tokens/train_per_sec_per_gpu": 31.92, "tokens/trainable": 69586}
-{"epoch": 0.59765625, "grad_norm": 0.08056659996509552, "learning_rate": 8.572411436841618e-05, "loss": 0.002611964475363493, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00262, "step": 153, "tokens/total": 4627936, "tokens/train_per_sec_per_gpu": 32.81, "tokens/trainable": 70061}
-{"epoch": 0.6015625, "grad_norm": 0.162270650267601, "learning_rate": 8.551137711130922e-05, "loss": 0.0033612053375691175, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00337, "step": 154, "tokens/total": 4658352, "tokens/train_per_sec_per_gpu": 27.93, "tokens/trainable": 70467}
-{"epoch": 0.60546875, "grad_norm": 0.17613235116004944, "learning_rate": 8.529737015125824e-05, "loss": 0.004349880386143923, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00436, "step": 155, "tokens/total": 4688896, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 70933}
-{"epoch": 0.609375, "grad_norm": 0.2590649127960205, "learning_rate": 8.508210239396639e-05, "loss": 0.010615494102239609, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.01067, "step": 156, "tokens/total": 4719168, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 71382}
-{"epoch": 0.61328125, "grad_norm": 0.15640626847743988, "learning_rate": 8.486558279760375e-05, "loss": 0.002627612790092826, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00263, "step": 157, "tokens/total": 4749136, "tokens/train_per_sec_per_gpu": 30.22, "tokens/trainable": 71808}
-{"epoch": 0.6171875, "grad_norm": 0.34429433941841125, "learning_rate": 8.464782037243449e-05, "loss": 0.010873161256313324, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.01093, "step": 158, "tokens/total": 4779472, "tokens/train_per_sec_per_gpu": 32.87, "tokens/trainable": 72249}
-{"epoch": 0.62109375, "grad_norm": 0.3858713209629059, "learning_rate": 8.442882418044202e-05, "loss": 0.020050909370183945, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.02025, "step": 159, "tokens/total": 4809632, "tokens/train_per_sec_per_gpu": 35.19, "tokens/trainable": 72726}
-{"epoch": 0.625, "grad_norm": 0.3002466857433319, "learning_rate": 8.420860333495179e-05, "loss": 0.009756345301866531, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.0098, "step": 160, "tokens/total": 4840112, "tokens/train_per_sec_per_gpu": 32.73, "tokens/trainable": 73148}
-{"epoch": 0.62890625, "grad_norm": 0.202926903963089, "learning_rate": 8.398716700025208e-05, "loss": 0.009084527380764484, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 34.04, "memory/max_allocated (GiB)": 34.04, "ppl": 1.00913, "step": 161, "tokens/total": 4870656, "tokens/train_per_sec_per_gpu": 31.46, "tokens/trainable": 73600}
-{"epoch": 0.6328125, "grad_norm": 0.29608848690986633, "learning_rate": 8.376452439121266e-05, "loss": 0.0125912856310606, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.01267, "step": 162, "tokens/total": 4900608, "tokens/train_per_sec_per_gpu": 35.71, "tokens/trainable": 74068}
-{"epoch": 0.63671875, "grad_norm": 0.11453798413276672, "learning_rate": 8.354068477290124e-05, "loss": 0.0031142355874180794, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00312, "step": 163, "tokens/total": 4930752, "tokens/train_per_sec_per_gpu": 35.71, "tokens/trainable": 74527}
-{"epoch": 0.640625, "grad_norm": 0.12609371542930603, "learning_rate": 8.331565746019807e-05, "loss": 0.0056648110039532185, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00568, "step": 164, "tokens/total": 4961008, "tokens/train_per_sec_per_gpu": 31.81, "tokens/trainable": 74992}
-{"epoch": 0.64453125, "grad_norm": 0.49591395258903503, "learning_rate": 8.308945181740812e-05, "loss": 0.012539982795715332, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01262, "step": 165, "tokens/total": 4991200, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 75442}
-{"epoch": 0.6484375, "grad_norm": 0.17120927572250366, "learning_rate": 8.286207725787153e-05, "loss": 0.007677854970097542, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00771, "step": 166, "tokens/total": 5021600, "tokens/train_per_sec_per_gpu": 31.19, "tokens/trainable": 75887}
-{"epoch": 0.65234375, "grad_norm": 0.19032779335975647, "learning_rate": 8.263354324357182e-05, "loss": 0.007361435331404209, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00739, "step": 167, "tokens/total": 5052032, "tokens/train_per_sec_per_gpu": 34.11, "tokens/trainable": 76331}
-{"epoch": 0.65625, "grad_norm": 0.08688601851463318, "learning_rate": 8.240385928474219e-05, "loss": 0.0016894477885216475, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00169, "step": 168, "tokens/total": 5080256, "tokens/train_per_sec_per_gpu": 35.9, "tokens/trainable": 76745}
-{"epoch": 0.66015625, "grad_norm": 1.221700668334961, "learning_rate": 8.217303493946967e-05, "loss": 0.02566530555486679, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.026, "step": 169, "tokens/total": 5110784, "tokens/train_per_sec_per_gpu": 34.3, "tokens/trainable": 77201}
-{"epoch": 0.6640625, "grad_norm": 0.22431711852550507, "learning_rate": 8.194107981329746e-05, "loss": 0.005605725571513176, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00562, "step": 170, "tokens/total": 5141200, "tokens/train_per_sec_per_gpu": 33.38, "tokens/trainable": 77655}
-{"epoch": 0.66796875, "grad_norm": 0.06314318627119064, "learning_rate": 8.170800355882518e-05, "loss": 0.0013656741939485073, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00137, "step": 171, "tokens/total": 5171760, "tokens/train_per_sec_per_gpu": 39.25, "tokens/trainable": 78122}
-{"epoch": 0.671875, "grad_norm": 0.3465789258480072, "learning_rate": 8.147381587530713e-05, "loss": 0.008099250495433807, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00813, "step": 172, "tokens/total": 5202272, "tokens/train_per_sec_per_gpu": 33.59, "tokens/trainable": 78576}
-{"epoch": 0.67578125, "grad_norm": 0.453427255153656, "learning_rate": 8.123852650824877e-05, "loss": 0.020783744752407074, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.021, "step": 173, "tokens/total": 5232800, "tokens/train_per_sec_per_gpu": 36.49, "tokens/trainable": 79059}
-{"epoch": 0.6796875, "grad_norm": 0.3241178095340729, "learning_rate": 8.100214524900103e-05, "loss": 0.010957238264381886, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.01102, "step": 174, "tokens/total": 5262672, "tokens/train_per_sec_per_gpu": 29.73, "tokens/trainable": 79498}
-{"epoch": 0.68359375, "grad_norm": 0.5538946986198425, "learning_rate": 8.076468193435301e-05, "loss": 0.009046275168657303, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00909, "step": 175, "tokens/total": 5293072, "tokens/train_per_sec_per_gpu": 30.45, "tokens/trainable": 79922}
-{"epoch": 0.6875, "grad_norm": 0.26320791244506836, "learning_rate": 8.052614644612253e-05, "loss": 0.014828844927251339, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.46, "memory/max_allocated (GiB)": 33.46, "ppl": 1.01494, "step": 176, "tokens/total": 5321520, "tokens/train_per_sec_per_gpu": 35.67, "tokens/trainable": 80383}
-{"epoch": 0.69140625, "grad_norm": 0.20839811861515045, "learning_rate": 8.028654871074489e-05, "loss": 0.006698478478938341, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00672, "step": 177, "tokens/total": 5351904, "tokens/train_per_sec_per_gpu": 33.09, "tokens/trainable": 80824}
-{"epoch": 0.6953125, "grad_norm": 0.3556506335735321, "learning_rate": 8.004589869885986e-05, "loss": 0.009760214015841484, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00981, "step": 178, "tokens/total": 5382432, "tokens/train_per_sec_per_gpu": 32.27, "tokens/trainable": 81243}
-{"epoch": 0.69921875, "grad_norm": 0.21442855894565582, "learning_rate": 7.980420642489674e-05, "loss": 0.009938797913491726, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00999, "step": 179, "tokens/total": 5412960, "tokens/train_per_sec_per_gpu": 34.36, "tokens/trainable": 81716}
-{"epoch": 0.703125, "grad_norm": 0.13008078932762146, "learning_rate": 7.95614819466576e-05, "loss": 0.005616464652121067, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00563, "step": 180, "tokens/total": 5443232, "tokens/train_per_sec_per_gpu": 35.61, "tokens/trainable": 82181}
-{"epoch": 0.70703125, "grad_norm": 0.23816989362239838, "learning_rate": 7.931773536489872e-05, "loss": 0.0075413500890135765, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.29, "memory/max_allocated (GiB)": 33.29, "ppl": 1.00757, "step": 181, "tokens/total": 5471440, "tokens/train_per_sec_per_gpu": 34.63, "tokens/trainable": 82626}
-{"epoch": 0.7109375, "grad_norm": 0.13832826912403107, "learning_rate": 7.907297682291035e-05, "loss": 0.0035294657573103905, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00354, "step": 182, "tokens/total": 5501744, "tokens/train_per_sec_per_gpu": 35.17, "tokens/trainable": 83089}
-{"epoch": 0.71484375, "grad_norm": 0.08244283497333527, "learning_rate": 7.882721650609442e-05, "loss": 0.0022330794017761946, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00224, "step": 183, "tokens/total": 5532272, "tokens/train_per_sec_per_gpu": 35.37, "tokens/trainable": 83590}
-{"epoch": 0.71875, "grad_norm": 0.26471880078315735, "learning_rate": 7.85804646415409e-05, "loss": 0.011201716959476471, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.01126, "step": 184, "tokens/total": 5562784, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 84046}
-{"epoch": 0.72265625, "grad_norm": 0.10434233397245407, "learning_rate": 7.833273149760207e-05, "loss": 0.0033001210540533066, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00331, "step": 185, "tokens/total": 5592800, "tokens/train_per_sec_per_gpu": 34.05, "tokens/trainable": 84517}
-{"epoch": 0.7265625, "grad_norm": 0.6545754075050354, "learning_rate": 7.808402738346527e-05, "loss": 0.033577777445316315, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.03415, "step": 186, "tokens/total": 5623088, "tokens/train_per_sec_per_gpu": 33.82, "tokens/trainable": 84974}
-{"epoch": 0.73046875, "grad_norm": 0.016698423773050308, "learning_rate": 7.783436264872382e-05, "loss": 0.000414226611610502, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00041, "step": 187, "tokens/total": 5653344, "tokens/train_per_sec_per_gpu": 35.08, "tokens/trainable": 85460}
-{"epoch": 0.734375, "grad_norm": 0.16612493991851807, "learning_rate": 7.758374768294647e-05, "loss": 0.002929423237219453, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00293, "step": 188, "tokens/total": 5683600, "tokens/train_per_sec_per_gpu": 35.9, "tokens/trainable": 85939}
-{"epoch": 0.73828125, "grad_norm": 0.3205801248550415, "learning_rate": 7.733219291524489e-05, "loss": 0.0012500635348260403, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00125, "step": 189, "tokens/total": 5711952, "tokens/train_per_sec_per_gpu": 38.26, "tokens/trainable": 86377}
-{"epoch": 0.7421875, "grad_norm": 0.5194064378738403, "learning_rate": 7.707970881383977e-05, "loss": 0.009376855567097664, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00942, "step": 190, "tokens/total": 5742336, "tokens/train_per_sec_per_gpu": 33.85, "tokens/trainable": 86834}
-{"epoch": 0.74609375, "grad_norm": 0.6358630061149597, "learning_rate": 7.682630588562518e-05, "loss": 0.009413158521056175, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00946, "step": 191, "tokens/total": 5772560, "tokens/train_per_sec_per_gpu": 32.08, "tokens/trainable": 87265}
-{"epoch": 0.75, "grad_norm": 0.1093795970082283, "learning_rate": 7.657199467573129e-05, "loss": 0.0017574415542185307, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00176, "step": 192, "tokens/total": 5803136, "tokens/train_per_sec_per_gpu": 34.18, "tokens/trainable": 87747}
-{"epoch": 0.75390625, "grad_norm": 0.0865081176161766, "learning_rate": 7.631678576708561e-05, "loss": 0.0017616486875340343, "memory/device_reserved (GiB)": 38.57, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00176, "step": 193, "tokens/total": 5833440, "tokens/train_per_sec_per_gpu": 36.22, "tokens/trainable": 88239}
-{"epoch": 0.7578125, "grad_norm": 0.01772180385887623, "learning_rate": 7.606068977997255e-05, "loss": 0.00022867789084557444, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00023, "step": 194, "tokens/total": 5863552, "tokens/train_per_sec_per_gpu": 36.24, "tokens/trainable": 88718}
-{"epoch": 0.76171875, "grad_norm": 0.2393598109483719, "learning_rate": 7.580371737159148e-05, "loss": 0.003605358535423875, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00361, "step": 195, "tokens/total": 5893680, "tokens/train_per_sec_per_gpu": 31.27, "tokens/trainable": 89129}
-{"epoch": 0.765625, "grad_norm": 0.006237801164388657, "learning_rate": 7.554587923561324e-05, "loss": 0.000122636032756418, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00012, "step": 196, "tokens/total": 5923744, "tokens/train_per_sec_per_gpu": 32.25, "tokens/trainable": 89567}
-{"epoch": 0.76953125, "grad_norm": 0.9131373763084412, "learning_rate": 7.528718610173511e-05, "loss": 0.03544272854924202, "memory/device_reserved (GiB)": 35.57, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.03608, "step": 197, "tokens/total": 5954128, "tokens/train_per_sec_per_gpu": 30.25, "tokens/trainable": 89988}
-{"epoch": 0.7734375, "grad_norm": 0.11690016835927963, "learning_rate": 7.502764873523431e-05, "loss": 0.0010152912000194192, "memory/device_reserved (GiB)": 35.57, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00102, "step": 198, "tokens/total": 5984352, "tokens/train_per_sec_per_gpu": 36.1, "tokens/trainable": 90434}
-{"epoch": 0.77734375, "grad_norm": 0.5135827660560608, "learning_rate": 7.476727793652011e-05, "loss": 0.009797877632081509, "memory/device_reserved (GiB)": 35.61, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00985, "step": 199, "tokens/total": 6014704, "tokens/train_per_sec_per_gpu": 34.17, "tokens/trainable": 90913}
-{"epoch": 0.78125, "grad_norm": 0.26704609394073486, "learning_rate": 7.450608454068415e-05, "loss": 0.009519001469016075, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00956, "step": 200, "tokens/total": 6045056, "tokens/train_per_sec_per_gpu": 34.17, "tokens/trainable": 91355}
-{"epoch": 0.78515625, "grad_norm": 0.3149840235710144, "learning_rate": 7.424407941704987e-05, "loss": 0.006803824566304684, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00683, "step": 201, "tokens/total": 6075504, "tokens/train_per_sec_per_gpu": 37.38, "tokens/trainable": 91856}
-{"epoch": 0.7890625, "grad_norm": 0.5193817615509033, "learning_rate": 7.398127346871986e-05, "loss": 0.01742384023964405, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01758, "step": 202, "tokens/total": 6105904, "tokens/train_per_sec_per_gpu": 34.11, "tokens/trainable": 92311}
-{"epoch": 0.79296875, "grad_norm": 0.12959794700145721, "learning_rate": 7.371767763212238e-05, "loss": 0.0025574658066034317, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00256, "step": 203, "tokens/total": 6136272, "tokens/train_per_sec_per_gpu": 34.55, "tokens/trainable": 92764}
-{"epoch": 0.796875, "grad_norm": 0.17874683439731598, "learning_rate": 7.345330287655617e-05, "loss": 0.004190261010080576, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.65, "memory/max_allocated (GiB)": 33.65, "ppl": 1.0042, "step": 204, "tokens/total": 6166336, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 93227}
-{"epoch": 0.80078125, "grad_norm": 0.38072845339775085, "learning_rate": 7.31881602037339e-05, "loss": 0.010198371484875679, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.01025, "step": 205, "tokens/total": 6196720, "tokens/train_per_sec_per_gpu": 33.07, "tokens/trainable": 93675}
-{"epoch": 0.8046875, "grad_norm": 0.39566439390182495, "learning_rate": 7.29222606473245e-05, "loss": 0.008401261642575264, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00844, "step": 206, "tokens/total": 6227424, "tokens/train_per_sec_per_gpu": 32.33, "tokens/trainable": 94120}
-{"epoch": 0.80859375, "grad_norm": 0.12372284382581711, "learning_rate": 7.265561527249383e-05, "loss": 0.002036123536527157, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00204, "step": 207, "tokens/total": 6257616, "tokens/train_per_sec_per_gpu": 35.21, "tokens/trainable": 94557}
-{"epoch": 0.8125, "grad_norm": 0.20433077216148376, "learning_rate": 7.238823517544436e-05, "loss": 0.010479221120476723, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01053, "step": 208, "tokens/total": 6288000, "tokens/train_per_sec_per_gpu": 40.52, "tokens/trainable": 95035}
-{"epoch": 0.81640625, "grad_norm": 0.06323215365409851, "learning_rate": 7.212013148295333e-05, "loss": 0.0014629911165684462, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00146, "step": 209, "tokens/total": 6318336, "tokens/train_per_sec_per_gpu": 37.41, "tokens/trainable": 95517}
-{"epoch": 0.8203125, "grad_norm": 0.17430178821086884, "learning_rate": 7.185131535190975e-05, "loss": 0.007305104751139879, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00733, "step": 210, "tokens/total": 6348656, "tokens/train_per_sec_per_gpu": 36.31, "tokens/trainable": 96026}
-{"epoch": 0.82421875, "grad_norm": 0.08656168729066849, "learning_rate": 7.158179796885005e-05, "loss": 0.002277363557368517, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00228, "step": 211, "tokens/total": 6379040, "tokens/train_per_sec_per_gpu": 35.44, "tokens/trainable": 96477}
-{"epoch": 0.828125, "grad_norm": 0.10843207687139511, "learning_rate": 7.131159054949273e-05, "loss": 0.0033393804915249348, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00334, "step": 212, "tokens/total": 6409312, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 96951}
-{"epoch": 0.83203125, "grad_norm": 0.20112648606300354, "learning_rate": 7.104070433827139e-05, "loss": 0.00444969953969121, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00446, "step": 213, "tokens/total": 6439520, "tokens/train_per_sec_per_gpu": 30.68, "tokens/trainable": 97350}
-{"epoch": 0.8359375, "grad_norm": 0.0895208865404129, "learning_rate": 7.076915060786705e-05, "loss": 0.0011141544673591852, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00111, "step": 214, "tokens/total": 6469888, "tokens/train_per_sec_per_gpu": 29.79, "tokens/trainable": 97792}
-{"epoch": 0.83984375, "grad_norm": 0.2406485378742218, "learning_rate": 7.049694065873882e-05, "loss": 0.0024814538192003965, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00248, "step": 215, "tokens/total": 6500128, "tokens/train_per_sec_per_gpu": 36.28, "tokens/trainable": 98257}
-{"epoch": 0.84375, "grad_norm": 0.21981537342071533, "learning_rate": 7.022408581865382e-05, "loss": 0.0057976399548351765, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00581, "step": 216, "tokens/total": 6530832, "tokens/train_per_sec_per_gpu": 32.46, "tokens/trainable": 98731}
-{"epoch": 0.84765625, "grad_norm": 0.020311880856752396, "learning_rate": 6.99505974422157e-05, "loss": 0.0002352493756916374, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00024, "step": 217, "tokens/total": 6561248, "tokens/train_per_sec_per_gpu": 36.37, "tokens/trainable": 99212}
-{"epoch": 0.8515625, "grad_norm": 0.047305941581726074, "learning_rate": 6.967648691039213e-05, "loss": 0.000759766495320946, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00076, "step": 218, "tokens/total": 6591648, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 99654}
-{"epoch": 0.85546875, "grad_norm": 0.3724987208843231, "learning_rate": 6.940176563004123e-05, "loss": 0.0064449617639184, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00647, "step": 219, "tokens/total": 6622368, "tokens/train_per_sec_per_gpu": 32.69, "tokens/trainable": 100086}
-{"epoch": 0.859375, "grad_norm": 0.044390205293893814, "learning_rate": 6.912644503343682e-05, "loss": 0.0006100431783124804, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00061, "step": 220, "tokens/total": 6652848, "tokens/train_per_sec_per_gpu": 33.21, "tokens/trainable": 100524}
-{"epoch": 0.86328125, "grad_norm": 0.42042797803878784, "learning_rate": 6.885053657779273e-05, "loss": 0.010451005771756172, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01051, "step": 221, "tokens/total": 6683392, "tokens/train_per_sec_per_gpu": 33.27, "tokens/trainable": 100996}
-{"epoch": 0.8671875, "grad_norm": 0.039993204176425934, "learning_rate": 6.857405174478604e-05, "loss": 0.0005216938443481922, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00052, "step": 222, "tokens/total": 6713712, "tokens/train_per_sec_per_gpu": 31.9, "tokens/trainable": 101449}
-{"epoch": 0.87109375, "grad_norm": 0.4914291501045227, "learning_rate": 6.82970020400792e-05, "loss": 0.01120755635201931, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01127, "step": 223, "tokens/total": 6744176, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 101905}
-{"epoch": 0.875, "grad_norm": 0.1168161928653717, "learning_rate": 6.801939899284132e-05, "loss": 0.0011214457917958498, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00112, "step": 224, "tokens/total": 6774416, "tokens/train_per_sec_per_gpu": 36.36, "tokens/trainable": 102411}
-{"epoch": 0.87890625, "grad_norm": 0.08470873534679413, "learning_rate": 6.774125415526827e-05, "loss": 0.0012768175220116973, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00128, "step": 225, "tokens/total": 6804592, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 102882}
-{"epoch": 0.8828125, "grad_norm": 0.039867568761110306, "learning_rate": 6.746257910210214e-05, "loss": 0.0003806123568210751, "memory/device_reserved (GiB)": 35.49, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00038, "step": 226, "tokens/total": 6834976, "tokens/train_per_sec_per_gpu": 34.57, "tokens/trainable": 103332}
-{"epoch": 0.88671875, "grad_norm": 0.4414898157119751, "learning_rate": 6.718338543014937e-05, "loss": 0.008082674816250801, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00812, "step": 227, "tokens/total": 6865408, "tokens/train_per_sec_per_gpu": 35.08, "tokens/trainable": 103790}
-{"epoch": 0.890625, "grad_norm": 0.2599055767059326, "learning_rate": 6.69036847577983e-05, "loss": 0.004410556983202696, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00442, "step": 228, "tokens/total": 6895664, "tokens/train_per_sec_per_gpu": 34.35, "tokens/trainable": 104246}
-{"epoch": 0.89453125, "grad_norm": 0.518774151802063, "learning_rate": 6.662348872453553e-05, "loss": 0.018912211060523987, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01909, "step": 229, "tokens/total": 6926016, "tokens/train_per_sec_per_gpu": 37.66, "tokens/trainable": 104707}
-{"epoch": 0.8984375, "grad_norm": 0.25264421105384827, "learning_rate": 6.63428089904618e-05, "loss": 0.006381561979651451, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0064, "step": 230, "tokens/total": 6956384, "tokens/train_per_sec_per_gpu": 36.83, "tokens/trainable": 105167}
-{"epoch": 0.90234375, "grad_norm": 0.04987993463873863, "learning_rate": 6.60616572358065e-05, "loss": 0.0007956874324008822, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.0008, "step": 231, "tokens/total": 6986768, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 105576}
-{"epoch": 0.90625, "grad_norm": 0.4156399965286255, "learning_rate": 6.578004516044172e-05, "loss": 0.0147963035851717, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.01491, "step": 232, "tokens/total": 7017232, "tokens/train_per_sec_per_gpu": 36.76, "tokens/trainable": 106043}
-{"epoch": 0.91015625, "grad_norm": 0.1599927842617035, "learning_rate": 6.549798448339548e-05, "loss": 0.0033814553171396255, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00339, "step": 233, "tokens/total": 7047568, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 106506}
-{"epoch": 0.9140625, "grad_norm": 0.0438290536403656, "learning_rate": 6.521548694236384e-05, "loss": 0.0007064358796924353, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00071, "step": 234, "tokens/total": 7077760, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 106951}
-{"epoch": 0.91796875, "grad_norm": 0.13300912082195282, "learning_rate": 6.493256429322259e-05, "loss": 0.003099396824836731, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0031, "step": 235, "tokens/total": 7107760, "tokens/train_per_sec_per_gpu": 30.78, "tokens/trainable": 107382}
-{"epoch": 0.921875, "grad_norm": 0.3202158510684967, "learning_rate": 6.464922830953799e-05, "loss": 0.0032203267328441143, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00323, "step": 236, "tokens/total": 7138144, "tokens/train_per_sec_per_gpu": 32.92, "tokens/trainable": 107814}
-{"epoch": 0.92578125, "grad_norm": 0.15484245121479034, "learning_rate": 6.436549078207688e-05, "loss": 0.002883841749280691, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00289, "step": 237, "tokens/total": 7168352, "tokens/train_per_sec_per_gpu": 35.37, "tokens/trainable": 108274}
-{"epoch": 0.9296875, "grad_norm": 0.09864962846040726, "learning_rate": 6.408136351831592e-05, "loss": 0.0021360500250011683, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00214, "step": 238, "tokens/total": 7198624, "tokens/train_per_sec_per_gpu": 29.22, "tokens/trainable": 108714}
-{"epoch": 0.93359375, "grad_norm": 0.06800950318574905, "learning_rate": 6.379685834195036e-05, "loss": 0.0014171022921800613, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00142, "step": 239, "tokens/total": 7229216, "tokens/train_per_sec_per_gpu": 36.79, "tokens/trainable": 109220}
-{"epoch": 0.9375, "grad_norm": 0.21696041524410248, "learning_rate": 6.351198709240186e-05, "loss": 0.002807284239679575, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00281, "step": 240, "tokens/total": 7259648, "tokens/train_per_sec_per_gpu": 32.12, "tokens/trainable": 109672}
-{"epoch": 0.94140625, "grad_norm": 0.43646690249443054, "learning_rate": 6.32267616243259e-05, "loss": 0.017607467249035835, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.01776, "step": 241, "tokens/total": 7289824, "tokens/train_per_sec_per_gpu": 35.18, "tokens/trainable": 110135}
-{"epoch": 0.9453125, "grad_norm": 0.06252250075340271, "learning_rate": 6.294119380711849e-05, "loss": 0.0006150953122414649, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00062, "step": 242, "tokens/total": 7320288, "tokens/train_per_sec_per_gpu": 30.54, "tokens/trainable": 110563}
-{"epoch": 0.94921875, "grad_norm": 0.16029377281665802, "learning_rate": 6.265529552442209e-05, "loss": 0.0031884105410426855, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00319, "step": 243, "tokens/total": 7350736, "tokens/train_per_sec_per_gpu": 37.12, "tokens/trainable": 111049}
-{"epoch": 0.953125, "grad_norm": 0.17883111536502838, "learning_rate": 6.236907867363127e-05, "loss": 0.0007043592631816864, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0007, "step": 244, "tokens/total": 7381280, "tokens/train_per_sec_per_gpu": 33.41, "tokens/trainable": 111495}
-{"epoch": 0.95703125, "grad_norm": 0.052345480769872665, "learning_rate": 6.208255516539749e-05, "loss": 0.0006958417361602187, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0007, "step": 245, "tokens/total": 7411632, "tokens/train_per_sec_per_gpu": 31.82, "tokens/trainable": 111968}
-{"epoch": 0.9609375, "grad_norm": 0.17381155490875244, "learning_rate": 6.179573692313344e-05, "loss": 0.008788044564425945, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00883, "step": 246, "tokens/total": 7441904, "tokens/train_per_sec_per_gpu": 36.55, "tokens/trainable": 112416}
-{"epoch": 0.96484375, "grad_norm": 0.10432726889848709, "learning_rate": 6.150863588251694e-05, "loss": 0.0013522123917937279, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00135, "step": 247, "tokens/total": 7472368, "tokens/train_per_sec_per_gpu": 34.06, "tokens/trainable": 112882}
-{"epoch": 0.96875, "grad_norm": 0.07330253720283508, "learning_rate": 6.122126399099419e-05, "loss": 0.0010365882189944386, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00104, "step": 248, "tokens/total": 7502656, "tokens/train_per_sec_per_gpu": 40.07, "tokens/trainable": 113390}
-{"epoch": 0.97265625, "grad_norm": 0.7874143123626709, "learning_rate": 6.0933633207282615e-05, "loss": 0.010244790464639664, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0103, "step": 249, "tokens/total": 7532800, "tokens/train_per_sec_per_gpu": 37.84, "tokens/trainable": 113904}
-{"epoch": 0.9765625, "grad_norm": 0.04100371524691582, "learning_rate": 6.064575550087316e-05, "loss": 0.000650806468911469, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00065, "step": 250, "tokens/total": 7563264, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 114351}
-{"epoch": 0.98046875, "grad_norm": 0.14704902470111847, "learning_rate": 6.0357642851532245e-05, "loss": 0.0022576111368834972, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00226, "step": 251, "tokens/total": 7593840, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 114842}
-{"epoch": 0.984375, "grad_norm": 0.2090071737766266, "learning_rate": 6.0069307248803294e-05, "loss": 0.0027604042552411556, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00276, "step": 252, "tokens/total": 7624416, "tokens/train_per_sec_per_gpu": 28.92, "tokens/trainable": 115263}
-{"epoch": 0.98828125, "grad_norm": 0.11346573382616043, "learning_rate": 5.9780760691507635e-05, "loss": 0.0015118042938411236, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00151, "step": 253, "tokens/total": 7654688, "tokens/train_per_sec_per_gpu": 32.76, "tokens/trainable": 115703}
-{"epoch": 0.9921875, "grad_norm": 0.20446987450122833, "learning_rate": 5.9492015187245334e-05, "loss": 0.002259923843666911, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00226, "step": 254, "tokens/total": 7685088, "tokens/train_per_sec_per_gpu": 33.31, "tokens/trainable": 116157}
-{"epoch": 0.99609375, "grad_norm": 0.6619936227798462, "learning_rate": 5.920308275189541e-05, "loss": 0.012125558219850063, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0122, "step": 255, "tokens/total": 7715552, "tokens/train_per_sec_per_gpu": 32.06, "tokens/trainable": 116567}
-{"epoch": 1.0, "grad_norm": 0.017078718170523643, "learning_rate": 5.8913975409115874e-05, "loss": 0.00028051040135324, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00028, "step": 256, "tokens/total": 7745872, "tokens/train_per_sec_per_gpu": 31.26, "tokens/trainable": 117030}
-{"epoch": 1.00390625, "grad_norm": 0.007685024291276932, "learning_rate": 5.8624705189843395e-05, "loss": 0.00016534165479242802, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00017, "step": 257, "tokens/total": 7776208, "tokens/train_per_sec_per_gpu": 34.89, "tokens/trainable": 117517}
-{"epoch": 1.0078125, "grad_norm": 0.046725187450647354, "learning_rate": 5.833528413179249e-05, "loss": 0.0004824839415960014, "memory/device_reserved (GiB)": 35.62, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00048, "step": 258, "tokens/total": 7806480, "tokens/train_per_sec_per_gpu": 38.44, "tokens/trainable": 118015}
-{"epoch": 1.01171875, "grad_norm": 0.093961201608181, "learning_rate": 5.80457242789548e-05, "loss": 0.00111109868157655, "memory/device_reserved (GiB)": 35.62, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00111, "step": 259, "tokens/total": 7836800, "tokens/train_per_sec_per_gpu": 36.3, "tokens/trainable": 118520}
-{"epoch": 1.015625, "grad_norm": 0.024271946400403976, "learning_rate": 5.77560376810977e-05, "loss": 0.00036734913010150194, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00037, "step": 260, "tokens/total": 7867040, "tokens/train_per_sec_per_gpu": 36.24, "tokens/trainable": 118992}
-{"epoch": 1.01953125, "grad_norm": 0.10853405296802521, "learning_rate": 5.7466236393263005e-05, "loss": 0.002126566832885146, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00213, "step": 261, "tokens/total": 7897408, "tokens/train_per_sec_per_gpu": 33.76, "tokens/trainable": 119438}
-{"epoch": 1.0234375, "grad_norm": 0.1838080883026123, "learning_rate": 5.717633247526522e-05, "loss": 0.0012134769931435585, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00121, "step": 262, "tokens/total": 7927648, "tokens/train_per_sec_per_gpu": 32.26, "tokens/trainable": 119881}
-{"epoch": 1.02734375, "grad_norm": 0.28273531794548035, "learning_rate": 5.688633799118971e-05, "loss": 0.0020987153984606266, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0021, "step": 263, "tokens/total": 7957968, "tokens/train_per_sec_per_gpu": 27.91, "tokens/trainable": 120285}
-{"epoch": 1.03125, "grad_norm": 0.07407250255346298, "learning_rate": 5.659626500889066e-05, "loss": 0.00043982663191854954, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.65, "memory/max_allocated (GiB)": 33.65, "ppl": 1.00044, "step": 264, "tokens/total": 7987888, "tokens/train_per_sec_per_gpu": 33.2, "tokens/trainable": 120755}
-{"epoch": 1.03515625, "grad_norm": 0.01878584362566471, "learning_rate": 5.6306125599488905e-05, "loss": 0.00014881066454108804, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00015, "step": 265, "tokens/total": 8018048, "tokens/train_per_sec_per_gpu": 35.69, "tokens/trainable": 121194}
-{"epoch": 1.0390625, "grad_norm": 0.012090266682207584, "learning_rate": 5.601593183686955e-05, "loss": 0.00013956695329397917, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00014, "step": 266, "tokens/total": 8048448, "tokens/train_per_sec_per_gpu": 36.68, "tokens/trainable": 121670}
-{"epoch": 1.04296875, "grad_norm": 0.03192078694701195, "learning_rate": 5.572569579717961e-05, "loss": 0.000175558976479806, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00018, "step": 267, "tokens/total": 8078848, "tokens/train_per_sec_per_gpu": 33.72, "tokens/trainable": 122142}
-{"epoch": 1.046875, "grad_norm": 0.008871566504240036, "learning_rate": 5.543542955832538e-05, "loss": 0.00010361030581407249, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0001, "step": 268, "tokens/total": 8109360, "tokens/train_per_sec_per_gpu": 32.41, "tokens/trainable": 122585}
-{"epoch": 1.05078125, "grad_norm": 0.37323296070098877, "learning_rate": 5.514514519946986e-05, "loss": 0.0028822675812989473, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00289, "step": 269, "tokens/total": 8139888, "tokens/train_per_sec_per_gpu": 38.85, "tokens/trainable": 123091}
-{"epoch": 1.0546875, "grad_norm": 0.019474836066365242, "learning_rate": 5.485485480053015e-05, "loss": 0.0003204788372386247, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00032, "step": 270, "tokens/total": 8170480, "tokens/train_per_sec_per_gpu": 30.9, "tokens/trainable": 123505}
-{"epoch": 1.05859375, "grad_norm": 0.05259509012103081, "learning_rate": 5.4564570441674645e-05, "loss": 0.00033461389830335975, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.41, "memory/max_allocated (GiB)": 33.41, "ppl": 1.00033, "step": 271, "tokens/total": 8198848, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 123953}
-{"epoch": 1.0625, "grad_norm": 0.09935376048088074, "learning_rate": 5.42743042028204e-05, "loss": 0.00043552459101192653, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00044, "step": 272, "tokens/total": 8229200, "tokens/train_per_sec_per_gpu": 28.66, "tokens/trainable": 124400}
-{"epoch": 1.06640625, "grad_norm": 0.3927276134490967, "learning_rate": 5.3984068163130464e-05, "loss": 0.00702043017372489, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00705, "step": 273, "tokens/total": 8259536, "tokens/train_per_sec_per_gpu": 35.05, "tokens/trainable": 124870}
-{"epoch": 1.0703125, "grad_norm": 0.03849954903125763, "learning_rate": 5.369387440051111e-05, "loss": 0.0003886982740368694, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00039, "step": 274, "tokens/total": 8289904, "tokens/train_per_sec_per_gpu": 32.07, "tokens/trainable": 125333}
-{"epoch": 1.07421875, "grad_norm": 0.010367084294557571, "learning_rate": 5.340373499110935e-05, "loss": 8.032341429498047e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00008, "step": 275, "tokens/total": 8320176, "tokens/train_per_sec_per_gpu": 36.59, "tokens/trainable": 125834}
-{"epoch": 1.078125, "grad_norm": 0.05538506433367729, "learning_rate": 5.3113662008810304e-05, "loss": 0.00026508988230489194, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00027, "step": 276, "tokens/total": 8350832, "tokens/train_per_sec_per_gpu": 37.97, "tokens/trainable": 126316}
-{"epoch": 1.08203125, "grad_norm": 0.20107394456863403, "learning_rate": 5.282366752473479e-05, "loss": 0.0007178307860158384, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00072, "step": 277, "tokens/total": 8380976, "tokens/train_per_sec_per_gpu": 35.29, "tokens/trainable": 126798}
-{"epoch": 1.0859375, "grad_norm": 0.005950715858489275, "learning_rate": 5.2533763606737005e-05, "loss": 4.905788227915764e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00005, "step": 278, "tokens/total": 8411072, "tokens/train_per_sec_per_gpu": 29.57, "tokens/trainable": 127223}
-{"epoch": 1.08984375, "grad_norm": 0.009278975427150726, "learning_rate": 5.224396231890232e-05, "loss": 8.996425458462909e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00009, "step": 279, "tokens/total": 8440736, "tokens/train_per_sec_per_gpu": 36.01, "tokens/trainable": 127672}
-{"epoch": 1.09375, "grad_norm": 0.11463552713394165, "learning_rate": 5.195427572104522e-05, "loss": 0.0006871019722893834, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00069, "step": 280, "tokens/total": 8470944, "tokens/train_per_sec_per_gpu": 32.07, "tokens/trainable": 128116}
-{"epoch": 1.09765625, "grad_norm": 0.004335940349847078, "learning_rate": 5.166471586820751e-05, "loss": 4.704743332695216e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00005, "step": 281, "tokens/total": 8501104, "tokens/train_per_sec_per_gpu": 32.7, "tokens/trainable": 128589}
-{"epoch": 1.1015625, "grad_norm": 0.21418413519859314, "learning_rate": 5.1375294810156615e-05, "loss": 0.01315401028841734, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01324, "step": 282, "tokens/total": 8531696, "tokens/train_per_sec_per_gpu": 33.78, "tokens/trainable": 129036}
-{"epoch": 1.10546875, "grad_norm": 0.0006189382984302938, "learning_rate": 5.1086024590884144e-05, "loss": 1.5588291716994718e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00002, "step": 283, "tokens/total": 8561936, "tokens/train_per_sec_per_gpu": 35.16, "tokens/trainable": 129485}
-{"epoch": 1.109375, "grad_norm": 0.005335172638297081, "learning_rate": 5.079691724810461e-05, "loss": 8.037629595492035e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00008, "step": 284, "tokens/total": 8592192, "tokens/train_per_sec_per_gpu": 29.49, "tokens/trainable": 129920}
-{"epoch": 1.11328125, "grad_norm": 0.003026328282430768, "learning_rate": 5.0507984812754684e-05, "loss": 4.424918006407097e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00004, "step": 285, "tokens/total": 8622432, "tokens/train_per_sec_per_gpu": 36.06, "tokens/trainable": 130417}
-{"epoch": 1.1171875, "grad_norm": 0.0032119054812937975, "learning_rate": 5.021923930849237e-05, "loss": 5.364553726394661e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00005, "step": 286, "tokens/total": 8652768, "tokens/train_per_sec_per_gpu": 35.89, "tokens/trainable": 130903}
-{"epoch": 1.12109375, "grad_norm": 0.015378961339592934, "learning_rate": 4.99306927511967e-05, "loss": 9.19914455153048e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00009, "step": 287, "tokens/total": 8683296, "tokens/train_per_sec_per_gpu": 35.45, "tokens/trainable": 131343}
-{"epoch": 1.125, "grad_norm": 0.8838728070259094, "learning_rate": 4.964235714846775e-05, "loss": 0.00693545350804925, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00696, "step": 288, "tokens/total": 8713504, "tokens/train_per_sec_per_gpu": 32.19, "tokens/trainable": 131763}
-{"epoch": 1.12890625, "grad_norm": 0.0019988964777439833, "learning_rate": 4.9354244499126866e-05, "loss": 2.3260268790181726e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00002, "step": 289, "tokens/total": 8743888, "tokens/train_per_sec_per_gpu": 34.0, "tokens/trainable": 132229}
-{"epoch": 1.1328125, "grad_norm": 0.0004976001800969243, "learning_rate": 4.90663667927174e-05, "loss": 1.430663360224571e-05, "memory/device_reserved (GiB)": 35.54, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 290, "tokens/total": 8774336, "tokens/train_per_sec_per_gpu": 35.46, "tokens/trainable": 132678}
-{"epoch": 1.13671875, "grad_norm": 0.01294003613293171, "learning_rate": 4.877873600900581e-05, "loss": 0.0001438881445210427, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00014, "step": 291, "tokens/total": 8804816, "tokens/train_per_sec_per_gpu": 29.39, "tokens/trainable": 133136}
-{"epoch": 1.140625, "grad_norm": 0.3451043963432312, "learning_rate": 4.849136411748306e-05, "loss": 0.0030744036193937063, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00308, "step": 292, "tokens/total": 8835024, "tokens/train_per_sec_per_gpu": 35.31, "tokens/trainable": 133608}
-{"epoch": 1.14453125, "grad_norm": 0.024186862632632256, "learning_rate": 4.8204263076866574e-05, "loss": 0.000165810008184053, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00017, "step": 293, "tokens/total": 8865408, "tokens/train_per_sec_per_gpu": 40.44, "tokens/trainable": 134108}
-{"epoch": 1.1484375, "grad_norm": 0.001723558409139514, "learning_rate": 4.791744483460251e-05, "loss": 3.2396514143329114e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00003, "step": 294, "tokens/total": 8895824, "tokens/train_per_sec_per_gpu": 30.72, "tokens/trainable": 134541}
-{"epoch": 1.15234375, "grad_norm": 0.010282398201525211, "learning_rate": 4.7630921326368736e-05, "loss": 0.00010598616790957749, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00011, "step": 295, "tokens/total": 8926336, "tokens/train_per_sec_per_gpu": 29.51, "tokens/trainable": 134966}
-{"epoch": 1.15625, "grad_norm": 0.02922157198190689, "learning_rate": 4.7344704475577916e-05, "loss": 0.0002625146880745888, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00026, "step": 296, "tokens/total": 8956656, "tokens/train_per_sec_per_gpu": 33.56, "tokens/trainable": 135402}
-{"epoch": 1.16015625, "grad_norm": 0.4516298770904541, "learning_rate": 4.705880619288153e-05, "loss": 0.004878990352153778, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00489, "step": 297, "tokens/total": 8986752, "tokens/train_per_sec_per_gpu": 30.05, "tokens/trainable": 135804}
-{"epoch": 1.1640625, "grad_norm": 0.07809585332870483, "learning_rate": 4.677323837567412e-05, "loss": 0.00019118667114526033, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00019, "step": 298, "tokens/total": 9017024, "tokens/train_per_sec_per_gpu": 31.19, "tokens/trainable": 136231}
-{"epoch": 1.16796875, "grad_norm": 0.008364620618522167, "learning_rate": 4.6488012907598146e-05, "loss": 6.207433034433052e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00006, "step": 299, "tokens/total": 9047424, "tokens/train_per_sec_per_gpu": 34.27, "tokens/trainable": 136705}
-{"epoch": 1.171875, "grad_norm": 0.35634350776672363, "learning_rate": 4.620314165804964e-05, "loss": 0.008261370472609997, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0083, "step": 300, "tokens/total": 9077648, "tokens/train_per_sec_per_gpu": 34.86, "tokens/trainable": 137178}
-{"epoch": 1.17578125, "grad_norm": 0.1326446235179901, "learning_rate": 4.591863648168407e-05, "loss": 0.0006729009910486639, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00067, "step": 301, "tokens/total": 9108032, "tokens/train_per_sec_per_gpu": 34.42, "tokens/trainable": 137643}
-{"epoch": 1.1796875, "grad_norm": 0.11166927963495255, "learning_rate": 4.5634509217923135e-05, "loss": 0.000889140646904707, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00089, "step": 302, "tokens/total": 9138240, "tokens/train_per_sec_per_gpu": 31.94, "tokens/trainable": 138078}
-{"epoch": 1.18359375, "grad_norm": 0.656753420829773, "learning_rate": 4.535077169046201e-05, "loss": 0.004501559771597385, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00451, "step": 303, "tokens/total": 9168352, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 138515}
-{"epoch": 1.1875, "grad_norm": 0.00743053387850523, "learning_rate": 4.506743570677743e-05, "loss": 9.650958236306906e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0001, "step": 304, "tokens/total": 9198864, "tokens/train_per_sec_per_gpu": 34.99, "tokens/trainable": 138948}
-{"epoch": 1.19140625, "grad_norm": 0.0033850902691483498, "learning_rate": 4.478451305763618e-05, "loss": 5.173611862119287e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00005, "step": 305, "tokens/total": 9229104, "tokens/train_per_sec_per_gpu": 38.17, "tokens/trainable": 139408}
-{"epoch": 1.1953125, "grad_norm": 0.0017230098601430655, "learning_rate": 4.450201551660454e-05, "loss": 3.198062040610239e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.38, "memory/max_allocated (GiB)": 33.38, "ppl": 1.00003, "step": 306, "tokens/total": 9257344, "tokens/train_per_sec_per_gpu": 33.48, "tokens/trainable": 139840}
-{"epoch": 1.19921875, "grad_norm": 0.06396278738975525, "learning_rate": 4.4219954839558276e-05, "loss": 0.0004305330221541226, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00043, "step": 307, "tokens/total": 9287520, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 140281}
-{"epoch": 1.203125, "grad_norm": 0.1433790922164917, "learning_rate": 4.393834276419352e-05, "loss": 0.0006829933845438063, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00068, "step": 308, "tokens/total": 9317888, "tokens/train_per_sec_per_gpu": 37.44, "tokens/trainable": 140776}
-{"epoch": 1.20703125, "grad_norm": 0.3920465111732483, "learning_rate": 4.36571910095382e-05, "loss": 0.005061979405581951, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00507, "step": 309, "tokens/total": 9348256, "tokens/train_per_sec_per_gpu": 36.89, "tokens/trainable": 141228}
-{"epoch": 1.2109375, "grad_norm": 0.016641858965158463, "learning_rate": 4.337651127546448e-05, "loss": 0.0001797095756046474, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00018, "step": 310, "tokens/total": 9378560, "tokens/train_per_sec_per_gpu": 37.42, "tokens/trainable": 141679}
-{"epoch": 1.21484375, "grad_norm": 0.0008492676424793899, "learning_rate": 4.3096315242201736e-05, "loss": 1.9429104213486426e-05, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00002, "step": 311, "tokens/total": 9408848, "tokens/train_per_sec_per_gpu": 32.27, "tokens/trainable": 142122}
-{"epoch": 1.21875, "grad_norm": 0.04300769418478012, "learning_rate": 4.2816614569850635e-05, "loss": 0.0005121674039401114, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00051, "step": 312, "tokens/total": 9439344, "tokens/train_per_sec_per_gpu": 33.11, "tokens/trainable": 142561}
-{"epoch": 1.22265625, "grad_norm": 0.023457281291484833, "learning_rate": 4.2537420897897864e-05, "loss": 0.000151450076373294, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00015, "step": 313, "tokens/total": 9469792, "tokens/train_per_sec_per_gpu": 35.66, "tokens/trainable": 143046}
-{"epoch": 1.2265625, "grad_norm": 0.06128578260540962, "learning_rate": 4.225874584473174e-05, "loss": 0.0006227570120245218, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00062, "step": 314, "tokens/total": 9500128, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 143493}
-{"epoch": 1.23046875, "grad_norm": 0.007639073766767979, "learning_rate": 4.19806010071587e-05, "loss": 7.468041440006346e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00007, "step": 315, "tokens/total": 9530480, "tokens/train_per_sec_per_gpu": 35.24, "tokens/trainable": 143956}
-{"epoch": 1.234375, "grad_norm": 0.35354724526405334, "learning_rate": 4.170299795992081e-05, "loss": 0.004370104521512985, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00438, "step": 316, "tokens/total": 9560912, "tokens/train_per_sec_per_gpu": 33.93, "tokens/trainable": 144411}
-{"epoch": 1.23828125, "grad_norm": 0.5020444989204407, "learning_rate": 4.142594825521398e-05, "loss": 0.010973826982080936, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.01103, "step": 317, "tokens/total": 9591344, "tokens/train_per_sec_per_gpu": 38.48, "tokens/trainable": 144892}
-{"epoch": 1.2421875, "grad_norm": 0.0027349034789949656, "learning_rate": 4.114946342220728e-05, "loss": 4.4591506593860686e-05, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00004, "step": 318, "tokens/total": 9621392, "tokens/train_per_sec_per_gpu": 32.6, "tokens/trainable": 145339}
-{"epoch": 1.24609375, "grad_norm": 0.013429106213152409, "learning_rate": 4.087355496656321e-05, "loss": 8.016972424229607e-05, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00008, "step": 319, "tokens/total": 9651728, "tokens/train_per_sec_per_gpu": 36.44, "tokens/trainable": 145811}
-{"epoch": 1.25, "grad_norm": 0.02254675142467022, "learning_rate": 4.05982343699588e-05, "loss": 0.00017163707525469363, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00017, "step": 320, "tokens/total": 9682224, "tokens/train_per_sec_per_gpu": 38.76, "tokens/trainable": 146314}
-{"epoch": 1.25390625, "grad_norm": 0.04359544813632965, "learning_rate": 4.0323513089607876e-05, "loss": 0.0005239051533862948, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00052, "step": 321, "tokens/total": 9712656, "tokens/train_per_sec_per_gpu": 31.85, "tokens/trainable": 146781}
-{"epoch": 1.2578125, "grad_norm": 0.02377651259303093, "learning_rate": 4.004940255778431e-05, "loss": 0.0001890905696200207, "memory/device_reserved (GiB)": 34.89, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00019, "step": 322, "tokens/total": 9743088, "tokens/train_per_sec_per_gpu": 35.55, "tokens/trainable": 147225}
-{"epoch": 1.26171875, "grad_norm": 0.01256605889648199, "learning_rate": 3.977591418134619e-05, "loss": 8.730488480068743e-05, "memory/device_reserved (GiB)": 35.17, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00009, "step": 323, "tokens/total": 9773808, "tokens/train_per_sec_per_gpu": 34.82, "tokens/trainable": 147673}
-{"epoch": 1.265625, "grad_norm": 0.21066401898860931, "learning_rate": 3.95030593412612e-05, "loss": 0.0028422519098967314, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00285, "step": 324, "tokens/total": 9804016, "tokens/train_per_sec_per_gpu": 33.22, "tokens/trainable": 148103}
-{"epoch": 1.26953125, "grad_norm": 0.025614596903324127, "learning_rate": 3.923084939213296e-05, "loss": 0.00016133410099428147, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00016, "step": 325, "tokens/total": 9834592, "tokens/train_per_sec_per_gpu": 31.73, "tokens/trainable": 148535}
-{"epoch": 1.2734375, "grad_norm": 0.033190563321113586, "learning_rate": 3.895929566172861e-05, "loss": 0.00033758440986275673, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00034, "step": 326, "tokens/total": 9864848, "tokens/train_per_sec_per_gpu": 34.07, "tokens/trainable": 148999}
-{"epoch": 1.27734375, "grad_norm": 0.07407072931528091, "learning_rate": 3.868840945050728e-05, "loss": 0.0007672893116250634, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00077, "step": 327, "tokens/total": 9895168, "tokens/train_per_sec_per_gpu": 31.64, "tokens/trainable": 149431}
-{"epoch": 1.28125, "grad_norm": 0.05904461070895195, "learning_rate": 3.841820203114995e-05, "loss": 0.000713472138158977, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00071, "step": 328, "tokens/total": 9925696, "tokens/train_per_sec_per_gpu": 34.11, "tokens/trainable": 149886}
-{"epoch": 1.28515625, "grad_norm": 0.007630123756825924, "learning_rate": 3.814868464809027e-05, "loss": 7.993751933099702e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00008, "step": 329, "tokens/total": 9955648, "tokens/train_per_sec_per_gpu": 30.98, "tokens/trainable": 150288}
-{"epoch": 1.2890625, "grad_norm": 0.024742672219872475, "learning_rate": 3.787986851704667e-05, "loss": 0.00019552679441403598, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.0002, "step": 330, "tokens/total": 9985856, "tokens/train_per_sec_per_gpu": 33.85, "tokens/trainable": 150733}
-{"epoch": 1.29296875, "grad_norm": 0.006353198084980249, "learning_rate": 3.7611764824555654e-05, "loss": 0.00010314649261999875, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0001, "step": 331, "tokens/total": 10016208, "tokens/train_per_sec_per_gpu": 36.7, "tokens/trainable": 151207}
-{"epoch": 1.296875, "grad_norm": 0.16203969717025757, "learning_rate": 3.734438472750619e-05, "loss": 0.0014735229779034853, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00147, "step": 332, "tokens/total": 10046512, "tokens/train_per_sec_per_gpu": 37.5, "tokens/trainable": 151694}
-{"epoch": 1.30078125, "grad_norm": 0.041821569204330444, "learning_rate": 3.707773935267552e-05, "loss": 0.0004190094769001007, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00042, "step": 333, "tokens/total": 10076944, "tokens/train_per_sec_per_gpu": 38.27, "tokens/trainable": 152188}
-{"epoch": 1.3046875, "grad_norm": 0.0011248595546931028, "learning_rate": 3.68118397962661e-05, "loss": 3.350014958414249e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00003, "step": 334, "tokens/total": 10107296, "tokens/train_per_sec_per_gpu": 30.6, "tokens/trainable": 152596}
-{"epoch": 1.30859375, "grad_norm": 0.003707638941705227, "learning_rate": 3.654669712344384e-05, "loss": 4.684936357080005e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00005, "step": 335, "tokens/total": 10137568, "tokens/train_per_sec_per_gpu": 36.44, "tokens/trainable": 153052}
-{"epoch": 1.3125, "grad_norm": 0.0017528374446555972, "learning_rate": 3.628232236787763e-05, "loss": 2.3632102966075763e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00002, "step": 336, "tokens/total": 10167952, "tokens/train_per_sec_per_gpu": 30.19, "tokens/trainable": 153491}
-{"epoch": 1.31640625, "grad_norm": 0.0626155436038971, "learning_rate": 3.6018726531280144e-05, "loss": 0.0006341143744066358, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00063, "step": 337, "tokens/total": 10198512, "tokens/train_per_sec_per_gpu": 40.28, "tokens/trainable": 153983}
-{"epoch": 1.3203125, "grad_norm": 0.0166204534471035, "learning_rate": 3.575592058295017e-05, "loss": 0.00015405623707920313, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00015, "step": 338, "tokens/total": 10228752, "tokens/train_per_sec_per_gpu": 36.59, "tokens/trainable": 154464}
-{"epoch": 1.32421875, "grad_norm": 0.0004508346610236913, "learning_rate": 3.549391545931585e-05, "loss": 8.604627510067075e-06, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 339, "tokens/total": 10259104, "tokens/train_per_sec_per_gpu": 35.73, "tokens/trainable": 154924}
-{"epoch": 1.328125, "grad_norm": 0.0024134982377290726, "learning_rate": 3.5232722063479914e-05, "loss": 4.25071848439984e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00004, "step": 340, "tokens/total": 10289520, "tokens/train_per_sec_per_gpu": 30.76, "tokens/trainable": 155373}
-{"epoch": 1.33203125, "grad_norm": 0.047796547412872314, "learning_rate": 3.49723512647657e-05, "loss": 0.0004414983559399843, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00044, "step": 341, "tokens/total": 10320016, "tokens/train_per_sec_per_gpu": 36.28, "tokens/trainable": 155873}
-{"epoch": 1.3359375, "grad_norm": 0.0013580132508650422, "learning_rate": 3.471281389826491e-05, "loss": 3.1043862691149116e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 342, "tokens/total": 10350368, "tokens/train_per_sec_per_gpu": 30.34, "tokens/trainable": 156278}
-{"epoch": 1.33984375, "grad_norm": 0.013898961246013641, "learning_rate": 3.4454120764386764e-05, "loss": 9.816634701564908e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0001, "step": 343, "tokens/total": 10380624, "tokens/train_per_sec_per_gpu": 35.31, "tokens/trainable": 156733}
-{"epoch": 1.34375, "grad_norm": 0.0031031679827719927, "learning_rate": 3.4196282628408526e-05, "loss": 4.329531657276675e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00004, "step": 344, "tokens/total": 10411024, "tokens/train_per_sec_per_gpu": 33.76, "tokens/trainable": 157203}
-{"epoch": 1.34765625, "grad_norm": 0.009028271771967411, "learning_rate": 3.3939310220027456e-05, "loss": 0.00010301935981260613, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0001, "step": 345, "tokens/total": 10441248, "tokens/train_per_sec_per_gpu": 37.71, "tokens/trainable": 157707}
-{"epoch": 1.3515625, "grad_norm": 0.0036287850234657526, "learning_rate": 3.3683214232914404e-05, "loss": 4.620348772732541e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00005, "step": 346, "tokens/total": 10471712, "tokens/train_per_sec_per_gpu": 34.96, "tokens/trainable": 158180}
-{"epoch": 1.35546875, "grad_norm": 0.08971801400184631, "learning_rate": 3.342800532426873e-05, "loss": 0.0009501657332293689, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00095, "step": 347, "tokens/total": 10502288, "tokens/train_per_sec_per_gpu": 34.32, "tokens/trainable": 158646}
-{"epoch": 1.359375, "grad_norm": 0.00890435092151165, "learning_rate": 3.317369411437484e-05, "loss": 7.783617911627516e-05, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00008, "step": 348, "tokens/total": 10532304, "tokens/train_per_sec_per_gpu": 35.59, "tokens/trainable": 159115}
-{"epoch": 1.36328125, "grad_norm": 0.023319199681282043, "learning_rate": 3.292029118616024e-05, "loss": 0.00021630006085615605, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00022, "step": 349, "tokens/total": 10562608, "tokens/train_per_sec_per_gpu": 31.51, "tokens/trainable": 159538}
-{"epoch": 1.3671875, "grad_norm": 0.047880928963422775, "learning_rate": 3.266780708475511e-05, "loss": 0.00044884331873618066, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00045, "step": 350, "tokens/total": 10592992, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 160016}
-{"epoch": 1.37109375, "grad_norm": 0.008666309528052807, "learning_rate": 3.241625231705354e-05, "loss": 5.017036892240867e-05, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00005, "step": 351, "tokens/total": 10623328, "tokens/train_per_sec_per_gpu": 34.47, "tokens/trainable": 160475}
-{"epoch": 1.375, "grad_norm": 0.06452610343694687, "learning_rate": 3.216563735127618e-05, "loss": 0.0007272367365658283, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00073, "step": 352, "tokens/total": 10653632, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 160952}
-{"epoch": 1.37890625, "grad_norm": 0.370597779750824, "learning_rate": 3.191597261653475e-05, "loss": 0.007550997193902731, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00758, "step": 353, "tokens/total": 10684176, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 161420}
-{"epoch": 1.3828125, "grad_norm": 0.0031376827973872423, "learning_rate": 3.166726850239794e-05, "loss": 2.3453332687495276e-05, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 354, "tokens/total": 10714352, "tokens/train_per_sec_per_gpu": 34.02, "tokens/trainable": 161868}
-{"epoch": 1.38671875, "grad_norm": 0.0021399864926934242, "learning_rate": 3.141953535845912e-05, "loss": 2.5832894607447088e-05, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00003, "step": 355, "tokens/total": 10744464, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 162318}
-{"epoch": 1.390625, "grad_norm": 0.0028148347046226263, "learning_rate": 3.11727834939056e-05, "loss": 3.7286932638380677e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00004, "step": 356, "tokens/total": 10774976, "tokens/train_per_sec_per_gpu": 31.45, "tokens/trainable": 162789}
-{"epoch": 1.39453125, "grad_norm": 0.1642119586467743, "learning_rate": 3.092702317708967e-05, "loss": 0.0027845175936818123, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00279, "step": 357, "tokens/total": 10805600, "tokens/train_per_sec_per_gpu": 36.32, "tokens/trainable": 163254}
-{"epoch": 1.3984375, "grad_norm": 0.066535085439682, "learning_rate": 3.0682264635101276e-05, "loss": 0.0002488417667336762, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00025, "step": 358, "tokens/total": 10835920, "tokens/train_per_sec_per_gpu": 34.41, "tokens/trainable": 163715}
-{"epoch": 1.40234375, "grad_norm": 0.006617655046284199, "learning_rate": 3.0438518053342407e-05, "loss": 7.841112528694794e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00008, "step": 359, "tokens/total": 10866048, "tokens/train_per_sec_per_gpu": 34.4, "tokens/trainable": 164197}
-{"epoch": 1.40625, "grad_norm": 0.010409035719931126, "learning_rate": 3.0195793575103266e-05, "loss": 0.00010489403939573094, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0001, "step": 360, "tokens/total": 10896288, "tokens/train_per_sec_per_gpu": 31.6, "tokens/trainable": 164661}
-{"epoch": 1.41015625, "grad_norm": 0.002426267135888338, "learning_rate": 2.9954101301140146e-05, "loss": 4.0343060391023755e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00004, "step": 361, "tokens/total": 10926816, "tokens/train_per_sec_per_gpu": 32.97, "tokens/trainable": 165111}
-{"epoch": 1.4140625, "grad_norm": 0.015666797757148743, "learning_rate": 2.9713451289255123e-05, "loss": 0.00014991794887464494, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.31, "memory/max_allocated (GiB)": 33.31, "ppl": 1.00015, "step": 362, "tokens/total": 10954928, "tokens/train_per_sec_per_gpu": 30.34, "tokens/trainable": 165552}
-{"epoch": 1.41796875, "grad_norm": 0.2695651650428772, "learning_rate": 2.9473853553877484e-05, "loss": 0.0014983330620452762, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.0015, "step": 363, "tokens/total": 10985328, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 166026}
-{"epoch": 1.421875, "grad_norm": 0.03586212173104286, "learning_rate": 2.9235318065647e-05, "loss": 0.00030509717180393636, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00031, "step": 364, "tokens/total": 11015664, "tokens/train_per_sec_per_gpu": 37.86, "tokens/trainable": 166486}
-{"epoch": 1.42578125, "grad_norm": 0.0019308789633214474, "learning_rate": 2.8997854750998964e-05, "loss": 2.4129443772835657e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00002, "step": 365, "tokens/total": 11046256, "tokens/train_per_sec_per_gpu": 36.34, "tokens/trainable": 166959}
-{"epoch": 1.4296875, "grad_norm": 0.005940837785601616, "learning_rate": 2.8761473491751258e-05, "loss": 2.8238933737156913e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.68, "memory/max_allocated (GiB)": 33.68, "ppl": 1.00003, "step": 366, "tokens/total": 11076288, "tokens/train_per_sec_per_gpu": 33.64, "tokens/trainable": 167394}
-{"epoch": 1.43359375, "grad_norm": 0.002125627128407359, "learning_rate": 2.8526184124692883e-05, "loss": 1.602789416210726e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00002, "step": 367, "tokens/total": 11106416, "tokens/train_per_sec_per_gpu": 28.5, "tokens/trainable": 167832}
-{"epoch": 1.4375, "grad_norm": 0.2510211765766144, "learning_rate": 2.829199644117484e-05, "loss": 0.00010395953722763807, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0001, "step": 368, "tokens/total": 11136848, "tokens/train_per_sec_per_gpu": 30.86, "tokens/trainable": 168242}
-{"epoch": 1.44140625, "grad_norm": 0.001904280623421073, "learning_rate": 2.8058920186702553e-05, "loss": 2.6418363631819375e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00003, "step": 369, "tokens/total": 11167376, "tokens/train_per_sec_per_gpu": 35.63, "tokens/trainable": 168715}
-{"epoch": 1.4453125, "grad_norm": 0.0028152521699666977, "learning_rate": 2.782696506053033e-05, "loss": 2.9974533390486613e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00003, "step": 370, "tokens/total": 11197776, "tokens/train_per_sec_per_gpu": 31.7, "tokens/trainable": 169147}
-{"epoch": 1.44921875, "grad_norm": 0.0010822078911587596, "learning_rate": 2.7596140715257824e-05, "loss": 1.34217716549756e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00001, "step": 371, "tokens/total": 11227840, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 169628}
-{"epoch": 1.453125, "grad_norm": 0.005367911420762539, "learning_rate": 2.7366456756428184e-05, "loss": 6.65646803099662e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00007, "step": 372, "tokens/total": 11258176, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 170085}
-{"epoch": 1.45703125, "grad_norm": 0.047570567578077316, "learning_rate": 2.7137922742128486e-05, "loss": 0.0003471036907285452, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00035, "step": 373, "tokens/total": 11288336, "tokens/train_per_sec_per_gpu": 35.51, "tokens/trainable": 170584}
-{"epoch": 1.4609375, "grad_norm": 0.040886107832193375, "learning_rate": 2.691054818259188e-05, "loss": 0.0002880148240365088, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00029, "step": 374, "tokens/total": 11318848, "tokens/train_per_sec_per_gpu": 34.56, "tokens/trainable": 171058}
-{"epoch": 1.46484375, "grad_norm": 0.0031216361094266176, "learning_rate": 2.6684342539801933e-05, "loss": 2.690855944820214e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00003, "step": 375, "tokens/total": 11349168, "tokens/train_per_sec_per_gpu": 33.84, "tokens/trainable": 171518}
-{"epoch": 1.46875, "grad_norm": 0.012176645919680595, "learning_rate": 2.645931522709877e-05, "loss": 8.402287494391203e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00008, "step": 376, "tokens/total": 11379328, "tokens/train_per_sec_per_gpu": 40.48, "tokens/trainable": 172007}
-{"epoch": 1.47265625, "grad_norm": 0.24136756360530853, "learning_rate": 2.6235475608787365e-05, "loss": 0.0013303000014275312, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00133, "step": 377, "tokens/total": 11409568, "tokens/train_per_sec_per_gpu": 34.71, "tokens/trainable": 172468}
-{"epoch": 1.4765625, "grad_norm": 0.021714258939027786, "learning_rate": 2.6012832999747916e-05, "loss": 0.00020160498388577253, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.0002, "step": 378, "tokens/total": 11439968, "tokens/train_per_sec_per_gpu": 33.64, "tokens/trainable": 172923}
-{"epoch": 1.48046875, "grad_norm": 0.35412073135375977, "learning_rate": 2.579139666504821e-05, "loss": 0.015761105343699455, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01589, "step": 379, "tokens/total": 11470496, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 173370}
-{"epoch": 1.484375, "grad_norm": 0.005766173824667931, "learning_rate": 2.557117581955798e-05, "loss": 2.5790239305933937e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00003, "step": 380, "tokens/total": 11500720, "tokens/train_per_sec_per_gpu": 34.34, "tokens/trainable": 173837}
-{"epoch": 1.48828125, "grad_norm": 0.005898744333535433, "learning_rate": 2.5352179627565532e-05, "loss": 5.821501690661535e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00006, "step": 381, "tokens/total": 11531280, "tokens/train_per_sec_per_gpu": 35.56, "tokens/trainable": 174294}
-{"epoch": 1.4921875, "grad_norm": 0.005451703444123268, "learning_rate": 2.5134417202396277e-05, "loss": 4.0181505028158426e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 382, "tokens/total": 11561264, "tokens/train_per_sec_per_gpu": 30.78, "tokens/trainable": 174701}
-{"epoch": 1.49609375, "grad_norm": 0.0038302047178149223, "learning_rate": 2.491789760603361e-05, "loss": 3.403786467970349e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00003, "step": 383, "tokens/total": 11591616, "tokens/train_per_sec_per_gpu": 32.04, "tokens/trainable": 175131}
-{"epoch": 1.5, "grad_norm": 0.004227485507726669, "learning_rate": 2.4702629848741764e-05, "loss": 3.562243364285678e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00004, "step": 384, "tokens/total": 11622000, "tokens/train_per_sec_per_gpu": 33.96, "tokens/trainable": 175586}
-{"epoch": 1.50390625, "grad_norm": 0.013361346907913685, "learning_rate": 2.4488622888690785e-05, "loss": 0.0001464220113120973, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00015, "step": 385, "tokens/total": 11652352, "tokens/train_per_sec_per_gpu": 34.12, "tokens/trainable": 176026}
-{"epoch": 1.5078125, "grad_norm": 0.008831475861370564, "learning_rate": 2.427588563158384e-05, "loss": 8.2662510976661e-05, "memory/device_reserved (GiB)": 34.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00008, "step": 386, "tokens/total": 11682736, "tokens/train_per_sec_per_gpu": 33.84, "tokens/trainable": 176512}
-{"epoch": 1.51171875, "grad_norm": 0.009311008267104626, "learning_rate": 2.406442693028651e-05, "loss": 0.0001226613821927458, "memory/device_reserved (GiB)": 34.98, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00012, "step": 387, "tokens/total": 11713136, "tokens/train_per_sec_per_gpu": 30.38, "tokens/trainable": 176943}
-{"epoch": 1.515625, "grad_norm": 0.0017816838808357716, "learning_rate": 2.3854255584458547e-05, "loss": 2.3221660740091465e-05, "memory/device_reserved (GiB)": 35.23, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00002, "step": 388, "tokens/total": 11743808, "tokens/train_per_sec_per_gpu": 32.06, "tokens/trainable": 177370}
-{"epoch": 1.51953125, "grad_norm": 0.023602057248353958, "learning_rate": 2.3645380340187508e-05, "loss": 0.00012638044427148998, "memory/device_reserved (GiB)": 35.58, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00013, "step": 389, "tokens/total": 11774144, "tokens/train_per_sec_per_gpu": 35.14, "tokens/trainable": 177851}
-{"epoch": 1.5234375, "grad_norm": 0.04383797571063042, "learning_rate": 2.3437809889624914e-05, "loss": 0.0004640861588995904, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00046, "step": 390, "tokens/total": 11804272, "tokens/train_per_sec_per_gpu": 34.68, "tokens/trainable": 178310}
-{"epoch": 1.52734375, "grad_norm": 0.061932601034641266, "learning_rate": 2.3231552870624487e-05, "loss": 0.00033243370126001537, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00033, "step": 391, "tokens/total": 11832432, "tokens/train_per_sec_per_gpu": 36.8, "tokens/trainable": 178736}
-{"epoch": 1.53125, "grad_norm": 0.04983547702431679, "learning_rate": 2.3026617866382657e-05, "loss": 0.000599355495069176, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0006, "step": 392, "tokens/total": 11862960, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 179233}
-{"epoch": 1.53515625, "grad_norm": 0.012036106549203396, "learning_rate": 2.2823013405081507e-05, "loss": 0.0001237639953615144, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00012, "step": 393, "tokens/total": 11893232, "tokens/train_per_sec_per_gpu": 38.49, "tokens/trainable": 179730}
-{"epoch": 1.5390625, "grad_norm": 0.008992817252874374, "learning_rate": 2.2620747959533722e-05, "loss": 0.00010385089262854308, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0001, "step": 394, "tokens/total": 11923664, "tokens/train_per_sec_per_gpu": 37.81, "tokens/trainable": 180227}
-{"epoch": 1.54296875, "grad_norm": 0.24631057679653168, "learning_rate": 2.2419829946830123e-05, "loss": 0.0031685903668403625, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00317, "step": 395, "tokens/total": 11954032, "tokens/train_per_sec_per_gpu": 32.85, "tokens/trainable": 180687}
-{"epoch": 1.546875, "grad_norm": 0.06871633976697922, "learning_rate": 2.2220267727989325e-05, "loss": 0.0005168755305930972, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00052, "step": 396, "tokens/total": 11984512, "tokens/train_per_sec_per_gpu": 32.76, "tokens/trainable": 181141}
-{"epoch": 1.55078125, "grad_norm": 0.011674679815769196, "learning_rate": 2.202206960760984e-05, "loss": 0.00017496946384198964, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00017, "step": 397, "tokens/total": 12014880, "tokens/train_per_sec_per_gpu": 35.95, "tokens/trainable": 181648}
-{"epoch": 1.5546875, "grad_norm": 0.14610664546489716, "learning_rate": 2.182524383352446e-05, "loss": 0.0014660547021776438, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00147, "step": 398, "tokens/total": 12044928, "tokens/train_per_sec_per_gpu": 33.89, "tokens/trainable": 182109}
-{"epoch": 1.55859375, "grad_norm": 0.016359565779566765, "learning_rate": 2.1629798596457056e-05, "loss": 7.928608101792634e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00008, "step": 399, "tokens/total": 12075344, "tokens/train_per_sec_per_gpu": 31.5, "tokens/trainable": 182559}
-{"epoch": 1.5625, "grad_norm": 0.0010484450031071901, "learning_rate": 2.1435742029681725e-05, "loss": 1.526270352769643e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00002, "step": 400, "tokens/total": 12105616, "tokens/train_per_sec_per_gpu": 33.21, "tokens/trainable": 182985}
-{"epoch": 1.56640625, "grad_norm": 0.0005900752730667591, "learning_rate": 2.124308220868431e-05, "loss": 1.2166316082584672e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00001, "step": 401, "tokens/total": 12136080, "tokens/train_per_sec_per_gpu": 32.69, "tokens/trainable": 183440}
-{"epoch": 1.5703125, "grad_norm": 0.013194791041314602, "learning_rate": 2.105182715082638e-05, "loss": 0.0001970212033484131, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.0002, "step": 402, "tokens/total": 12166240, "tokens/train_per_sec_per_gpu": 34.52, "tokens/trainable": 183917}
-{"epoch": 1.57421875, "grad_norm": 0.02566305734217167, "learning_rate": 2.0861984815011552e-05, "loss": 0.00022389904188457876, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00022, "step": 403, "tokens/total": 12196544, "tokens/train_per_sec_per_gpu": 30.39, "tokens/trainable": 184348}
-{"epoch": 1.578125, "grad_norm": 0.008275284431874752, "learning_rate": 2.0673563101354323e-05, "loss": 8.35009923321195e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00008, "step": 404, "tokens/total": 12226784, "tokens/train_per_sec_per_gpu": 31.53, "tokens/trainable": 184786}
-{"epoch": 1.58203125, "grad_norm": 0.038177311420440674, "learning_rate": 2.0486569850851317e-05, "loss": 0.00018527230713516474, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00019, "step": 405, "tokens/total": 12257264, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 185249}
-{"epoch": 1.5859375, "grad_norm": 0.007672510575503111, "learning_rate": 2.0301012845054956e-05, "loss": 0.00011439670925028622, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00011, "step": 406, "tokens/total": 12287584, "tokens/train_per_sec_per_gpu": 31.53, "tokens/trainable": 185680}
-{"epoch": 1.58984375, "grad_norm": 0.003159885760396719, "learning_rate": 2.011689980574966e-05, "loss": 4.649449692806229e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00005, "step": 407, "tokens/total": 12317984, "tokens/train_per_sec_per_gpu": 31.72, "tokens/trainable": 186100}
-{"epoch": 1.59375, "grad_norm": 0.32744285464286804, "learning_rate": 1.993423839463052e-05, "loss": 0.002739987801760435, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00274, "step": 408, "tokens/total": 12348144, "tokens/train_per_sec_per_gpu": 39.82, "tokens/trainable": 186565}
-{"epoch": 1.59765625, "grad_norm": 0.0006675662589259446, "learning_rate": 1.975303621298445e-05, "loss": 1.2970660463906825e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 409, "tokens/total": 12378544, "tokens/train_per_sec_per_gpu": 35.84, "tokens/trainable": 187017}
-{"epoch": 1.6015625, "grad_norm": 0.003976705949753523, "learning_rate": 1.957330080137385e-05, "loss": 3.857718547806144e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00004, "step": 410, "tokens/total": 12408784, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 187468}
-{"epoch": 1.60546875, "grad_norm": 0.0024577646981924772, "learning_rate": 1.9395039639322864e-05, "loss": 2.431379834888503e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00002, "step": 411, "tokens/total": 12438832, "tokens/train_per_sec_per_gpu": 37.61, "tokens/trainable": 187961}
-{"epoch": 1.609375, "grad_norm": 0.003754909848794341, "learning_rate": 1.9218260145006073e-05, "loss": 4.119630830246024e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00004, "step": 412, "tokens/total": 12469296, "tokens/train_per_sec_per_gpu": 35.84, "tokens/trainable": 188447}
-{"epoch": 1.61328125, "grad_norm": 0.3583323657512665, "learning_rate": 1.904296967493982e-05, "loss": 0.004735157359391451, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00475, "step": 413, "tokens/total": 12499392, "tokens/train_per_sec_per_gpu": 35.95, "tokens/trainable": 188903}
-{"epoch": 1.6171875, "grad_norm": 0.049485232681035995, "learning_rate": 1.8869175523676064e-05, "loss": 0.00019404500199016184, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 34.0, "memory/max_allocated (GiB)": 34.0, "ppl": 1.00019, "step": 414, "tokens/total": 12529952, "tokens/train_per_sec_per_gpu": 36.7, "tokens/trainable": 189398}
-{"epoch": 1.62109375, "grad_norm": 0.0030447980388998985, "learning_rate": 1.869688492349885e-05, "loss": 2.0038012735312805e-05, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 415, "tokens/total": 12560240, "tokens/train_per_sec_per_gpu": 34.16, "tokens/trainable": 189850}
-{"epoch": 1.625, "grad_norm": 0.25415608286857605, "learning_rate": 1.85261050441233e-05, "loss": 0.0016201161779463291, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00162, "step": 416, "tokens/total": 12590736, "tokens/train_per_sec_per_gpu": 32.64, "tokens/trainable": 190306}
-{"epoch": 1.62890625, "grad_norm": 0.006711115129292011, "learning_rate": 1.8356842992397304e-05, "loss": 4.321872620494105e-05, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00004, "step": 417, "tokens/total": 12621168, "tokens/train_per_sec_per_gpu": 34.19, "tokens/trainable": 190746}
-{"epoch": 1.6328125, "grad_norm": 0.004931971430778503, "learning_rate": 1.8189105812005714e-05, "loss": 4.740363510791212e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00005, "step": 418, "tokens/total": 12651456, "tokens/train_per_sec_per_gpu": 31.07, "tokens/trainable": 191156}
-{"epoch": 1.63671875, "grad_norm": 0.007677357643842697, "learning_rate": 1.802290048317732e-05, "loss": 7.545409607701004e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00008, "step": 419, "tokens/total": 12681760, "tokens/train_per_sec_per_gpu": 29.47, "tokens/trainable": 191573}
-{"epoch": 1.640625, "grad_norm": 0.015445503406226635, "learning_rate": 1.785823392239424e-05, "loss": 0.0001045453900587745, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.0001, "step": 420, "tokens/total": 12712144, "tokens/train_per_sec_per_gpu": 32.73, "tokens/trainable": 192014}
-{"epoch": 1.64453125, "grad_norm": 0.16882061958312988, "learning_rate": 1.7695112982104225e-05, "loss": 0.0024146074429154396, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00242, "step": 421, "tokens/total": 12742400, "tokens/train_per_sec_per_gpu": 32.82, "tokens/trainable": 192453}
-{"epoch": 1.6484375, "grad_norm": 0.002461223630234599, "learning_rate": 1.7533544450435433e-05, "loss": 3.336594454594888e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00003, "step": 422, "tokens/total": 12772288, "tokens/train_per_sec_per_gpu": 33.21, "tokens/trainable": 192886}
-{"epoch": 1.65234375, "grad_norm": 0.0005388484569266438, "learning_rate": 1.7373535050913946e-05, "loss": 8.696397344465367e-06, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00001, "step": 423, "tokens/total": 12802880, "tokens/train_per_sec_per_gpu": 29.85, "tokens/trainable": 193323}
-{"epoch": 1.65625, "grad_norm": 0.005009706597775221, "learning_rate": 1.721509144218405e-05, "loss": 4.8607362259645015e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00005, "step": 424, "tokens/total": 12833456, "tokens/train_per_sec_per_gpu": 32.6, "tokens/trainable": 193753}
-{"epoch": 1.66015625, "grad_norm": 0.001982118235900998, "learning_rate": 1.705822021773101e-05, "loss": 3.262238169554621e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00003, "step": 425, "tokens/total": 12863952, "tokens/train_per_sec_per_gpu": 36.33, "tokens/trainable": 194240}
-{"epoch": 1.6640625, "grad_norm": 0.05305991321802139, "learning_rate": 1.69029279056068e-05, "loss": 0.0005229170201346278, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00052, "step": 426, "tokens/total": 12894352, "tokens/train_per_sec_per_gpu": 35.53, "tokens/trainable": 194680}
-{"epoch": 1.66796875, "grad_norm": 0.001265498693101108, "learning_rate": 1.6749220968158415e-05, "loss": 1.5427456673933193e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 427, "tokens/total": 12924688, "tokens/train_per_sec_per_gpu": 38.73, "tokens/trainable": 195186}
-{"epoch": 1.671875, "grad_norm": 0.0016263640718534589, "learning_rate": 1.659710580175893e-05, "loss": 2.7778178264270537e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00003, "step": 428, "tokens/total": 12955072, "tokens/train_per_sec_per_gpu": 36.51, "tokens/trainable": 195693}
-{"epoch": 1.67578125, "grad_norm": 0.013330154120922089, "learning_rate": 1.644658873654133e-05, "loss": 6.972272240091115e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00007, "step": 429, "tokens/total": 12985328, "tokens/train_per_sec_per_gpu": 36.93, "tokens/trainable": 196130}
-{"epoch": 1.6796875, "grad_norm": 0.0034615101758390665, "learning_rate": 1.629767603613508e-05, "loss": 4.500148497754708e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00005, "step": 430, "tokens/total": 13015600, "tokens/train_per_sec_per_gpu": 29.58, "tokens/trainable": 196530}
-{"epoch": 1.68359375, "grad_norm": 0.03424045443534851, "learning_rate": 1.615037389740547e-05, "loss": 0.0002949235204141587, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00029, "step": 431, "tokens/total": 13045728, "tokens/train_per_sec_per_gpu": 29.88, "tokens/trainable": 196974}
-{"epoch": 1.6875, "grad_norm": 0.0064499350264668465, "learning_rate": 1.600468845019576e-05, "loss": 7.787663344061002e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00008, "step": 432, "tokens/total": 13075920, "tokens/train_per_sec_per_gpu": 35.25, "tokens/trainable": 197454}
-{"epoch": 1.69140625, "grad_norm": 0.00021908426424488425, "learning_rate": 1.5860625757072092e-05, "loss": 6.603059773624409e-06, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 433, "tokens/total": 13106128, "tokens/train_per_sec_per_gpu": 31.4, "tokens/trainable": 197902}
-{"epoch": 1.6953125, "grad_norm": 0.0002612156968098134, "learning_rate": 1.571819181307116e-05, "loss": 6.425582796509843e-06, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00001, "step": 434, "tokens/total": 13136368, "tokens/train_per_sec_per_gpu": 36.0, "tokens/trainable": 198374}
-{"epoch": 1.69921875, "grad_norm": 0.001639618189074099, "learning_rate": 1.557739254545075e-05, "loss": 2.5485322112217546e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00003, "step": 435, "tokens/total": 13166704, "tokens/train_per_sec_per_gpu": 37.54, "tokens/trainable": 198846}
-{"epoch": 1.703125, "grad_norm": 0.01887959986925125, "learning_rate": 1.543823381344311e-05, "loss": 0.0002141120348824188, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00021, "step": 436, "tokens/total": 13197344, "tokens/train_per_sec_per_gpu": 37.67, "tokens/trainable": 199320}
-{"epoch": 1.70703125, "grad_norm": 0.0012294527841731906, "learning_rate": 1.5300721408011114e-05, "loss": 2.316079735464882e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 437, "tokens/total": 13227664, "tokens/train_per_sec_per_gpu": 37.43, "tokens/trainable": 199777}
-{"epoch": 1.7109375, "grad_norm": 0.005585651844739914, "learning_rate": 1.5164861051607254e-05, "loss": 4.5935248635942116e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00005, "step": 438, "tokens/total": 13257632, "tokens/train_per_sec_per_gpu": 30.82, "tokens/trainable": 200194}
-{"epoch": 1.71484375, "grad_norm": 0.001553745474666357, "learning_rate": 1.5030658397935521e-05, "loss": 2.2851421817904338e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 439, "tokens/total": 13288240, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 200663}
-{"epoch": 1.71875, "grad_norm": 0.0029791900888085365, "learning_rate": 1.4898119031716104e-05, "loss": 4.093274037586525e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00004, "step": 440, "tokens/total": 13318992, "tokens/train_per_sec_per_gpu": 33.79, "tokens/trainable": 201121}
-{"epoch": 1.72265625, "grad_norm": 0.0017392379231750965, "learning_rate": 1.476724846845306e-05, "loss": 7.29740804672474e-06, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00001, "step": 441, "tokens/total": 13349392, "tokens/train_per_sec_per_gpu": 39.74, "tokens/trainable": 201598}
-{"epoch": 1.7265625, "grad_norm": 0.0009943305049091578, "learning_rate": 1.463805215420471e-05, "loss": 1.7162077710963786e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 442, "tokens/total": 13379648, "tokens/train_per_sec_per_gpu": 34.94, "tokens/trainable": 202068}
-{"epoch": 1.73046875, "grad_norm": 0.0011473585618659854, "learning_rate": 1.451053546535705e-05, "loss": 2.3819740817998536e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00002, "step": 443, "tokens/total": 13409936, "tokens/train_per_sec_per_gpu": 35.57, "tokens/trainable": 202524}
-{"epoch": 1.734375, "grad_norm": 0.006799118127673864, "learning_rate": 1.438470370840001e-05, "loss": 6.556943117175251e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00007, "step": 444, "tokens/total": 13440288, "tokens/train_per_sec_per_gpu": 36.34, "tokens/trainable": 202996}
-{"epoch": 1.73828125, "grad_norm": 0.0011645558988675475, "learning_rate": 1.4260562119706606e-05, "loss": 1.8389995602774434e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00002, "step": 445, "tokens/total": 13468880, "tokens/train_per_sec_per_gpu": 39.14, "tokens/trainable": 203470}
-{"epoch": 1.7421875, "grad_norm": 0.020891210064291954, "learning_rate": 1.413811586531508e-05, "loss": 0.00016399261949118227, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00016, "step": 446, "tokens/total": 13499184, "tokens/train_per_sec_per_gpu": 35.2, "tokens/trainable": 203919}
-{"epoch": 1.74609375, "grad_norm": 0.09615519642829895, "learning_rate": 1.4017370040713884e-05, "loss": 0.0010675137164071202, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00107, "step": 447, "tokens/total": 13529632, "tokens/train_per_sec_per_gpu": 33.3, "tokens/trainable": 204376}
-{"epoch": 1.75, "grad_norm": 0.09744399785995483, "learning_rate": 1.3898329670629645e-05, "loss": 0.001196134602651, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.0012, "step": 448, "tokens/total": 13560080, "tokens/train_per_sec_per_gpu": 32.0, "tokens/trainable": 204821}
-{"epoch": 1.75390625, "grad_norm": 0.0007810555398464203, "learning_rate": 1.3780999708818058e-05, "loss": 8.019956112548243e-06, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00001, "step": 449, "tokens/total": 13590496, "tokens/train_per_sec_per_gpu": 33.63, "tokens/trainable": 205271}
-{"epoch": 1.7578125, "grad_norm": 0.034087613224983215, "learning_rate": 1.3665385037857758e-05, "loss": 0.00010244422446703538, "memory/device_reserved (GiB)": 34.92, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0001, "step": 450, "tokens/total": 13620960, "tokens/train_per_sec_per_gpu": 30.95, "tokens/trainable": 205723}
-{"epoch": 1.76171875, "grad_norm": 0.02930639684200287, "learning_rate": 1.3551490468947126e-05, "loss": 0.00021926099725533277, "memory/device_reserved (GiB)": 35.04, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00022, "step": 451, "tokens/total": 13651344, "tokens/train_per_sec_per_gpu": 32.42, "tokens/trainable": 206169}
-{"epoch": 1.765625, "grad_norm": 0.0109552051872015, "learning_rate": 1.3439320741704075e-05, "loss": 7.999646186362952e-05, "memory/device_reserved (GiB)": 35.04, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00008, "step": 452, "tokens/total": 13681792, "tokens/train_per_sec_per_gpu": 32.74, "tokens/trainable": 206592}
-{"epoch": 1.76953125, "grad_norm": 0.004453443922102451, "learning_rate": 1.3328880523968808e-05, "loss": 3.79011980840005e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00004, "step": 453, "tokens/total": 13712256, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 207055}
-{"epoch": 1.7734375, "grad_norm": 0.0005650459788739681, "learning_rate": 1.3220174411609587e-05, "loss": 1.2244867320987396e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00001, "step": 454, "tokens/total": 13742128, "tokens/train_per_sec_per_gpu": 34.65, "tokens/trainable": 207494}
-{"epoch": 1.77734375, "grad_norm": 0.0007141040405258536, "learning_rate": 1.3113206928331471e-05, "loss": 1.3571594536188059e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 455, "tokens/total": 13772160, "tokens/train_per_sec_per_gpu": 32.97, "tokens/trainable": 207956}
-{"epoch": 1.78125, "grad_norm": 0.0008097058744169772, "learning_rate": 1.300798252548806e-05, "loss": 1.1659047231660224e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 456, "tokens/total": 13802448, "tokens/train_per_sec_per_gpu": 35.53, "tokens/trainable": 208408}
-{"epoch": 1.78515625, "grad_norm": 0.0012649539858102798, "learning_rate": 1.2904505581896265e-05, "loss": 9.200517524732277e-06, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00001, "step": 457, "tokens/total": 13832880, "tokens/train_per_sec_per_gpu": 36.0, "tokens/trainable": 208889}
-{"epoch": 1.7890625, "grad_norm": 0.014978500083088875, "learning_rate": 1.2802780403654082e-05, "loss": 0.0001208957692142576, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00012, "step": 458, "tokens/total": 13862880, "tokens/train_per_sec_per_gpu": 34.08, "tokens/trainable": 209336}
-{"epoch": 1.79296875, "grad_norm": 0.004875013139098883, "learning_rate": 1.2702811223961408e-05, "loss": 3.241455851821229e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00003, "step": 459, "tokens/total": 13893440, "tokens/train_per_sec_per_gpu": 33.26, "tokens/trainable": 209797}
-{"epoch": 1.796875, "grad_norm": 0.0009627968538552523, "learning_rate": 1.2604602202943861e-05, "loss": 1.980438355531078e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00002, "step": 460, "tokens/total": 13923824, "tokens/train_per_sec_per_gpu": 36.12, "tokens/trainable": 210240}
-{"epoch": 1.80078125, "grad_norm": 0.019912317395210266, "learning_rate": 1.2508157427479686e-05, "loss": 0.00010681153071345761, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00011, "step": 461, "tokens/total": 13954208, "tokens/train_per_sec_per_gpu": 34.91, "tokens/trainable": 210702}
-{"epoch": 1.8046875, "grad_norm": 0.0005370390135794878, "learning_rate": 1.2413480911029655e-05, "loss": 1.1081473530794028e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 462, "tokens/total": 13984784, "tokens/train_per_sec_per_gpu": 31.57, "tokens/trainable": 211140}
-{"epoch": 1.80859375, "grad_norm": 0.000815370527561754, "learning_rate": 1.2320576593470082e-05, "loss": 1.5510300727328286e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00002, "step": 463, "tokens/total": 14014896, "tokens/train_per_sec_per_gpu": 32.6, "tokens/trainable": 211575}
-{"epoch": 1.8125, "grad_norm": 0.02875823900103569, "learning_rate": 1.2229448340928828e-05, "loss": 0.00013141569797880948, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00013, "step": 464, "tokens/total": 14045312, "tokens/train_per_sec_per_gpu": 39.03, "tokens/trainable": 212083}
-{"epoch": 1.81640625, "grad_norm": 0.04558868706226349, "learning_rate": 1.2140099945624458e-05, "loss": 0.0002969688503071666, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0003, "step": 465, "tokens/total": 14075840, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 212558}
-{"epoch": 1.8203125, "grad_norm": 0.0004122587270103395, "learning_rate": 1.205253512570841e-05, "loss": 9.937594768416602e-06, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 466, "tokens/total": 14106016, "tokens/train_per_sec_per_gpu": 32.91, "tokens/trainable": 213014}
-{"epoch": 1.82421875, "grad_norm": 0.0010108116548508406, "learning_rate": 1.1966757525110255e-05, "loss": 1.5579567843815312e-05, "memory/device_reserved (GiB)": 37.99, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00002, "step": 467, "tokens/total": 14136512, "tokens/train_per_sec_per_gpu": 34.03, "tokens/trainable": 213465}
-{"epoch": 1.828125, "grad_norm": 0.0043095871806144714, "learning_rate": 1.1882770713386095e-05, "loss": 3.2609641493763775e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 468, "tokens/total": 14166832, "tokens/train_per_sec_per_gpu": 36.62, "tokens/trainable": 213940}
-{"epoch": 1.83203125, "grad_norm": 0.005317045841366053, "learning_rate": 1.180057818556998e-05, "loss": 2.144884638255462e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 469, "tokens/total": 14197360, "tokens/train_per_sec_per_gpu": 35.03, "tokens/trainable": 214415}
-{"epoch": 1.8359375, "grad_norm": 0.0003264908737037331, "learning_rate": 1.1720183362028494e-05, "loss": 5.6935500651889015e-06, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 470, "tokens/total": 14227600, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 214880}
-{"epoch": 1.83984375, "grad_norm": 0.013723314739763737, "learning_rate": 1.1641589588318387e-05, "loss": 0.00010619591921567917, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00011, "step": 471, "tokens/total": 14257888, "tokens/train_per_sec_per_gpu": 33.41, "tokens/trainable": 215334}
-{"epoch": 1.84375, "grad_norm": 0.00020912640320602804, "learning_rate": 1.1564800135047418e-05, "loss": 4.586940121953376e-06, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0, "step": 472, "tokens/total": 14288208, "tokens/train_per_sec_per_gpu": 34.94, "tokens/trainable": 215822}
-{"epoch": 1.84765625, "grad_norm": 0.10667876899242401, "learning_rate": 1.148981819773816e-05, "loss": 0.0006915333215147257, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00069, "step": 473, "tokens/total": 14318448, "tokens/train_per_sec_per_gpu": 30.86, "tokens/trainable": 216239}
-{"epoch": 1.8515625, "grad_norm": 0.013282700441777706, "learning_rate": 1.1416646896695086e-05, "loss": 6.861680594738573e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00007, "step": 474, "tokens/total": 14348912, "tokens/train_per_sec_per_gpu": 33.37, "tokens/trainable": 216691}
-{"epoch": 1.85546875, "grad_norm": 0.0074070473201572895, "learning_rate": 1.1345289276874717e-05, "loss": 8.334079757332802e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00008, "step": 475, "tokens/total": 14379440, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 217162}
-{"epoch": 1.859375, "grad_norm": 0.0017771078273653984, "learning_rate": 1.1275748307758873e-05, "loss": 1.803937993827276e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 476, "tokens/total": 14409840, "tokens/train_per_sec_per_gpu": 35.84, "tokens/trainable": 217638}
-{"epoch": 1.86328125, "grad_norm": 0.00044826362864114344, "learning_rate": 1.1208026883231147e-05, "loss": 9.577534910931718e-06, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00001, "step": 477, "tokens/total": 14440256, "tokens/train_per_sec_per_gpu": 32.81, "tokens/trainable": 218053}
-{"epoch": 1.8671875, "grad_norm": 0.0009560272446833551, "learning_rate": 1.1142127821456433e-05, "loss": 1.0945323083433323e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 478, "tokens/total": 14470704, "tokens/train_per_sec_per_gpu": 34.12, "tokens/trainable": 218503}
-{"epoch": 1.87109375, "grad_norm": 0.2086740881204605, "learning_rate": 1.1078053864763674e-05, "loss": 0.0012299084337428212, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00123, "step": 479, "tokens/total": 14501072, "tokens/train_per_sec_per_gpu": 36.19, "tokens/trainable": 218964}
-{"epoch": 1.875, "grad_norm": 0.019221138209104538, "learning_rate": 1.1015807679531756e-05, "loss": 0.00011615331459324807, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00012, "step": 480, "tokens/total": 14531200, "tokens/train_per_sec_per_gpu": 35.51, "tokens/trainable": 219412}
-{"epoch": 1.87890625, "grad_norm": 0.002074754796922207, "learning_rate": 1.0955391856078528e-05, "loss": 2.3620896172360517e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00002, "step": 481, "tokens/total": 14561152, "tokens/train_per_sec_per_gpu": 38.29, "tokens/trainable": 219886}
-{"epoch": 1.8828125, "grad_norm": 0.011151358485221863, "learning_rate": 1.0896808908553007e-05, "loss": 5.527659959625453e-05, "memory/device_reserved (GiB)": 37.77, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00006, "step": 482, "tokens/total": 14591264, "tokens/train_per_sec_per_gpu": 34.53, "tokens/trainable": 220343}
-{"epoch": 1.88671875, "grad_norm": 0.010913084261119366, "learning_rate": 1.0840061274830763e-05, "loss": 6.711594323860481e-05, "memory/device_reserved (GiB)": 37.77, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00007, "step": 483, "tokens/total": 14621568, "tokens/train_per_sec_per_gpu": 35.24, "tokens/trainable": 220825}
-{"epoch": 1.890625, "grad_norm": 0.3236236274242401, "learning_rate": 1.0785151316412473e-05, "loss": 0.002836203668266535, "memory/device_reserved (GiB)": 37.77, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00284, "step": 484, "tokens/total": 14652064, "tokens/train_per_sec_per_gpu": 32.52, "tokens/trainable": 221282}
-{"epoch": 1.89453125, "grad_norm": 0.0002806074626278132, "learning_rate": 1.0732081318325639e-05, "loss": 5.300881639414001e-06, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00001, "step": 485, "tokens/total": 14682608, "tokens/train_per_sec_per_gpu": 37.86, "tokens/trainable": 221780}
-{"epoch": 1.8984375, "grad_norm": 0.009111613035202026, "learning_rate": 1.0680853489029501e-05, "loss": 3.664881660370156e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00004, "step": 486, "tokens/total": 14713168, "tokens/train_per_sec_per_gpu": 31.52, "tokens/trainable": 222210}
-{"epoch": 1.90234375, "grad_norm": 0.04484618082642555, "learning_rate": 1.0631469960323152e-05, "loss": 0.0002056290686596185, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00021, "step": 487, "tokens/total": 14743536, "tokens/train_per_sec_per_gpu": 29.33, "tokens/trainable": 222646}
-{"epoch": 1.90625, "grad_norm": 0.2784590721130371, "learning_rate": 1.0583932787256783e-05, "loss": 0.008498833514750004, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00854, "step": 488, "tokens/total": 14773728, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 223066}
-{"epoch": 1.91015625, "grad_norm": 0.0006373917567543685, "learning_rate": 1.0538243948046206e-05, "loss": 9.457155101699755e-06, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00001, "step": 489, "tokens/total": 14803744, "tokens/train_per_sec_per_gpu": 34.19, "tokens/trainable": 223512}
-{"epoch": 1.9140625, "grad_norm": 0.00038701840094290674, "learning_rate": 1.0494405343990523e-05, "loss": 6.92771118337987e-06, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00001, "step": 490, "tokens/total": 14832112, "tokens/train_per_sec_per_gpu": 38.96, "tokens/trainable": 223956}
-{"epoch": 1.91796875, "grad_norm": 0.0015320515958592296, "learning_rate": 1.0452418799392985e-05, "loss": 2.272017445648089e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.44, "memory/max_allocated (GiB)": 33.44, "ppl": 1.00002, "step": 491, "tokens/total": 14860560, "tokens/train_per_sec_per_gpu": 36.33, "tokens/trainable": 224416}
-{"epoch": 1.921875, "grad_norm": 0.016094110906124115, "learning_rate": 1.0412286061485102e-05, "loss": 0.00011934755457332358, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00012, "step": 492, "tokens/total": 14890848, "tokens/train_per_sec_per_gpu": 34.09, "tokens/trainable": 224883}
-{"epoch": 1.92578125, "grad_norm": 0.0016222589183598757, "learning_rate": 1.03740088003539e-05, "loss": 1.4765413652639836e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 493, "tokens/total": 14921008, "tokens/train_per_sec_per_gpu": 35.47, "tokens/trainable": 225324}
-{"epoch": 1.9296875, "grad_norm": 0.004442836623638868, "learning_rate": 1.0337588608872463e-05, "loss": 3.44704239978455e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00003, "step": 494, "tokens/total": 14951360, "tokens/train_per_sec_per_gpu": 34.9, "tokens/trainable": 225783}
-{"epoch": 1.93359375, "grad_norm": 0.002437110524624586, "learning_rate": 1.0303027002633622e-05, "loss": 1.8051592633128166e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 495, "tokens/total": 14981552, "tokens/train_per_sec_per_gpu": 36.46, "tokens/trainable": 226281}
-{"epoch": 1.9375, "grad_norm": 0.0015804325230419636, "learning_rate": 1.0270325419886884e-05, "loss": 1.691515171842184e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 496, "tokens/total": 15011584, "tokens/train_per_sec_per_gpu": 36.72, "tokens/trainable": 226735}
-{"epoch": 1.94140625, "grad_norm": 0.0036208343226462603, "learning_rate": 1.0239485221478599e-05, "loss": 4.592180994222872e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00005, "step": 497, "tokens/total": 15042016, "tokens/train_per_sec_per_gpu": 36.31, "tokens/trainable": 227198}
-{"epoch": 1.9453125, "grad_norm": 0.0005918977549299598, "learning_rate": 1.0210507690795292e-05, "loss": 1.5305626220651902e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.68, "memory/max_allocated (GiB)": 33.68, "ppl": 1.00002, "step": 498, "tokens/total": 15071984, "tokens/train_per_sec_per_gpu": 31.07, "tokens/trainable": 227623}
-{"epoch": 1.94921875, "grad_norm": 0.0013667694292962551, "learning_rate": 1.0183394033710305e-05, "loss": 1.3490713172359392e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00001, "step": 499, "tokens/total": 15102192, "tokens/train_per_sec_per_gpu": 32.15, "tokens/trainable": 228074}
-{"epoch": 1.953125, "grad_norm": 0.0027930096257478, "learning_rate": 1.0158145378533583e-05, "loss": 2.8409334845491685e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00003, "step": 500, "tokens/total": 15132560, "tokens/train_per_sec_per_gpu": 35.57, "tokens/trainable": 228541}
-{"epoch": 1.95703125, "grad_norm": 0.0028320043347775936, "learning_rate": 1.0134762775964726e-05, "loss": 3.1829928047955036e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00003, "step": 501, "tokens/total": 15163152, "tokens/train_per_sec_per_gpu": 37.77, "tokens/trainable": 229031}
-{"epoch": 1.9609375, "grad_norm": 0.0030925904866307974, "learning_rate": 1.0113247199049278e-05, "loss": 2.238066190329846e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 502, "tokens/total": 15193456, "tokens/train_per_sec_per_gpu": 38.68, "tokens/trainable": 229512}
-{"epoch": 1.96484375, "grad_norm": 0.06389517337083817, "learning_rate": 1.0093599543138205e-05, "loss": 0.0004140711098443717, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00041, "step": 503, "tokens/total": 15223680, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 229989}
-{"epoch": 1.96875, "grad_norm": 0.003928070887923241, "learning_rate": 1.0075820625850675e-05, "loss": 4.3977754103252664e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00004, "step": 504, "tokens/total": 15254336, "tokens/train_per_sec_per_gpu": 35.81, "tokens/trainable": 230447}
-{"epoch": 1.97265625, "grad_norm": 0.0022790853399783373, "learning_rate": 1.0059911187040013e-05, "loss": 2.180483534175437e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00002, "step": 505, "tokens/total": 15284496, "tokens/train_per_sec_per_gpu": 34.29, "tokens/trainable": 230879}
-{"epoch": 1.9765625, "grad_norm": 0.0022161444649100304, "learning_rate": 1.0045871888762893e-05, "loss": 2.00994672923116e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00002, "step": 506, "tokens/total": 15314976, "tokens/train_per_sec_per_gpu": 34.12, "tokens/trainable": 231348}
-{"epoch": 1.98046875, "grad_norm": 0.0109304990619421, "learning_rate": 1.003370331525184e-05, "loss": 9.907589264912531e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0001, "step": 507, "tokens/total": 15345280, "tokens/train_per_sec_per_gpu": 35.22, "tokens/trainable": 231819}
-{"epoch": 1.984375, "grad_norm": 0.0008610524819232523, "learning_rate": 1.002340597289085e-05, "loss": 1.4185266991262324e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00001, "step": 508, "tokens/total": 15375424, "tokens/train_per_sec_per_gpu": 34.57, "tokens/trainable": 232269}
-{"epoch": 1.98828125, "grad_norm": 0.18468032777309418, "learning_rate": 1.0014980290194387e-05, "loss": 0.0011317847529426217, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00113, "step": 509, "tokens/total": 15405872, "tokens/train_per_sec_per_gpu": 34.06, "tokens/trainable": 232716}
-{"epoch": 1.9921875, "grad_norm": 0.0012291419552639127, "learning_rate": 1.0008426617789489e-05, "loss": 1.4661756722489372e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 510, "tokens/total": 15436128, "tokens/train_per_sec_per_gpu": 36.09, "tokens/trainable": 233178}
-{"epoch": 1.99609375, "grad_norm": 0.003051232313737273, "learning_rate": 1.0003745228401215e-05, "loss": 2.0963430870324373e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 511, "tokens/total": 15466464, "tokens/train_per_sec_per_gpu": 35.95, "tokens/trainable": 233647}
-{"epoch": 2.0, "grad_norm": 0.001929801655933261, "learning_rate": 1.0000936316841296e-05, "loss": 1.9281407730886713e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.67, "memory/max_allocated (GiB)": 33.67, "ppl": 1.00002, "step": 512, "tokens/total": 15496368, "tokens/train_per_sec_per_gpu": 29.42, "tokens/trainable": 234060}
+{"epoch": 0.00390625, "grad_norm": 0.8591235280036926, "learning_rate": 0.0, "loss": 0.10251401364803314, "memory/device_reserved (GiB)": 34.94, "memory/max_active (GiB)": 32.98, "memory/max_allocated (GiB)": 32.98, "ppl": 1.10795, "step": 1, "tokens/total": 30464, "tokens/train_per_sec_per_gpu": 30.01, "tokens/trainable": 470}
+{"epoch": 0.0078125, "grad_norm": 0.8033757209777832, "learning_rate": 4.000000000000001e-06, "loss": 0.09678442776203156, "memory/device_reserved (GiB)": 35.83, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.10162, "step": 2, "tokens/total": 60800, "tokens/train_per_sec_per_gpu": 35.1, "tokens/trainable": 922}
+{"epoch": 0.01171875, "grad_norm": 1.0102914571762085, "learning_rate": 8.000000000000001e-06, "loss": 0.10876177996397018, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.1149, "step": 3, "tokens/total": 91136, "tokens/train_per_sec_per_gpu": 34.67, "tokens/trainable": 1396}
+{"epoch": 0.015625, "grad_norm": 2.2042534351348877, "learning_rate": 1.2e-05, "loss": 0.1031389832496643, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.10865, "step": 4, "tokens/total": 121552, "tokens/train_per_sec_per_gpu": 38.0, "tokens/trainable": 1850}
+{"epoch": 0.01953125, "grad_norm": 2.5755860805511475, "learning_rate": 1.6000000000000003e-05, "loss": 0.1097191572189331, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.11596, "step": 5, "tokens/total": 152304, "tokens/train_per_sec_per_gpu": 34.92, "tokens/trainable": 2302}
+{"epoch": 0.0234375, "grad_norm": 1.498002052307129, "learning_rate": 2e-05, "loss": 0.08722387254238129, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.09114, "step": 6, "tokens/total": 182448, "tokens/train_per_sec_per_gpu": 31.65, "tokens/trainable": 2742}
+{"epoch": 0.02734375, "grad_norm": 1.280110478401184, "learning_rate": 2.4e-05, "loss": 0.07383580505847931, "memory/device_reserved (GiB)": 35.85, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.07663, "step": 7, "tokens/total": 212736, "tokens/train_per_sec_per_gpu": 36.7, "tokens/trainable": 3208}
+{"epoch": 0.03125, "grad_norm": 1.6952791213989258, "learning_rate": 2.8000000000000003e-05, "loss": 0.08001460134983063, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.0833, "step": 8, "tokens/total": 243024, "tokens/train_per_sec_per_gpu": 31.94, "tokens/trainable": 3655}
+{"epoch": 0.03515625, "grad_norm": 1.2223162651062012, "learning_rate": 3.2000000000000005e-05, "loss": 0.047361284494400024, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0485, "step": 9, "tokens/total": 271264, "tokens/train_per_sec_per_gpu": 35.78, "tokens/trainable": 4091}
+{"epoch": 0.0390625, "grad_norm": 2.902157783508301, "learning_rate": 3.6e-05, "loss": 0.09570425748825073, "memory/device_reserved (GiB)": 36.07, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.10043, "step": 10, "tokens/total": 301520, "tokens/train_per_sec_per_gpu": 34.98, "tokens/trainable": 4553}
+{"epoch": 0.04296875, "grad_norm": 2.1767208576202393, "learning_rate": 4e-05, "loss": 0.0828268975019455, "memory/device_reserved (GiB)": 36.08, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.08635, "step": 11, "tokens/total": 331808, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 4992}
+{"epoch": 0.046875, "grad_norm": 3.15231990814209, "learning_rate": 4.4000000000000006e-05, "loss": 0.12141455709934235, "memory/device_reserved (GiB)": 36.08, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.12909, "step": 12, "tokens/total": 362224, "tokens/train_per_sec_per_gpu": 38.32, "tokens/trainable": 5494}
+{"epoch": 0.05078125, "grad_norm": 2.3834526538848877, "learning_rate": 4.8e-05, "loss": 0.037937015295028687, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.03867, "step": 13, "tokens/total": 392432, "tokens/train_per_sec_per_gpu": 35.04, "tokens/trainable": 5933}
+{"epoch": 0.0546875, "grad_norm": 3.2587738037109375, "learning_rate": 5.2000000000000004e-05, "loss": 0.06514571607112885, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.06731, "step": 14, "tokens/total": 422784, "tokens/train_per_sec_per_gpu": 29.67, "tokens/trainable": 6369}
+{"epoch": 0.05859375, "grad_norm": 1.5818979740142822, "learning_rate": 5.6000000000000006e-05, "loss": 0.05656517297029495, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.0582, "step": 15, "tokens/total": 453376, "tokens/train_per_sec_per_gpu": 33.08, "tokens/trainable": 6820}
+{"epoch": 0.0625, "grad_norm": 1.4215443134307861, "learning_rate": 6e-05, "loss": 0.06070145219564438, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.06258, "step": 16, "tokens/total": 483504, "tokens/train_per_sec_per_gpu": 30.85, "tokens/trainable": 7258}
+{"epoch": 0.06640625, "grad_norm": 1.3065693378448486, "learning_rate": 6.400000000000001e-05, "loss": 0.05027393624186516, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.05156, "step": 17, "tokens/total": 514032, "tokens/train_per_sec_per_gpu": 35.29, "tokens/trainable": 7710}
+{"epoch": 0.0703125, "grad_norm": 0.6123363375663757, "learning_rate": 6.800000000000001e-05, "loss": 0.028499452397227287, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.02891, "step": 18, "tokens/total": 544432, "tokens/train_per_sec_per_gpu": 31.76, "tokens/trainable": 8174}
+{"epoch": 0.07421875, "grad_norm": 0.648410439491272, "learning_rate": 7.2e-05, "loss": 0.031143227592110634, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.03163, "step": 19, "tokens/total": 574880, "tokens/train_per_sec_per_gpu": 34.47, "tokens/trainable": 8617}
+{"epoch": 0.078125, "grad_norm": 0.6737155318260193, "learning_rate": 7.6e-05, "loss": 0.030753308907151222, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.03123, "step": 20, "tokens/total": 605408, "tokens/train_per_sec_per_gpu": 30.37, "tokens/trainable": 9037}
+{"epoch": 0.08203125, "grad_norm": 0.7464718222618103, "learning_rate": 8e-05, "loss": 0.02451065182685852, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.02481, "step": 21, "tokens/total": 635584, "tokens/train_per_sec_per_gpu": 36.63, "tokens/trainable": 9503}
+{"epoch": 0.0859375, "grad_norm": 0.9435750246047974, "learning_rate": 8.4e-05, "loss": 0.017626767978072166, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01778, "step": 22, "tokens/total": 665952, "tokens/train_per_sec_per_gpu": 36.79, "tokens/trainable": 9972}
+{"epoch": 0.08984375, "grad_norm": 0.6369844079017639, "learning_rate": 8.800000000000001e-05, "loss": 0.013959845528006554, "memory/device_reserved (GiB)": 36.21, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01406, "step": 23, "tokens/total": 696240, "tokens/train_per_sec_per_gpu": 37.54, "tokens/trainable": 10447}
+{"epoch": 0.09375, "grad_norm": 1.0666130781173706, "learning_rate": 9.200000000000001e-05, "loss": 0.03303435444831848, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.03359, "step": 24, "tokens/total": 726464, "tokens/train_per_sec_per_gpu": 37.23, "tokens/trainable": 10899}
+{"epoch": 0.09765625, "grad_norm": 1.0491507053375244, "learning_rate": 9.6e-05, "loss": 0.030840374529361725, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.03132, "step": 25, "tokens/total": 756704, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 11360}
+{"epoch": 0.1015625, "grad_norm": 0.8108148574829102, "learning_rate": 0.0001, "loss": 0.03408072516322136, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.03467, "step": 26, "tokens/total": 786912, "tokens/train_per_sec_per_gpu": 29.3, "tokens/trainable": 11775}
+{"epoch": 0.10546875, "grad_norm": 0.507036030292511, "learning_rate": 9.99990636831587e-05, "loss": 0.014000408351421356, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.46, "memory/max_allocated (GiB)": 33.46, "ppl": 1.0141, "step": 27, "tokens/total": 815424, "tokens/train_per_sec_per_gpu": 39.89, "tokens/trainable": 12242}
+{"epoch": 0.109375, "grad_norm": 0.618457555770874, "learning_rate": 9.999625477159879e-05, "loss": 0.022290699183940887, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.02254, "step": 28, "tokens/total": 845584, "tokens/train_per_sec_per_gpu": 33.48, "tokens/trainable": 12696}
+{"epoch": 0.11328125, "grad_norm": 0.4989492893218994, "learning_rate": 9.999157338221051e-05, "loss": 0.025926023721694946, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.02627, "step": 29, "tokens/total": 875808, "tokens/train_per_sec_per_gpu": 34.34, "tokens/trainable": 13153}
+{"epoch": 0.1171875, "grad_norm": 0.3578357696533203, "learning_rate": 9.998501970980562e-05, "loss": 0.014475762844085693, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01458, "step": 30, "tokens/total": 904096, "tokens/train_per_sec_per_gpu": 39.7, "tokens/trainable": 13624}
+{"epoch": 0.12109375, "grad_norm": 0.4404466450214386, "learning_rate": 9.997659402710915e-05, "loss": 0.015927618369460106, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01606, "step": 31, "tokens/total": 934400, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 14064}
+{"epoch": 0.125, "grad_norm": 0.5913511514663696, "learning_rate": 9.996629668474818e-05, "loss": 0.019408874213695526, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.0196, "step": 32, "tokens/total": 964832, "tokens/train_per_sec_per_gpu": 38.92, "tokens/trainable": 14565}
+{"epoch": 0.12890625, "grad_norm": 0.2795853614807129, "learning_rate": 9.995412811123711e-05, "loss": 0.007002322003245354, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00703, "step": 33, "tokens/total": 995040, "tokens/train_per_sec_per_gpu": 33.77, "tokens/trainable": 15005}
+{"epoch": 0.1328125, "grad_norm": 1.1757413148880005, "learning_rate": 9.994008881295999e-05, "loss": 0.021448152139782906, "memory/device_reserved (GiB)": 34.72, "memory/max_active (GiB)": 33.64, "memory/max_allocated (GiB)": 33.64, "ppl": 1.02168, "step": 34, "tokens/total": 1024896, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 15459}
+{"epoch": 0.13671875, "grad_norm": 0.3860406279563904, "learning_rate": 9.992417937414932e-05, "loss": 0.01894465833902359, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01913, "step": 35, "tokens/total": 1054992, "tokens/train_per_sec_per_gpu": 38.21, "tokens/trainable": 15960}
+{"epoch": 0.140625, "grad_norm": 0.4803963005542755, "learning_rate": 9.99064004568618e-05, "loss": 0.013810254633426666, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.01391, "step": 36, "tokens/total": 1085280, "tokens/train_per_sec_per_gpu": 34.59, "tokens/trainable": 16428}
+{"epoch": 0.14453125, "grad_norm": 0.3357454836368561, "learning_rate": 9.988675280095074e-05, "loss": 0.008235199376940727, "memory/device_reserved (GiB)": 35.49, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00827, "step": 37, "tokens/total": 1115504, "tokens/train_per_sec_per_gpu": 31.89, "tokens/trainable": 16884}
+{"epoch": 0.1484375, "grad_norm": 0.9474877715110779, "learning_rate": 9.986523722403528e-05, "loss": 0.019564270973205566, "memory/device_reserved (GiB)": 35.49, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01976, "step": 38, "tokens/total": 1145712, "tokens/train_per_sec_per_gpu": 32.02, "tokens/trainable": 17341}
+{"epoch": 0.15234375, "grad_norm": 1.101553201675415, "learning_rate": 9.984185462146642e-05, "loss": 0.017880277708172798, "memory/device_reserved (GiB)": 35.49, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.01804, "step": 39, "tokens/total": 1176128, "tokens/train_per_sec_per_gpu": 34.16, "tokens/trainable": 17786}
+{"epoch": 0.15625, "grad_norm": 0.7563315033912659, "learning_rate": 9.98166059662897e-05, "loss": 0.019336596131324768, "memory/device_reserved (GiB)": 35.49, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01952, "step": 40, "tokens/total": 1206576, "tokens/train_per_sec_per_gpu": 35.01, "tokens/trainable": 18262}
+{"epoch": 0.16015625, "grad_norm": 0.41576531529426575, "learning_rate": 9.978949230920472e-05, "loss": 0.011620002798736095, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01169, "step": 41, "tokens/total": 1236848, "tokens/train_per_sec_per_gpu": 32.74, "tokens/trainable": 18716}
+{"epoch": 0.1640625, "grad_norm": 1.180986762046814, "learning_rate": 9.976051477852141e-05, "loss": 0.04661658778786659, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.04772, "step": 42, "tokens/total": 1267088, "tokens/train_per_sec_per_gpu": 32.88, "tokens/trainable": 19157}
+{"epoch": 0.16796875, "grad_norm": 0.7583066821098328, "learning_rate": 9.972967458011312e-05, "loss": 0.029224852100014687, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.02966, "step": 43, "tokens/total": 1297360, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 19647}
+{"epoch": 0.171875, "grad_norm": 0.18861345946788788, "learning_rate": 9.96969729973664e-05, "loss": 0.007798851002007723, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00783, "step": 44, "tokens/total": 1327744, "tokens/train_per_sec_per_gpu": 34.91, "tokens/trainable": 20119}
+{"epoch": 0.17578125, "grad_norm": 0.35095125436782837, "learning_rate": 9.966241139112754e-05, "loss": 0.012044938281178474, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01212, "step": 45, "tokens/total": 1358096, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 20560}
+{"epoch": 0.1796875, "grad_norm": 0.5071477890014648, "learning_rate": 9.96259911996461e-05, "loss": 0.017856616526842117, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01802, "step": 46, "tokens/total": 1388240, "tokens/train_per_sec_per_gpu": 32.63, "tokens/trainable": 20992}
+{"epoch": 0.18359375, "grad_norm": 0.5569872260093689, "learning_rate": 9.958771393851491e-05, "loss": 0.019440822303295135, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.26, "memory/max_allocated (GiB)": 33.26, "ppl": 1.01963, "step": 47, "tokens/total": 1416240, "tokens/train_per_sec_per_gpu": 31.55, "tokens/trainable": 21441}
+{"epoch": 0.1875, "grad_norm": 0.42062458395957947, "learning_rate": 9.954758120060702e-05, "loss": 0.013018792495131493, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0131, "step": 48, "tokens/total": 1446880, "tokens/train_per_sec_per_gpu": 33.68, "tokens/trainable": 21901}
+{"epoch": 0.19140625, "grad_norm": 0.30396750569343567, "learning_rate": 9.950559465600948e-05, "loss": 0.0077492957934737206, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00778, "step": 49, "tokens/total": 1477168, "tokens/train_per_sec_per_gpu": 33.33, "tokens/trainable": 22341}
+{"epoch": 0.1953125, "grad_norm": 2.044849395751953, "learning_rate": 9.946175605195379e-05, "loss": 0.014447445049881935, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01455, "step": 50, "tokens/total": 1507712, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 22833}
+{"epoch": 0.19921875, "grad_norm": 0.9357694983482361, "learning_rate": 9.941606721274322e-05, "loss": 0.012720856815576553, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.0128, "step": 51, "tokens/total": 1537936, "tokens/train_per_sec_per_gpu": 30.72, "tokens/trainable": 23277}
+{"epoch": 0.203125, "grad_norm": 0.2881873548030853, "learning_rate": 9.936853003967685e-05, "loss": 0.005877626594156027, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00589, "step": 52, "tokens/total": 1568352, "tokens/train_per_sec_per_gpu": 35.63, "tokens/trainable": 23759}
+{"epoch": 0.20703125, "grad_norm": 0.7577692270278931, "learning_rate": 9.93191465109705e-05, "loss": 0.01451955921947956, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01463, "step": 53, "tokens/total": 1598992, "tokens/train_per_sec_per_gpu": 32.95, "tokens/trainable": 24193}
+{"epoch": 0.2109375, "grad_norm": 0.5201014280319214, "learning_rate": 9.926791868167438e-05, "loss": 0.006856432184576988, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00688, "step": 54, "tokens/total": 1629488, "tokens/train_per_sec_per_gpu": 33.59, "tokens/trainable": 24619}
+{"epoch": 0.21484375, "grad_norm": 0.8399921655654907, "learning_rate": 9.921484868358753e-05, "loss": 0.037967782467603683, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0387, "step": 55, "tokens/total": 1659696, "tokens/train_per_sec_per_gpu": 32.85, "tokens/trainable": 25075}
+{"epoch": 0.21875, "grad_norm": 0.8203506469726562, "learning_rate": 9.915993872516924e-05, "loss": 0.012814272195100784, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.0129, "step": 56, "tokens/total": 1689872, "tokens/train_per_sec_per_gpu": 31.55, "tokens/trainable": 25519}
+{"epoch": 0.22265625, "grad_norm": 0.20874246954917908, "learning_rate": 9.9103191091447e-05, "loss": 0.00539036700502038, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0054, "step": 57, "tokens/total": 1720144, "tokens/train_per_sec_per_gpu": 32.69, "tokens/trainable": 25966}
+{"epoch": 0.2265625, "grad_norm": 0.4427756071090698, "learning_rate": 9.904460814392147e-05, "loss": 0.009390819817781448, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00944, "step": 58, "tokens/total": 1750448, "tokens/train_per_sec_per_gpu": 33.98, "tokens/trainable": 26423}
+{"epoch": 0.23046875, "grad_norm": 0.7457786798477173, "learning_rate": 9.898419232046825e-05, "loss": 0.019530881196260452, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.01972, "step": 59, "tokens/total": 1781088, "tokens/train_per_sec_per_gpu": 38.51, "tokens/trainable": 26934}
+{"epoch": 0.234375, "grad_norm": 0.13402195274829865, "learning_rate": 9.892194613523633e-05, "loss": 0.0014544172445312142, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00146, "step": 60, "tokens/total": 1811344, "tokens/train_per_sec_per_gpu": 32.96, "tokens/trainable": 27393}
+{"epoch": 0.23828125, "grad_norm": 1.0385031700134277, "learning_rate": 9.885787217854357e-05, "loss": 0.019916968420147896, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.02012, "step": 61, "tokens/total": 1841600, "tokens/train_per_sec_per_gpu": 32.94, "tokens/trainable": 27852}
+{"epoch": 0.2421875, "grad_norm": 1.2596747875213623, "learning_rate": 9.879197311676887e-05, "loss": 0.015884939581155777, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01601, "step": 62, "tokens/total": 1872048, "tokens/train_per_sec_per_gpu": 32.96, "tokens/trainable": 28292}
+{"epoch": 0.24609375, "grad_norm": 0.14031143486499786, "learning_rate": 9.872425169224113e-05, "loss": 0.002796083688735962, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.0028, "step": 63, "tokens/total": 1902592, "tokens/train_per_sec_per_gpu": 37.36, "tokens/trainable": 28798}
+{"epoch": 0.25, "grad_norm": 0.6247786283493042, "learning_rate": 9.865471072312528e-05, "loss": 0.017117884010076523, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01727, "step": 64, "tokens/total": 1933088, "tokens/train_per_sec_per_gpu": 35.79, "tokens/trainable": 29283}
+{"epoch": 0.25390625, "grad_norm": 0.3513385057449341, "learning_rate": 9.858335310330492e-05, "loss": 0.008029159158468246, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00806, "step": 65, "tokens/total": 1963296, "tokens/train_per_sec_per_gpu": 31.99, "tokens/trainable": 29745}
+{"epoch": 0.2578125, "grad_norm": 0.279448539018631, "learning_rate": 9.851018180226185e-05, "loss": 0.0161186084151268, "memory/device_reserved (GiB)": 34.87, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.01625, "step": 66, "tokens/total": 1993760, "tokens/train_per_sec_per_gpu": 31.19, "tokens/trainable": 30186}
+{"epoch": 0.26171875, "grad_norm": 0.31739094853401184, "learning_rate": 9.843519986495259e-05, "loss": 0.009091717191040516, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00913, "step": 67, "tokens/total": 2024144, "tokens/train_per_sec_per_gpu": 33.36, "tokens/trainable": 30622}
+{"epoch": 0.265625, "grad_norm": 0.3539387285709381, "learning_rate": 9.835841041168162e-05, "loss": 0.00908343680202961, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00912, "step": 68, "tokens/total": 2054608, "tokens/train_per_sec_per_gpu": 35.52, "tokens/trainable": 31097}
+{"epoch": 0.26953125, "grad_norm": 0.6497699618339539, "learning_rate": 9.82798166379715e-05, "loss": 0.03086906298995018, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.03135, "step": 69, "tokens/total": 2084912, "tokens/train_per_sec_per_gpu": 36.6, "tokens/trainable": 31595}
+{"epoch": 0.2734375, "grad_norm": 0.19664841890335083, "learning_rate": 9.819942181443002e-05, "loss": 0.0039155250415205956, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00392, "step": 70, "tokens/total": 2115216, "tokens/train_per_sec_per_gpu": 30.75, "tokens/trainable": 32036}
+{"epoch": 0.27734375, "grad_norm": 0.4300064146518707, "learning_rate": 9.811722928661392e-05, "loss": 0.007546662352979183, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00758, "step": 71, "tokens/total": 2145424, "tokens/train_per_sec_per_gpu": 28.09, "tokens/trainable": 32428}
+{"epoch": 0.28125, "grad_norm": 0.027750927954912186, "learning_rate": 9.803324247488975e-05, "loss": 0.0004817460721824318, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00048, "step": 72, "tokens/total": 2175648, "tokens/train_per_sec_per_gpu": 32.11, "tokens/trainable": 32880}
+{"epoch": 0.28515625, "grad_norm": 0.11202923208475113, "learning_rate": 9.794746487429161e-05, "loss": 0.0012140646576881409, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00121, "step": 73, "tokens/total": 2205856, "tokens/train_per_sec_per_gpu": 33.51, "tokens/trainable": 33323}
+{"epoch": 0.2890625, "grad_norm": 0.026963796466588974, "learning_rate": 9.785990005437554e-05, "loss": 0.00046908354852348566, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00047, "step": 74, "tokens/total": 2236352, "tokens/train_per_sec_per_gpu": 35.86, "tokens/trainable": 33792}
+{"epoch": 0.29296875, "grad_norm": 0.5172365307807922, "learning_rate": 9.777055165907117e-05, "loss": 0.029645610600709915, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.03009, "step": 75, "tokens/total": 2266480, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 34223}
+{"epoch": 0.296875, "grad_norm": 0.84085613489151, "learning_rate": 9.767942340652993e-05, "loss": 0.006980063393712044, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.007, "step": 76, "tokens/total": 2296496, "tokens/train_per_sec_per_gpu": 29.61, "tokens/trainable": 34649}
+{"epoch": 0.30078125, "grad_norm": 3.4935519695281982, "learning_rate": 9.758651908897035e-05, "loss": 0.008870027028024197, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00891, "step": 77, "tokens/total": 2327040, "tokens/train_per_sec_per_gpu": 35.64, "tokens/trainable": 35094}
+{"epoch": 0.3046875, "grad_norm": 0.9529178142547607, "learning_rate": 9.749184257252033e-05, "loss": 0.032071419060230255, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.03259, "step": 78, "tokens/total": 2357328, "tokens/train_per_sec_per_gpu": 31.82, "tokens/trainable": 35534}
+{"epoch": 0.30859375, "grad_norm": 0.5781691074371338, "learning_rate": 9.739539779705614e-05, "loss": 0.01475254725664854, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01486, "step": 79, "tokens/total": 2387664, "tokens/train_per_sec_per_gpu": 35.82, "tokens/trainable": 36029}
+{"epoch": 0.3125, "grad_norm": 0.15085959434509277, "learning_rate": 9.729718877603861e-05, "loss": 0.002578896936029196, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00258, "step": 80, "tokens/total": 2418000, "tokens/train_per_sec_per_gpu": 35.53, "tokens/trainable": 36469}
+{"epoch": 0.31640625, "grad_norm": 0.19004814326763153, "learning_rate": 9.719721959634592e-05, "loss": 0.004292497877031565, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0043, "step": 81, "tokens/total": 2448720, "tokens/train_per_sec_per_gpu": 30.75, "tokens/trainable": 36906}
+{"epoch": 0.3203125, "grad_norm": 0.4505981504917145, "learning_rate": 9.709549441810375e-05, "loss": 0.018529793247580528, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.0187, "step": 82, "tokens/total": 2479248, "tokens/train_per_sec_per_gpu": 38.95, "tokens/trainable": 37390}
+{"epoch": 0.32421875, "grad_norm": 0.4981430470943451, "learning_rate": 9.699201747451195e-05, "loss": 0.014818452298641205, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.01493, "step": 83, "tokens/total": 2509408, "tokens/train_per_sec_per_gpu": 30.8, "tokens/trainable": 37838}
+{"epoch": 0.328125, "grad_norm": 0.16379471123218536, "learning_rate": 9.688679307166854e-05, "loss": 0.003185997949913144, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00319, "step": 84, "tokens/total": 2539776, "tokens/train_per_sec_per_gpu": 35.02, "tokens/trainable": 38310}
+{"epoch": 0.33203125, "grad_norm": 0.40732133388519287, "learning_rate": 9.677982558839042e-05, "loss": 0.020803559571504593, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.02102, "step": 85, "tokens/total": 2569840, "tokens/train_per_sec_per_gpu": 34.03, "tokens/trainable": 38789}
+{"epoch": 0.3359375, "grad_norm": 0.3687220513820648, "learning_rate": 9.66711194760312e-05, "loss": 0.012931328266859055, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01302, "step": 86, "tokens/total": 2600192, "tokens/train_per_sec_per_gpu": 36.36, "tokens/trainable": 39277}
+{"epoch": 0.33984375, "grad_norm": 0.367418110370636, "learning_rate": 9.656067925829593e-05, "loss": 0.01382569782435894, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.01392, "step": 87, "tokens/total": 2630640, "tokens/train_per_sec_per_gpu": 35.68, "tokens/trainable": 39737}
+{"epoch": 0.34375, "grad_norm": 0.2704487144947052, "learning_rate": 9.644850953105288e-05, "loss": 0.008717816323041916, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00876, "step": 88, "tokens/total": 2660832, "tokens/train_per_sec_per_gpu": 29.6, "tokens/trainable": 40179}
+{"epoch": 0.34765625, "grad_norm": 3.374918222427368, "learning_rate": 9.633461496214225e-05, "loss": 0.01938408613204956, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01957, "step": 89, "tokens/total": 2691328, "tokens/train_per_sec_per_gpu": 32.66, "tokens/trainable": 40649}
+{"epoch": 0.3515625, "grad_norm": 0.4690157175064087, "learning_rate": 9.621900029118195e-05, "loss": 0.02013186179101467, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.41, "memory/max_allocated (GiB)": 33.41, "ppl": 1.02034, "step": 90, "tokens/total": 2719696, "tokens/train_per_sec_per_gpu": 31.46, "tokens/trainable": 41080}
+{"epoch": 0.35546875, "grad_norm": 0.08705829828977585, "learning_rate": 9.610167032937036e-05, "loss": 0.002885152120143175, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00289, "step": 91, "tokens/total": 2750272, "tokens/train_per_sec_per_gpu": 38.11, "tokens/trainable": 41599}
+{"epoch": 0.359375, "grad_norm": 8.197907447814941, "learning_rate": 9.598262995928611e-05, "loss": 0.00822490081191063, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00826, "step": 92, "tokens/total": 2780656, "tokens/train_per_sec_per_gpu": 36.82, "tokens/trainable": 42076}
+{"epoch": 0.36328125, "grad_norm": 0.14939527213573456, "learning_rate": 9.586188413468492e-05, "loss": 0.004234164021909237, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00424, "step": 93, "tokens/total": 2811088, "tokens/train_per_sec_per_gpu": 33.71, "tokens/trainable": 42522}
+{"epoch": 0.3671875, "grad_norm": 0.15404288470745087, "learning_rate": 9.57394378802934e-05, "loss": 0.009490479715168476, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00954, "step": 94, "tokens/total": 2841520, "tokens/train_per_sec_per_gpu": 34.4, "tokens/trainable": 42974}
+{"epoch": 0.37109375, "grad_norm": 0.5274825692176819, "learning_rate": 9.56152962916e-05, "loss": 0.02413639798760414, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.02443, "step": 95, "tokens/total": 2871600, "tokens/train_per_sec_per_gpu": 30.67, "tokens/trainable": 43380}
+{"epoch": 0.375, "grad_norm": 0.5182522535324097, "learning_rate": 9.548946453464296e-05, "loss": 0.009927366860210896, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00998, "step": 96, "tokens/total": 2902144, "tokens/train_per_sec_per_gpu": 34.14, "tokens/trainable": 43865}
+{"epoch": 0.37890625, "grad_norm": 0.5578822493553162, "learning_rate": 9.53619478457953e-05, "loss": 0.014485684223473072, "memory/device_reserved (GiB)": 36.1, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01459, "step": 97, "tokens/total": 2932272, "tokens/train_per_sec_per_gpu": 31.94, "tokens/trainable": 44328}
+{"epoch": 0.3828125, "grad_norm": 0.10520734637975693, "learning_rate": 9.523275153154695e-05, "loss": 0.0021552909165620804, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00216, "step": 98, "tokens/total": 2962032, "tokens/train_per_sec_per_gpu": 30.95, "tokens/trainable": 44778}
+{"epoch": 0.38671875, "grad_norm": 0.7544558644294739, "learning_rate": 9.51018809682839e-05, "loss": 0.01703210547566414, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01718, "step": 99, "tokens/total": 2992256, "tokens/train_per_sec_per_gpu": 37.12, "tokens/trainable": 45225}
+{"epoch": 0.390625, "grad_norm": 0.3857012689113617, "learning_rate": 9.49693416020645e-05, "loss": 0.00604570098221302, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00606, "step": 100, "tokens/total": 3022608, "tokens/train_per_sec_per_gpu": 35.75, "tokens/trainable": 45678}
+{"epoch": 0.39453125, "grad_norm": 0.21589453518390656, "learning_rate": 9.483513894839276e-05, "loss": 0.005753816105425358, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00577, "step": 101, "tokens/total": 3052992, "tokens/train_per_sec_per_gpu": 32.72, "tokens/trainable": 46125}
+{"epoch": 0.3984375, "grad_norm": 1.1246687173843384, "learning_rate": 9.469927859198888e-05, "loss": 0.006994037888944149, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00702, "step": 102, "tokens/total": 3083392, "tokens/train_per_sec_per_gpu": 39.6, "tokens/trainable": 46612}
+{"epoch": 0.40234375, "grad_norm": 0.267713725566864, "learning_rate": 9.456176618655689e-05, "loss": 0.013721915893256664, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01382, "step": 103, "tokens/total": 3113744, "tokens/train_per_sec_per_gpu": 32.41, "tokens/trainable": 47059}
+{"epoch": 0.40625, "grad_norm": 0.325897753238678, "learning_rate": 9.442260745454927e-05, "loss": 0.012948594987392426, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.01303, "step": 104, "tokens/total": 3144272, "tokens/train_per_sec_per_gpu": 33.93, "tokens/trainable": 47536}
+{"epoch": 0.41015625, "grad_norm": 0.531863808631897, "learning_rate": 9.428180818692884e-05, "loss": 0.02244492992758751, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.0227, "step": 105, "tokens/total": 3174512, "tokens/train_per_sec_per_gpu": 33.95, "tokens/trainable": 48037}
+{"epoch": 0.4140625, "grad_norm": 0.3957497775554657, "learning_rate": 9.413937424292791e-05, "loss": 0.015801995992660522, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01593, "step": 106, "tokens/total": 3204896, "tokens/train_per_sec_per_gpu": 32.49, "tokens/trainable": 48491}
+{"epoch": 0.41796875, "grad_norm": 0.4241825044155121, "learning_rate": 9.399531154980424e-05, "loss": 0.016320914030075073, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.01645, "step": 107, "tokens/total": 3235360, "tokens/train_per_sec_per_gpu": 32.89, "tokens/trainable": 48940}
+{"epoch": 0.421875, "grad_norm": 0.32064536213874817, "learning_rate": 9.384962610259455e-05, "loss": 0.010785036720335484, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01084, "step": 108, "tokens/total": 3265552, "tokens/train_per_sec_per_gpu": 31.52, "tokens/trainable": 49389}
+{"epoch": 0.42578125, "grad_norm": 0.17215749621391296, "learning_rate": 9.370232396386494e-05, "loss": 0.00814715214073658, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.00818, "step": 109, "tokens/total": 3293856, "tokens/train_per_sec_per_gpu": 35.91, "tokens/trainable": 49838}
+{"epoch": 0.4296875, "grad_norm": 0.38179653882980347, "learning_rate": 9.355341126345868e-05, "loss": 0.012128787115216255, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.0122, "step": 110, "tokens/total": 3323984, "tokens/train_per_sec_per_gpu": 33.93, "tokens/trainable": 50310}
+{"epoch": 0.43359375, "grad_norm": 0.49835413694381714, "learning_rate": 9.340289419824107e-05, "loss": 0.015248995274305344, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01537, "step": 111, "tokens/total": 3354320, "tokens/train_per_sec_per_gpu": 33.55, "tokens/trainable": 50755}
+{"epoch": 0.4375, "grad_norm": 0.43904298543930054, "learning_rate": 9.325077903184159e-05, "loss": 0.011272929608821869, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01134, "step": 112, "tokens/total": 3384880, "tokens/train_per_sec_per_gpu": 31.5, "tokens/trainable": 51213}
+{"epoch": 0.44140625, "grad_norm": 0.5670213103294373, "learning_rate": 9.30970720943932e-05, "loss": 0.0028921598568558693, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0029, "step": 113, "tokens/total": 3415104, "tokens/train_per_sec_per_gpu": 33.77, "tokens/trainable": 51660}
+{"epoch": 0.4453125, "grad_norm": 0.159476637840271, "learning_rate": 9.2941779782269e-05, "loss": 0.0025574099272489548, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00256, "step": 114, "tokens/total": 3445504, "tokens/train_per_sec_per_gpu": 32.34, "tokens/trainable": 52133}
+{"epoch": 0.44921875, "grad_norm": 0.795085608959198, "learning_rate": 9.278490855781596e-05, "loss": 0.024456799030303955, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02476, "step": 115, "tokens/total": 3475744, "tokens/train_per_sec_per_gpu": 37.25, "tokens/trainable": 52580}
+{"epoch": 0.453125, "grad_norm": 0.38324710726737976, "learning_rate": 9.262646494908604e-05, "loss": 0.004516632296144962, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00453, "step": 116, "tokens/total": 3506016, "tokens/train_per_sec_per_gpu": 34.74, "tokens/trainable": 53033}
+{"epoch": 0.45703125, "grad_norm": 0.3037130534648895, "learning_rate": 9.246645554956457e-05, "loss": 0.021973589435219765, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.02222, "step": 117, "tokens/total": 3536256, "tokens/train_per_sec_per_gpu": 35.22, "tokens/trainable": 53517}
+{"epoch": 0.4609375, "grad_norm": 1.3904820680618286, "learning_rate": 9.230488701789578e-05, "loss": 0.009210974909365177, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00925, "step": 118, "tokens/total": 3566480, "tokens/train_per_sec_per_gpu": 34.59, "tokens/trainable": 53967}
+{"epoch": 0.46484375, "grad_norm": 0.1571500301361084, "learning_rate": 9.214176607760577e-05, "loss": 0.0021451227366924286, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00215, "step": 119, "tokens/total": 3596624, "tokens/train_per_sec_per_gpu": 35.33, "tokens/trainable": 54430}
+{"epoch": 0.46875, "grad_norm": 0.39999091625213623, "learning_rate": 9.197709951682268e-05, "loss": 0.00788091216236353, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00791, "step": 120, "tokens/total": 3627168, "tokens/train_per_sec_per_gpu": 35.32, "tokens/trainable": 54896}
+{"epoch": 0.47265625, "grad_norm": 0.38124287128448486, "learning_rate": 9.181089418799428e-05, "loss": 0.010133227333426476, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01018, "step": 121, "tokens/total": 3657632, "tokens/train_per_sec_per_gpu": 37.76, "tokens/trainable": 55379}
+{"epoch": 0.4765625, "grad_norm": 0.0512852780520916, "learning_rate": 9.164315700760271e-05, "loss": 0.0007940311916172504, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00079, "step": 122, "tokens/total": 3687824, "tokens/train_per_sec_per_gpu": 31.71, "tokens/trainable": 55803}
+{"epoch": 0.48046875, "grad_norm": 0.13324694335460663, "learning_rate": 9.147389495587671e-05, "loss": 0.0023267122451215982, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00233, "step": 123, "tokens/total": 3718320, "tokens/train_per_sec_per_gpu": 36.46, "tokens/trainable": 56249}
+{"epoch": 0.484375, "grad_norm": 0.230186328291893, "learning_rate": 9.130311507650116e-05, "loss": 0.0037590472493320704, "memory/device_reserved (GiB)": 37.62, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00377, "step": 124, "tokens/total": 3748672, "tokens/train_per_sec_per_gpu": 32.43, "tokens/trainable": 56713}
+{"epoch": 0.48828125, "grad_norm": 0.30578872561454773, "learning_rate": 9.113082447632394e-05, "loss": 0.00473653944209218, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00475, "step": 125, "tokens/total": 3778976, "tokens/train_per_sec_per_gpu": 39.1, "tokens/trainable": 57196}
+{"epoch": 0.4921875, "grad_norm": 0.3714931607246399, "learning_rate": 9.09570303250602e-05, "loss": 0.005811735987663269, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00583, "step": 126, "tokens/total": 3809296, "tokens/train_per_sec_per_gpu": 36.55, "tokens/trainable": 57680}
+{"epoch": 0.49609375, "grad_norm": 0.11054892838001251, "learning_rate": 9.078173985499394e-05, "loss": 0.0032034481409937143, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00321, "step": 127, "tokens/total": 3839328, "tokens/train_per_sec_per_gpu": 31.88, "tokens/trainable": 58110}
+{"epoch": 0.5, "grad_norm": 0.09251131862401962, "learning_rate": 9.060496036067713e-05, "loss": 0.001724504167214036, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00173, "step": 128, "tokens/total": 3869824, "tokens/train_per_sec_per_gpu": 31.74, "tokens/trainable": 58534}
+{"epoch": 0.50390625, "grad_norm": 0.22758308053016663, "learning_rate": 9.042669919862615e-05, "loss": 0.004688034299761057, "memory/device_reserved (GiB)": 37.76, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0047, "step": 129, "tokens/total": 3900080, "tokens/train_per_sec_per_gpu": 35.19, "tokens/trainable": 58998}
+{"epoch": 0.5078125, "grad_norm": 0.2881723642349243, "learning_rate": 9.024696378701557e-05, "loss": 0.008266786113381386, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.0083, "step": 130, "tokens/total": 3930560, "tokens/train_per_sec_per_gpu": 31.83, "tokens/trainable": 59448}
+{"epoch": 0.51171875, "grad_norm": 0.18802326917648315, "learning_rate": 9.006576160536948e-05, "loss": 0.00283675454556942, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00284, "step": 131, "tokens/total": 3960496, "tokens/train_per_sec_per_gpu": 31.92, "tokens/trainable": 59906}
+{"epoch": 0.515625, "grad_norm": 0.6749681234359741, "learning_rate": 8.988310019425035e-05, "loss": 0.012044447474181652, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01212, "step": 132, "tokens/total": 3990928, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 60350}
+{"epoch": 0.51953125, "grad_norm": 0.5789079070091248, "learning_rate": 8.969898715494506e-05, "loss": 0.011312158778309822, "memory/device_reserved (GiB)": 37.16, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01138, "step": 133, "tokens/total": 4021264, "tokens/train_per_sec_per_gpu": 36.17, "tokens/trainable": 60831}
+{"epoch": 0.5234375, "grad_norm": 0.47060829401016235, "learning_rate": 8.951343014914869e-05, "loss": 0.0308814849704504, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.03136, "step": 134, "tokens/total": 4051728, "tokens/train_per_sec_per_gpu": 33.96, "tokens/trainable": 61305}
+{"epoch": 0.52734375, "grad_norm": 0.16633495688438416, "learning_rate": 8.932643689864568e-05, "loss": 0.005535767879337072, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00555, "step": 135, "tokens/total": 4081920, "tokens/train_per_sec_per_gpu": 37.62, "tokens/trainable": 61792}
+{"epoch": 0.53125, "grad_norm": 0.10699360817670822, "learning_rate": 8.913801518498845e-05, "loss": 0.002973862923681736, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00298, "step": 136, "tokens/total": 4112304, "tokens/train_per_sec_per_gpu": 31.37, "tokens/trainable": 62253}
+{"epoch": 0.53515625, "grad_norm": 0.2920030951499939, "learning_rate": 8.894817284917364e-05, "loss": 0.01169741339981556, "memory/device_reserved (GiB)": 38.55, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.01177, "step": 137, "tokens/total": 4142512, "tokens/train_per_sec_per_gpu": 31.65, "tokens/trainable": 62707}
+{"epoch": 0.5390625, "grad_norm": 0.3187088370323181, "learning_rate": 8.875691779131569e-05, "loss": 0.011357840150594711, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01142, "step": 138, "tokens/total": 4172992, "tokens/train_per_sec_per_gpu": 29.32, "tokens/trainable": 63145}
+{"epoch": 0.54296875, "grad_norm": 0.18588471412658691, "learning_rate": 8.856425797031829e-05, "loss": 0.006680965423583984, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.0067, "step": 139, "tokens/total": 4203136, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 63587}
+{"epoch": 0.546875, "grad_norm": 0.2760343551635742, "learning_rate": 8.837020140354295e-05, "loss": 0.01477967668324709, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01489, "step": 140, "tokens/total": 4233456, "tokens/train_per_sec_per_gpu": 36.74, "tokens/trainable": 64052}
+{"epoch": 0.55078125, "grad_norm": 0.49880966544151306, "learning_rate": 8.817475616647554e-05, "loss": 0.016770213842391968, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.01691, "step": 141, "tokens/total": 4263728, "tokens/train_per_sec_per_gpu": 34.38, "tokens/trainable": 64526}
+{"epoch": 0.5546875, "grad_norm": 0.181757390499115, "learning_rate": 8.797793039239017e-05, "loss": 0.006471520289778709, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.00649, "step": 142, "tokens/total": 4294432, "tokens/train_per_sec_per_gpu": 34.58, "tokens/trainable": 64983}
+{"epoch": 0.55859375, "grad_norm": 0.209610253572464, "learning_rate": 8.777973227201069e-05, "loss": 0.006492790300399065, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00651, "step": 143, "tokens/total": 4324960, "tokens/train_per_sec_per_gpu": 37.58, "tokens/trainable": 65492}
+{"epoch": 0.5625, "grad_norm": 0.13360266387462616, "learning_rate": 8.758017005316988e-05, "loss": 0.003702069167047739, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00371, "step": 144, "tokens/total": 4355424, "tokens/train_per_sec_per_gpu": 35.72, "tokens/trainable": 65925}
+{"epoch": 0.56640625, "grad_norm": 0.1640344262123108, "learning_rate": 8.737925204046629e-05, "loss": 0.006490131840109825, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00651, "step": 145, "tokens/total": 4385712, "tokens/train_per_sec_per_gpu": 31.39, "tokens/trainable": 66383}
+{"epoch": 0.5703125, "grad_norm": 0.13646955788135529, "learning_rate": 8.717698659491851e-05, "loss": 0.0026589329354465008, "memory/device_reserved (GiB)": 38.56, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00266, "step": 146, "tokens/total": 4416016, "tokens/train_per_sec_per_gpu": 31.39, "tokens/trainable": 66840}
+{"epoch": 0.57421875, "grad_norm": 0.4220513701438904, "learning_rate": 8.697338213361735e-05, "loss": 0.01334389764815569, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01343, "step": 147, "tokens/total": 4446368, "tokens/train_per_sec_per_gpu": 31.76, "tokens/trainable": 67297}
+{"epoch": 0.578125, "grad_norm": 0.37345919013023376, "learning_rate": 8.676844712937552e-05, "loss": 0.012794861570000648, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01288, "step": 148, "tokens/total": 4476880, "tokens/train_per_sec_per_gpu": 37.36, "tokens/trainable": 67769}
+{"epoch": 0.58203125, "grad_norm": 0.3093344271183014, "learning_rate": 8.656219011037509e-05, "loss": 0.012492594309151173, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01257, "step": 149, "tokens/total": 4507232, "tokens/train_per_sec_per_gpu": 32.83, "tokens/trainable": 68257}
+{"epoch": 0.5859375, "grad_norm": 0.2453778088092804, "learning_rate": 8.63546196598125e-05, "loss": 0.003456964623183012, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00346, "step": 150, "tokens/total": 4537376, "tokens/train_per_sec_per_gpu": 35.52, "tokens/trainable": 68706}
+{"epoch": 0.58984375, "grad_norm": 0.337802916765213, "learning_rate": 8.614574441554145e-05, "loss": 0.009631449356675148, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00968, "step": 151, "tokens/total": 4567584, "tokens/train_per_sec_per_gpu": 33.3, "tokens/trainable": 69131}
+{"epoch": 0.59375, "grad_norm": 0.34801673889160156, "learning_rate": 8.593557306971349e-05, "loss": 0.02037646435201168, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.02059, "step": 152, "tokens/total": 4597792, "tokens/train_per_sec_per_gpu": 31.93, "tokens/trainable": 69586}
+{"epoch": 0.59765625, "grad_norm": 0.03586283326148987, "learning_rate": 8.572411436841618e-05, "loss": 0.0008243054617196321, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00082, "step": 153, "tokens/total": 4627936, "tokens/train_per_sec_per_gpu": 32.79, "tokens/trainable": 70061}
+{"epoch": 0.6015625, "grad_norm": 0.1870104819536209, "learning_rate": 8.551137711130922e-05, "loss": 0.0038898580241948366, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0039, "step": 154, "tokens/total": 4658352, "tokens/train_per_sec_per_gpu": 27.95, "tokens/trainable": 70467}
+{"epoch": 0.60546875, "grad_norm": 0.2884272038936615, "learning_rate": 8.529737015125824e-05, "loss": 0.005026768893003464, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00504, "step": 155, "tokens/total": 4688896, "tokens/train_per_sec_per_gpu": 33.08, "tokens/trainable": 70933}
+{"epoch": 0.609375, "grad_norm": 0.07867873460054398, "learning_rate": 8.508210239396639e-05, "loss": 0.0024596985895186663, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00246, "step": 156, "tokens/total": 4719168, "tokens/train_per_sec_per_gpu": 34.29, "tokens/trainable": 71382}
+{"epoch": 0.61328125, "grad_norm": 0.056005269289016724, "learning_rate": 8.486558279760375e-05, "loss": 0.0012056033592671156, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00121, "step": 157, "tokens/total": 4749136, "tokens/train_per_sec_per_gpu": 30.83, "tokens/trainable": 71808}
+{"epoch": 0.6171875, "grad_norm": 0.34242892265319824, "learning_rate": 8.464782037243449e-05, "loss": 0.007983298972249031, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00802, "step": 158, "tokens/total": 4779472, "tokens/train_per_sec_per_gpu": 32.88, "tokens/trainable": 72249}
+{"epoch": 0.62109375, "grad_norm": 0.36051586270332336, "learning_rate": 8.442882418044202e-05, "loss": 0.011793753132224083, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.01186, "step": 159, "tokens/total": 4809632, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 72726}
+{"epoch": 0.625, "grad_norm": 0.376717746257782, "learning_rate": 8.420860333495179e-05, "loss": 0.018659302964806557, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.01883, "step": 160, "tokens/total": 4840112, "tokens/train_per_sec_per_gpu": 32.68, "tokens/trainable": 73148}
+{"epoch": 0.62890625, "grad_norm": 0.20132119953632355, "learning_rate": 8.398716700025208e-05, "loss": 0.007013507187366486, "memory/device_reserved (GiB)": 38.65, "memory/max_active (GiB)": 34.04, "memory/max_allocated (GiB)": 34.04, "ppl": 1.00704, "step": 161, "tokens/total": 4870656, "tokens/train_per_sec_per_gpu": 31.53, "tokens/trainable": 73600}
+{"epoch": 0.6328125, "grad_norm": 0.24618405103683472, "learning_rate": 8.376452439121266e-05, "loss": 0.00824644137173891, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00828, "step": 162, "tokens/total": 4900608, "tokens/train_per_sec_per_gpu": 35.76, "tokens/trainable": 74068}
+{"epoch": 0.63671875, "grad_norm": 0.2069157212972641, "learning_rate": 8.354068477290124e-05, "loss": 0.007023319602012634, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00705, "step": 163, "tokens/total": 4930752, "tokens/train_per_sec_per_gpu": 35.75, "tokens/trainable": 74527}
+{"epoch": 0.640625, "grad_norm": 0.1887698471546173, "learning_rate": 8.331565746019807e-05, "loss": 0.0082007497549057, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00823, "step": 164, "tokens/total": 4961008, "tokens/train_per_sec_per_gpu": 31.87, "tokens/trainable": 74992}
+{"epoch": 0.64453125, "grad_norm": 0.17459234595298767, "learning_rate": 8.308945181740812e-05, "loss": 0.004637294448912144, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00465, "step": 165, "tokens/total": 4991200, "tokens/train_per_sec_per_gpu": 35.29, "tokens/trainable": 75442}
+{"epoch": 0.6484375, "grad_norm": 0.26009130477905273, "learning_rate": 8.286207725787153e-05, "loss": 0.007355842739343643, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00738, "step": 166, "tokens/total": 5021600, "tokens/train_per_sec_per_gpu": 31.27, "tokens/trainable": 75887}
+{"epoch": 0.65234375, "grad_norm": 0.2539709806442261, "learning_rate": 8.263354324357182e-05, "loss": 0.010408539324998856, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.01046, "step": 167, "tokens/total": 5052032, "tokens/train_per_sec_per_gpu": 34.06, "tokens/trainable": 76331}
+{"epoch": 0.65625, "grad_norm": 0.12331778556108475, "learning_rate": 8.240385928474219e-05, "loss": 0.0022821722086519003, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00228, "step": 168, "tokens/total": 5080256, "tokens/train_per_sec_per_gpu": 35.92, "tokens/trainable": 76745}
+{"epoch": 0.66015625, "grad_norm": 0.110007144510746, "learning_rate": 8.217303493946967e-05, "loss": 0.0038710185326635838, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00388, "step": 169, "tokens/total": 5110784, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 77201}
+{"epoch": 0.6640625, "grad_norm": 0.03679708018898964, "learning_rate": 8.194107981329746e-05, "loss": 0.000850176380481571, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00085, "step": 170, "tokens/total": 5141200, "tokens/train_per_sec_per_gpu": 33.4, "tokens/trainable": 77655}
+{"epoch": 0.66796875, "grad_norm": 0.12713676691055298, "learning_rate": 8.170800355882518e-05, "loss": 0.0018071834929287434, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00181, "step": 171, "tokens/total": 5171760, "tokens/train_per_sec_per_gpu": 39.23, "tokens/trainable": 78122}
+{"epoch": 0.671875, "grad_norm": 0.1453125774860382, "learning_rate": 8.147381587530713e-05, "loss": 0.0017664346378296614, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00177, "step": 172, "tokens/total": 5202272, "tokens/train_per_sec_per_gpu": 33.53, "tokens/trainable": 78576}
+{"epoch": 0.67578125, "grad_norm": 0.21252205967903137, "learning_rate": 8.123852650824877e-05, "loss": 0.002328047761693597, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00233, "step": 173, "tokens/total": 5232800, "tokens/train_per_sec_per_gpu": 36.44, "tokens/trainable": 79059}
+{"epoch": 0.6796875, "grad_norm": 0.637407660484314, "learning_rate": 8.100214524900103e-05, "loss": 0.007709754630923271, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00774, "step": 174, "tokens/total": 5262672, "tokens/train_per_sec_per_gpu": 29.72, "tokens/trainable": 79498}
+{"epoch": 0.68359375, "grad_norm": 0.06158079952001572, "learning_rate": 8.076468193435301e-05, "loss": 0.0007811276591382921, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00078, "step": 175, "tokens/total": 5293072, "tokens/train_per_sec_per_gpu": 30.45, "tokens/trainable": 79922}
+{"epoch": 0.6875, "grad_norm": 0.04236136004328728, "learning_rate": 8.052614644612253e-05, "loss": 0.000772522296756506, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.46, "memory/max_allocated (GiB)": 33.46, "ppl": 1.00077, "step": 176, "tokens/total": 5321520, "tokens/train_per_sec_per_gpu": 35.6, "tokens/trainable": 80383}
+{"epoch": 0.69140625, "grad_norm": 0.0892096534371376, "learning_rate": 8.028654871074489e-05, "loss": 0.0023201322183012962, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00232, "step": 177, "tokens/total": 5351904, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 80824}
+{"epoch": 0.6953125, "grad_norm": 0.679317831993103, "learning_rate": 8.004589869885986e-05, "loss": 0.008408154360949993, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00844, "step": 178, "tokens/total": 5382432, "tokens/train_per_sec_per_gpu": 32.23, "tokens/trainable": 81243}
+{"epoch": 0.69921875, "grad_norm": 0.05571528524160385, "learning_rate": 7.980420642489674e-05, "loss": 0.00045867619337514043, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00046, "step": 179, "tokens/total": 5412960, "tokens/train_per_sec_per_gpu": 34.28, "tokens/trainable": 81716}
+{"epoch": 0.703125, "grad_norm": 0.36156049370765686, "learning_rate": 7.95614819466576e-05, "loss": 0.0047795758582651615, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00479, "step": 180, "tokens/total": 5443232, "tokens/train_per_sec_per_gpu": 35.55, "tokens/trainable": 82181}
+{"epoch": 0.70703125, "grad_norm": 0.3550747036933899, "learning_rate": 7.931773536489872e-05, "loss": 0.0044985488057136536, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.29, "memory/max_allocated (GiB)": 33.29, "ppl": 1.00451, "step": 181, "tokens/total": 5471440, "tokens/train_per_sec_per_gpu": 34.65, "tokens/trainable": 82626}
+{"epoch": 0.7109375, "grad_norm": 0.00732263270765543, "learning_rate": 7.907297682291035e-05, "loss": 0.00012463401071727276, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00012, "step": 182, "tokens/total": 5501744, "tokens/train_per_sec_per_gpu": 35.19, "tokens/trainable": 83089}
+{"epoch": 0.71484375, "grad_norm": 0.005601493176072836, "learning_rate": 7.882721650609442e-05, "loss": 0.00012698184582404792, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00013, "step": 183, "tokens/total": 5532272, "tokens/train_per_sec_per_gpu": 35.39, "tokens/trainable": 83590}
+{"epoch": 0.71875, "grad_norm": 0.03298855572938919, "learning_rate": 7.85804646415409e-05, "loss": 0.00026586768217384815, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00027, "step": 184, "tokens/total": 5562784, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 84046}
+{"epoch": 0.72265625, "grad_norm": 0.02470102533698082, "learning_rate": 7.833273149760207e-05, "loss": 0.00024917692644521594, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00025, "step": 185, "tokens/total": 5592800, "tokens/train_per_sec_per_gpu": 34.05, "tokens/trainable": 84517}
+{"epoch": 0.7265625, "grad_norm": 0.6944283246994019, "learning_rate": 7.808402738346527e-05, "loss": 0.016732344403862953, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01687, "step": 186, "tokens/total": 5623088, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 84974}
+{"epoch": 0.73046875, "grad_norm": 0.011723512783646584, "learning_rate": 7.783436264872382e-05, "loss": 0.00017266182112507522, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00017, "step": 187, "tokens/total": 5653344, "tokens/train_per_sec_per_gpu": 35.07, "tokens/trainable": 85460}
+{"epoch": 0.734375, "grad_norm": 0.34870269894599915, "learning_rate": 7.758374768294647e-05, "loss": 0.004548810888081789, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00456, "step": 188, "tokens/total": 5683600, "tokens/train_per_sec_per_gpu": 36.09, "tokens/trainable": 85939}
+{"epoch": 0.73828125, "grad_norm": 0.09110172092914581, "learning_rate": 7.733219291524489e-05, "loss": 0.001469930517487228, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00147, "step": 189, "tokens/total": 5711952, "tokens/train_per_sec_per_gpu": 38.43, "tokens/trainable": 86377}
+{"epoch": 0.7421875, "grad_norm": 0.6382125020027161, "learning_rate": 7.707970881383977e-05, "loss": 0.01117285992950201, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01124, "step": 190, "tokens/total": 5742336, "tokens/train_per_sec_per_gpu": 33.87, "tokens/trainable": 86834}
+{"epoch": 0.74609375, "grad_norm": 0.1396624743938446, "learning_rate": 7.682630588562518e-05, "loss": 0.0022487500682473183, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00225, "step": 191, "tokens/total": 5772560, "tokens/train_per_sec_per_gpu": 32.06, "tokens/trainable": 87265}
+{"epoch": 0.75, "grad_norm": 0.09118734300136566, "learning_rate": 7.657199467573129e-05, "loss": 0.0009341652039438486, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00093, "step": 192, "tokens/total": 5803136, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 87747}
+{"epoch": 0.75390625, "grad_norm": 0.10082298517227173, "learning_rate": 7.631678576708561e-05, "loss": 0.001603117911145091, "memory/device_reserved (GiB)": 38.58, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0016, "step": 193, "tokens/total": 5833440, "tokens/train_per_sec_per_gpu": 36.21, "tokens/trainable": 88239}
+{"epoch": 0.7578125, "grad_norm": 0.0594109408557415, "learning_rate": 7.606068977997255e-05, "loss": 0.0009742019465193152, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00097, "step": 194, "tokens/total": 5863552, "tokens/train_per_sec_per_gpu": 36.22, "tokens/trainable": 88718}
+{"epoch": 0.76171875, "grad_norm": 0.12520930171012878, "learning_rate": 7.580371737159148e-05, "loss": 0.0015380105469375849, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00154, "step": 195, "tokens/total": 5893680, "tokens/train_per_sec_per_gpu": 31.29, "tokens/trainable": 89129}
+{"epoch": 0.765625, "grad_norm": 0.32357996702194214, "learning_rate": 7.554587923561324e-05, "loss": 0.0033828348387032747, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00339, "step": 196, "tokens/total": 5923744, "tokens/train_per_sec_per_gpu": 32.29, "tokens/trainable": 89567}
+{"epoch": 0.76953125, "grad_norm": 0.30336976051330566, "learning_rate": 7.528718610173511e-05, "loss": 0.009017270989716053, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00906, "step": 197, "tokens/total": 5954128, "tokens/train_per_sec_per_gpu": 30.38, "tokens/trainable": 89988}
+{"epoch": 0.7734375, "grad_norm": 0.530124306678772, "learning_rate": 7.502764873523431e-05, "loss": 0.013890329748392105, "memory/device_reserved (GiB)": 35.56, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.01399, "step": 198, "tokens/total": 5984352, "tokens/train_per_sec_per_gpu": 36.15, "tokens/trainable": 90434}
+{"epoch": 0.77734375, "grad_norm": 0.024880079552531242, "learning_rate": 7.476727793652011e-05, "loss": 0.00029932294273748994, "memory/device_reserved (GiB)": 35.6, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0003, "step": 199, "tokens/total": 6014704, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 90913}
+{"epoch": 0.78125, "grad_norm": 0.07654840499162674, "learning_rate": 7.450608454068415e-05, "loss": 0.0013769213110208511, "memory/device_reserved (GiB)": 35.66, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00138, "step": 200, "tokens/total": 6045056, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 91355}
+{"epoch": 0.78515625, "grad_norm": 0.08619414269924164, "learning_rate": 7.424407941704987e-05, "loss": 0.0012924536131322384, "memory/device_reserved (GiB)": 35.66, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00129, "step": 201, "tokens/total": 6075504, "tokens/train_per_sec_per_gpu": 37.38, "tokens/trainable": 91856}
+{"epoch": 0.7890625, "grad_norm": 0.0749412402510643, "learning_rate": 7.398127346871986e-05, "loss": 0.0007854659343138337, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00079, "step": 202, "tokens/total": 6105904, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 92311}
+{"epoch": 0.79296875, "grad_norm": 0.12518921494483948, "learning_rate": 7.371767763212238e-05, "loss": 0.0026314458809792995, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00263, "step": 203, "tokens/total": 6136272, "tokens/train_per_sec_per_gpu": 34.73, "tokens/trainable": 92764}
+{"epoch": 0.796875, "grad_norm": 0.14211589097976685, "learning_rate": 7.345330287655617e-05, "loss": 0.00402654055505991, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.65, "memory/max_allocated (GiB)": 33.65, "ppl": 1.00403, "step": 204, "tokens/total": 6166336, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 93227}
+{"epoch": 0.80078125, "grad_norm": 0.9564501047134399, "learning_rate": 7.31881602037339e-05, "loss": 0.01280949730426073, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.01289, "step": 205, "tokens/total": 6196720, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 93675}
+{"epoch": 0.8046875, "grad_norm": 0.13096395134925842, "learning_rate": 7.29222606473245e-05, "loss": 0.0037373730447143316, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00374, "step": 206, "tokens/total": 6227424, "tokens/train_per_sec_per_gpu": 32.3, "tokens/trainable": 94120}
+{"epoch": 0.80859375, "grad_norm": 0.08379670232534409, "learning_rate": 7.265561527249383e-05, "loss": 0.0017476077191531658, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00175, "step": 207, "tokens/total": 6257616, "tokens/train_per_sec_per_gpu": 35.39, "tokens/trainable": 94557}
+{"epoch": 0.8125, "grad_norm": 0.05349349230527878, "learning_rate": 7.238823517544436e-05, "loss": 0.0018553336849436164, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00186, "step": 208, "tokens/total": 6288000, "tokens/train_per_sec_per_gpu": 40.66, "tokens/trainable": 95035}
+{"epoch": 0.81640625, "grad_norm": 0.11009859293699265, "learning_rate": 7.212013148295333e-05, "loss": 0.0024754812475293875, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00248, "step": 209, "tokens/total": 6318336, "tokens/train_per_sec_per_gpu": 37.59, "tokens/trainable": 95517}
+{"epoch": 0.8203125, "grad_norm": 0.08477463573217392, "learning_rate": 7.185131535190975e-05, "loss": 0.0019841620232909918, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00199, "step": 210, "tokens/total": 6348656, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 96026}
+{"epoch": 0.82421875, "grad_norm": 0.06025635451078415, "learning_rate": 7.158179796885005e-05, "loss": 0.0011887844884768128, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00119, "step": 211, "tokens/total": 6379040, "tokens/train_per_sec_per_gpu": 35.64, "tokens/trainable": 96477}
+{"epoch": 0.828125, "grad_norm": 0.07387597858905792, "learning_rate": 7.131159054949273e-05, "loss": 0.001786265056580305, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00179, "step": 212, "tokens/total": 6409312, "tokens/train_per_sec_per_gpu": 33.17, "tokens/trainable": 96951}
+{"epoch": 0.83203125, "grad_norm": 0.1013425812125206, "learning_rate": 7.104070433827139e-05, "loss": 0.0019797745626419783, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00198, "step": 213, "tokens/total": 6439520, "tokens/train_per_sec_per_gpu": 30.81, "tokens/trainable": 97350}
+{"epoch": 0.8359375, "grad_norm": 0.009712542407214642, "learning_rate": 7.076915060786705e-05, "loss": 0.00013215326180215925, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00013, "step": 214, "tokens/total": 6469888, "tokens/train_per_sec_per_gpu": 29.91, "tokens/trainable": 97792}
+{"epoch": 0.83984375, "grad_norm": 0.14567089080810547, "learning_rate": 7.049694065873882e-05, "loss": 0.0015704174293205142, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00157, "step": 215, "tokens/total": 6500128, "tokens/train_per_sec_per_gpu": 36.44, "tokens/trainable": 98257}
+{"epoch": 0.84375, "grad_norm": 0.021219903603196144, "learning_rate": 7.022408581865382e-05, "loss": 0.0003704531991388649, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00037, "step": 216, "tokens/total": 6530832, "tokens/train_per_sec_per_gpu": 32.74, "tokens/trainable": 98731}
+{"epoch": 0.84765625, "grad_norm": 0.22449812293052673, "learning_rate": 6.99505974422157e-05, "loss": 0.0037617988418787718, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00377, "step": 217, "tokens/total": 6561248, "tokens/train_per_sec_per_gpu": 36.5, "tokens/trainable": 99212}
+{"epoch": 0.8515625, "grad_norm": 0.6340874433517456, "learning_rate": 6.967648691039213e-05, "loss": 0.0011263209162279963, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00113, "step": 218, "tokens/total": 6591648, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 99654}
+{"epoch": 0.85546875, "grad_norm": 0.1624881774187088, "learning_rate": 6.940176563004123e-05, "loss": 0.0014779233606532216, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00148, "step": 219, "tokens/total": 6622368, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 100086}
+{"epoch": 0.859375, "grad_norm": 0.01793455332517624, "learning_rate": 6.912644503343682e-05, "loss": 0.0001897630572784692, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00019, "step": 220, "tokens/total": 6652848, "tokens/train_per_sec_per_gpu": 33.34, "tokens/trainable": 100524}
+{"epoch": 0.86328125, "grad_norm": 0.12492946535348892, "learning_rate": 6.885053657779273e-05, "loss": 0.000895547098480165, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0009, "step": 221, "tokens/total": 6683392, "tokens/train_per_sec_per_gpu": 33.46, "tokens/trainable": 100996}
+{"epoch": 0.8671875, "grad_norm": 0.015087602660059929, "learning_rate": 6.857405174478604e-05, "loss": 0.0001049312122631818, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0001, "step": 222, "tokens/total": 6713712, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 101449}
+{"epoch": 0.87109375, "grad_norm": 0.713071882724762, "learning_rate": 6.82970020400792e-05, "loss": 0.01887362264096737, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01905, "step": 223, "tokens/total": 6744176, "tokens/train_per_sec_per_gpu": 33.14, "tokens/trainable": 101905}
+{"epoch": 0.875, "grad_norm": 0.008568123914301395, "learning_rate": 6.801939899284132e-05, "loss": 7.857779564801604e-05, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00008, "step": 224, "tokens/total": 6774416, "tokens/train_per_sec_per_gpu": 36.43, "tokens/trainable": 102411}
+{"epoch": 0.87890625, "grad_norm": 0.6806920766830444, "learning_rate": 6.774125415526827e-05, "loss": 0.010111674666404724, "memory/device_reserved (GiB)": 37.19, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.01016, "step": 225, "tokens/total": 6804592, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 102882}
+{"epoch": 0.8828125, "grad_norm": 0.00420374283567071, "learning_rate": 6.746257910210214e-05, "loss": 5.371138468035497e-05, "memory/device_reserved (GiB)": 35.49, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00005, "step": 226, "tokens/total": 6834976, "tokens/train_per_sec_per_gpu": 34.69, "tokens/trainable": 103332}
+{"epoch": 0.88671875, "grad_norm": 0.02610113099217415, "learning_rate": 6.718338543014937e-05, "loss": 0.00038069591391831636, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00038, "step": 227, "tokens/total": 6865408, "tokens/train_per_sec_per_gpu": 35.18, "tokens/trainable": 103790}
+{"epoch": 0.890625, "grad_norm": 0.0028072672430425882, "learning_rate": 6.69036847577983e-05, "loss": 6.497368303826079e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00006, "step": 228, "tokens/total": 6895664, "tokens/train_per_sec_per_gpu": 34.28, "tokens/trainable": 104246}
+{"epoch": 0.89453125, "grad_norm": 0.8277482390403748, "learning_rate": 6.662348872453553e-05, "loss": 0.013669933192431927, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01376, "step": 229, "tokens/total": 6926016, "tokens/train_per_sec_per_gpu": 37.72, "tokens/trainable": 104707}
+{"epoch": 0.8984375, "grad_norm": 0.08310598134994507, "learning_rate": 6.63428089904618e-05, "loss": 0.0005468585877679288, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00055, "step": 230, "tokens/total": 6956384, "tokens/train_per_sec_per_gpu": 36.84, "tokens/trainable": 105167}
+{"epoch": 0.90234375, "grad_norm": 0.008127766661345959, "learning_rate": 6.60616572358065e-05, "loss": 0.00013037689495831728, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00013, "step": 231, "tokens/total": 6986768, "tokens/train_per_sec_per_gpu": 33.18, "tokens/trainable": 105576}
+{"epoch": 0.90625, "grad_norm": 0.03313232958316803, "learning_rate": 6.578004516044172e-05, "loss": 0.000351642636815086, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00035, "step": 232, "tokens/total": 7017232, "tokens/train_per_sec_per_gpu": 36.76, "tokens/trainable": 106043}
+{"epoch": 0.91015625, "grad_norm": 0.08690419793128967, "learning_rate": 6.549798448339548e-05, "loss": 0.001037480542436242, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00104, "step": 233, "tokens/total": 7047568, "tokens/train_per_sec_per_gpu": 35.19, "tokens/trainable": 106506}
+{"epoch": 0.9140625, "grad_norm": 0.12895406782627106, "learning_rate": 6.521548694236384e-05, "loss": 0.0019432969857007265, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00195, "step": 234, "tokens/total": 7077760, "tokens/train_per_sec_per_gpu": 34.15, "tokens/trainable": 106951}
+{"epoch": 0.91796875, "grad_norm": 0.25051259994506836, "learning_rate": 6.493256429322259e-05, "loss": 0.0025090454146265984, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00251, "step": 235, "tokens/total": 7107760, "tokens/train_per_sec_per_gpu": 30.73, "tokens/trainable": 107382}
+{"epoch": 0.921875, "grad_norm": 0.5061792731285095, "learning_rate": 6.464922830953799e-05, "loss": 0.00891919620335102, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00896, "step": 236, "tokens/total": 7138144, "tokens/train_per_sec_per_gpu": 32.82, "tokens/trainable": 107814}
+{"epoch": 0.92578125, "grad_norm": 0.3209003210067749, "learning_rate": 6.436549078207688e-05, "loss": 0.0041964175179600716, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00421, "step": 237, "tokens/total": 7168352, "tokens/train_per_sec_per_gpu": 35.25, "tokens/trainable": 108274}
+{"epoch": 0.9296875, "grad_norm": 0.3212501108646393, "learning_rate": 6.408136351831592e-05, "loss": 0.0037440985906869173, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00375, "step": 238, "tokens/total": 7198624, "tokens/train_per_sec_per_gpu": 29.09, "tokens/trainable": 108714}
+{"epoch": 0.93359375, "grad_norm": 0.021965481340885162, "learning_rate": 6.379685834195036e-05, "loss": 0.00035154391662217677, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00035, "step": 239, "tokens/total": 7229216, "tokens/train_per_sec_per_gpu": 36.7, "tokens/trainable": 109220}
+{"epoch": 0.9375, "grad_norm": 0.1164102703332901, "learning_rate": 6.351198709240186e-05, "loss": 0.0012684958055615425, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00127, "step": 240, "tokens/total": 7259648, "tokens/train_per_sec_per_gpu": 32.0, "tokens/trainable": 109672}
+{"epoch": 0.94140625, "grad_norm": 0.17972798645496368, "learning_rate": 6.32267616243259e-05, "loss": 0.0024644152726978064, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00247, "step": 241, "tokens/total": 7289824, "tokens/train_per_sec_per_gpu": 35.09, "tokens/trainable": 110135}
+{"epoch": 0.9453125, "grad_norm": 0.5711016654968262, "learning_rate": 6.294119380711849e-05, "loss": 0.01326853595674038, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01336, "step": 242, "tokens/total": 7320288, "tokens/train_per_sec_per_gpu": 30.6, "tokens/trainable": 110563}
+{"epoch": 0.94921875, "grad_norm": 0.7820162177085876, "learning_rate": 6.265529552442209e-05, "loss": 0.01847490295767784, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01865, "step": 243, "tokens/total": 7350736, "tokens/train_per_sec_per_gpu": 36.84, "tokens/trainable": 111049}
+{"epoch": 0.953125, "grad_norm": 0.2531258165836334, "learning_rate": 6.236907867363127e-05, "loss": 0.003868672763928771, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00388, "step": 244, "tokens/total": 7381280, "tokens/train_per_sec_per_gpu": 33.31, "tokens/trainable": 111495}
+{"epoch": 0.95703125, "grad_norm": 0.09388009458780289, "learning_rate": 6.208255516539749e-05, "loss": 0.0010953794699162245, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0011, "step": 245, "tokens/total": 7411632, "tokens/train_per_sec_per_gpu": 31.68, "tokens/trainable": 111968}
+{"epoch": 0.9609375, "grad_norm": 0.06202390044927597, "learning_rate": 6.179573692313344e-05, "loss": 0.0006574424332939088, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00066, "step": 246, "tokens/total": 7441904, "tokens/train_per_sec_per_gpu": 36.44, "tokens/trainable": 112416}
+{"epoch": 0.96484375, "grad_norm": 0.21640881896018982, "learning_rate": 6.150863588251694e-05, "loss": 0.00276574632152915, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00277, "step": 247, "tokens/total": 7472368, "tokens/train_per_sec_per_gpu": 34.05, "tokens/trainable": 112882}
+{"epoch": 0.96875, "grad_norm": 0.04909277334809303, "learning_rate": 6.122126399099419e-05, "loss": 0.0012688931310549378, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00127, "step": 248, "tokens/total": 7502656, "tokens/train_per_sec_per_gpu": 40.02, "tokens/trainable": 113390}
+{"epoch": 0.97265625, "grad_norm": 0.2183118760585785, "learning_rate": 6.0933633207282615e-05, "loss": 0.01150418072938919, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01157, "step": 249, "tokens/total": 7532800, "tokens/train_per_sec_per_gpu": 37.76, "tokens/trainable": 113904}
+{"epoch": 0.9765625, "grad_norm": 0.05114463344216347, "learning_rate": 6.064575550087316e-05, "loss": 0.0009117473382502794, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00091, "step": 250, "tokens/total": 7563264, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 114351}
+{"epoch": 0.98046875, "grad_norm": 0.10292479395866394, "learning_rate": 6.0357642851532245e-05, "loss": 0.0016239782562479377, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00163, "step": 251, "tokens/total": 7593840, "tokens/train_per_sec_per_gpu": 35.24, "tokens/trainable": 114842}
+{"epoch": 0.984375, "grad_norm": 0.057799480855464935, "learning_rate": 6.0069307248803294e-05, "loss": 0.0011053154012188315, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00111, "step": 252, "tokens/total": 7624416, "tokens/train_per_sec_per_gpu": 28.88, "tokens/trainable": 115263}
+{"epoch": 0.98828125, "grad_norm": 0.017502324655652046, "learning_rate": 5.9780760691507635e-05, "loss": 0.0003236275806557387, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00032, "step": 253, "tokens/total": 7654688, "tokens/train_per_sec_per_gpu": 32.66, "tokens/trainable": 115703}
+{"epoch": 0.9921875, "grad_norm": 0.2686062753200531, "learning_rate": 5.9492015187245334e-05, "loss": 0.003511242102831602, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00352, "step": 254, "tokens/total": 7685088, "tokens/train_per_sec_per_gpu": 33.26, "tokens/trainable": 116157}
+{"epoch": 0.99609375, "grad_norm": 0.25028568506240845, "learning_rate": 5.920308275189541e-05, "loss": 0.0028144530951976776, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00282, "step": 255, "tokens/total": 7715552, "tokens/train_per_sec_per_gpu": 31.95, "tokens/trainable": 116567}
+{"epoch": 1.0, "grad_norm": 0.010626083239912987, "learning_rate": 5.8913975409115874e-05, "loss": 0.0002322449436178431, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00023, "step": 256, "tokens/total": 7745872, "tokens/train_per_sec_per_gpu": 31.11, "tokens/trainable": 117030}
+{"epoch": 1.00390625, "grad_norm": 0.004322522785514593, "learning_rate": 5.8624705189843395e-05, "loss": 8.972767682280391e-05, "memory/device_reserved (GiB)": 38.23, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00009, "step": 257, "tokens/total": 7776208, "tokens/train_per_sec_per_gpu": 34.73, "tokens/trainable": 117517}
+{"epoch": 1.0078125, "grad_norm": 0.04855626821517944, "learning_rate": 5.833528413179249e-05, "loss": 0.0010425080545246601, "memory/device_reserved (GiB)": 35.62, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00104, "step": 258, "tokens/total": 7806480, "tokens/train_per_sec_per_gpu": 38.33, "tokens/trainable": 118015}
+{"epoch": 1.01171875, "grad_norm": 0.015271801501512527, "learning_rate": 5.80457242789548e-05, "loss": 0.0002960565616376698, "memory/device_reserved (GiB)": 35.62, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.0003, "step": 259, "tokens/total": 7836800, "tokens/train_per_sec_per_gpu": 36.2, "tokens/trainable": 118520}
+{"epoch": 1.015625, "grad_norm": 0.1238817349076271, "learning_rate": 5.77560376810977e-05, "loss": 0.001452557509765029, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00145, "step": 260, "tokens/total": 7867040, "tokens/train_per_sec_per_gpu": 36.15, "tokens/trainable": 118992}
+{"epoch": 1.01953125, "grad_norm": 0.008926448412239552, "learning_rate": 5.7466236393263005e-05, "loss": 0.0002352800511289388, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00024, "step": 261, "tokens/total": 7897408, "tokens/train_per_sec_per_gpu": 33.65, "tokens/trainable": 119438}
+{"epoch": 1.0234375, "grad_norm": 0.0867115706205368, "learning_rate": 5.717633247526522e-05, "loss": 0.0014156652614474297, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00142, "step": 262, "tokens/total": 7927648, "tokens/train_per_sec_per_gpu": 32.1, "tokens/trainable": 119881}
+{"epoch": 1.02734375, "grad_norm": 0.03644087538123131, "learning_rate": 5.688633799118971e-05, "loss": 0.0004944024258293211, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00049, "step": 263, "tokens/total": 7957968, "tokens/train_per_sec_per_gpu": 27.72, "tokens/trainable": 120285}
+{"epoch": 1.03125, "grad_norm": 0.4136442542076111, "learning_rate": 5.659626500889066e-05, "loss": 0.005234354641288519, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.65, "memory/max_allocated (GiB)": 33.65, "ppl": 1.00525, "step": 264, "tokens/total": 7987888, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 120755}
+{"epoch": 1.03515625, "grad_norm": 0.011108173988759518, "learning_rate": 5.6306125599488905e-05, "loss": 0.00019686836458276957, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0002, "step": 265, "tokens/total": 8018048, "tokens/train_per_sec_per_gpu": 35.55, "tokens/trainable": 121194}
+{"epoch": 1.0390625, "grad_norm": 0.2199329286813736, "learning_rate": 5.601593183686955e-05, "loss": 0.005357124377042055, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00537, "step": 266, "tokens/total": 8048448, "tokens/train_per_sec_per_gpu": 36.54, "tokens/trainable": 121670}
+{"epoch": 1.04296875, "grad_norm": 0.15096357464790344, "learning_rate": 5.572569579717961e-05, "loss": 0.0024120814632624388, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00241, "step": 267, "tokens/total": 8078848, "tokens/train_per_sec_per_gpu": 33.58, "tokens/trainable": 122142}
+{"epoch": 1.046875, "grad_norm": 0.0024968513753265142, "learning_rate": 5.543542955832538e-05, "loss": 4.619035462383181e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00005, "step": 268, "tokens/total": 8109360, "tokens/train_per_sec_per_gpu": 32.36, "tokens/trainable": 122585}
+{"epoch": 1.05078125, "grad_norm": 0.018697405233979225, "learning_rate": 5.514514519946986e-05, "loss": 0.00024445558665320277, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00024, "step": 269, "tokens/total": 8139888, "tokens/train_per_sec_per_gpu": 38.78, "tokens/trainable": 123091}
+{"epoch": 1.0546875, "grad_norm": 0.04383962228894234, "learning_rate": 5.485485480053015e-05, "loss": 0.0005069951876066625, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00051, "step": 270, "tokens/total": 8170480, "tokens/train_per_sec_per_gpu": 30.91, "tokens/trainable": 123505}
+{"epoch": 1.05859375, "grad_norm": 0.018031178042292595, "learning_rate": 5.4564570441674645e-05, "loss": 0.00028141128132119775, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.41, "memory/max_allocated (GiB)": 33.41, "ppl": 1.00028, "step": 271, "tokens/total": 8198848, "tokens/train_per_sec_per_gpu": 35.31, "tokens/trainable": 123953}
+{"epoch": 1.0625, "grad_norm": 0.016046874225139618, "learning_rate": 5.42743042028204e-05, "loss": 0.00015048845671117306, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00015, "step": 272, "tokens/total": 8229200, "tokens/train_per_sec_per_gpu": 28.61, "tokens/trainable": 124400}
+{"epoch": 1.06640625, "grad_norm": 0.004127623979002237, "learning_rate": 5.3984068163130464e-05, "loss": 7.019042095635086e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00007, "step": 273, "tokens/total": 8259536, "tokens/train_per_sec_per_gpu": 35.1, "tokens/trainable": 124870}
+{"epoch": 1.0703125, "grad_norm": 0.016368450596928596, "learning_rate": 5.369387440051111e-05, "loss": 0.00023265022900886834, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00023, "step": 274, "tokens/total": 8289904, "tokens/train_per_sec_per_gpu": 32.05, "tokens/trainable": 125333}
+{"epoch": 1.07421875, "grad_norm": 0.0009975603315979242, "learning_rate": 5.340373499110935e-05, "loss": 2.3090822651283816e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00002, "step": 275, "tokens/total": 8320176, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 125834}
+{"epoch": 1.078125, "grad_norm": 0.0020015796180814505, "learning_rate": 5.3113662008810304e-05, "loss": 2.2906289814272895e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00002, "step": 276, "tokens/total": 8350832, "tokens/train_per_sec_per_gpu": 38.02, "tokens/trainable": 126316}
+{"epoch": 1.08203125, "grad_norm": 0.005128095392137766, "learning_rate": 5.282366752473479e-05, "loss": 6.587664393009618e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00007, "step": 277, "tokens/total": 8380976, "tokens/train_per_sec_per_gpu": 35.33, "tokens/trainable": 126798}
+{"epoch": 1.0859375, "grad_norm": 0.0013011472765356302, "learning_rate": 5.2533763606737005e-05, "loss": 2.66208026005188e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00003, "step": 278, "tokens/total": 8411072, "tokens/train_per_sec_per_gpu": 29.64, "tokens/trainable": 127223}
+{"epoch": 1.08984375, "grad_norm": 0.001754116965457797, "learning_rate": 5.224396231890232e-05, "loss": 2.587787457741797e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00003, "step": 279, "tokens/total": 8440736, "tokens/train_per_sec_per_gpu": 36.28, "tokens/trainable": 127672}
+{"epoch": 1.09375, "grad_norm": 0.005082705058157444, "learning_rate": 5.195427572104522e-05, "loss": 8.052532211877406e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00008, "step": 280, "tokens/total": 8470944, "tokens/train_per_sec_per_gpu": 32.1, "tokens/trainable": 128116}
+{"epoch": 1.09765625, "grad_norm": 0.0014385804533958435, "learning_rate": 5.166471586820751e-05, "loss": 2.603003304102458e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00003, "step": 281, "tokens/total": 8501104, "tokens/train_per_sec_per_gpu": 32.82, "tokens/trainable": 128589}
+{"epoch": 1.1015625, "grad_norm": 0.7784804105758667, "learning_rate": 5.1375294810156615e-05, "loss": 0.008352375589311123, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00839, "step": 282, "tokens/total": 8531696, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 129036}
+{"epoch": 1.10546875, "grad_norm": 0.0021191469859331846, "learning_rate": 5.1086024590884144e-05, "loss": 3.5222510632593185e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00004, "step": 283, "tokens/total": 8561936, "tokens/train_per_sec_per_gpu": 35.21, "tokens/trainable": 129485}
+{"epoch": 1.109375, "grad_norm": 0.019356347620487213, "learning_rate": 5.079691724810461e-05, "loss": 0.0002056795492535457, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00021, "step": 284, "tokens/total": 8592192, "tokens/train_per_sec_per_gpu": 29.55, "tokens/trainable": 129920}
+{"epoch": 1.11328125, "grad_norm": 0.030793415382504463, "learning_rate": 5.0507984812754684e-05, "loss": 0.00022263142454903573, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00022, "step": 285, "tokens/total": 8622432, "tokens/train_per_sec_per_gpu": 36.14, "tokens/trainable": 130417}
+{"epoch": 1.1171875, "grad_norm": 0.2944176197052002, "learning_rate": 5.021923930849237e-05, "loss": 0.0028715431690216064, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00288, "step": 286, "tokens/total": 8652768, "tokens/train_per_sec_per_gpu": 35.95, "tokens/trainable": 130903}
+{"epoch": 1.12109375, "grad_norm": 0.0015030609210953116, "learning_rate": 4.99306927511967e-05, "loss": 2.242590744572226e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 287, "tokens/total": 8683296, "tokens/train_per_sec_per_gpu": 35.5, "tokens/trainable": 131343}
+{"epoch": 1.125, "grad_norm": 0.22225140035152435, "learning_rate": 4.964235714846775e-05, "loss": 0.0026556546799838543, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00266, "step": 288, "tokens/total": 8713504, "tokens/train_per_sec_per_gpu": 32.23, "tokens/trainable": 131763}
+{"epoch": 1.12890625, "grad_norm": 0.018996328115463257, "learning_rate": 4.9354244499126866e-05, "loss": 4.354536213213578e-05, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00004, "step": 289, "tokens/total": 8743888, "tokens/train_per_sec_per_gpu": 34.0, "tokens/trainable": 132229}
+{"epoch": 1.1328125, "grad_norm": 0.001890109502710402, "learning_rate": 4.90663667927174e-05, "loss": 2.4500381186953746e-05, "memory/device_reserved (GiB)": 35.54, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00002, "step": 290, "tokens/total": 8774336, "tokens/train_per_sec_per_gpu": 35.19, "tokens/trainable": 132678}
+{"epoch": 1.13671875, "grad_norm": 0.0009667908307164907, "learning_rate": 4.877873600900581e-05, "loss": 2.0667655917350203e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 291, "tokens/total": 8804816, "tokens/train_per_sec_per_gpu": 29.28, "tokens/trainable": 133136}
+{"epoch": 1.140625, "grad_norm": 0.0010472588473930955, "learning_rate": 4.849136411748306e-05, "loss": 2.282073546666652e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00002, "step": 292, "tokens/total": 8835024, "tokens/train_per_sec_per_gpu": 35.25, "tokens/trainable": 133608}
+{"epoch": 1.14453125, "grad_norm": 0.028602443635463715, "learning_rate": 4.8204263076866574e-05, "loss": 0.0002425020356895402, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00024, "step": 293, "tokens/total": 8865408, "tokens/train_per_sec_per_gpu": 40.43, "tokens/trainable": 134108}
+{"epoch": 1.1484375, "grad_norm": 0.0024220976047217846, "learning_rate": 4.791744483460251e-05, "loss": 3.14589160552714e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00003, "step": 294, "tokens/total": 8895824, "tokens/train_per_sec_per_gpu": 30.68, "tokens/trainable": 134541}
+{"epoch": 1.15234375, "grad_norm": 0.00373630179092288, "learning_rate": 4.7630921326368736e-05, "loss": 5.82915308768861e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00006, "step": 295, "tokens/total": 8926336, "tokens/train_per_sec_per_gpu": 29.44, "tokens/trainable": 134966}
+{"epoch": 1.15625, "grad_norm": 0.023777559399604797, "learning_rate": 4.7344704475577916e-05, "loss": 0.0002222473849542439, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00022, "step": 296, "tokens/total": 8956656, "tokens/train_per_sec_per_gpu": 33.53, "tokens/trainable": 135402}
+{"epoch": 1.16015625, "grad_norm": 0.03583608567714691, "learning_rate": 4.705880619288153e-05, "loss": 0.000519716995768249, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00052, "step": 297, "tokens/total": 8986752, "tokens/train_per_sec_per_gpu": 30.02, "tokens/trainable": 135804}
+{"epoch": 1.1640625, "grad_norm": 0.10825730860233307, "learning_rate": 4.677323837567412e-05, "loss": 0.0007585557177662849, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00076, "step": 298, "tokens/total": 9017024, "tokens/train_per_sec_per_gpu": 31.13, "tokens/trainable": 136231}
+{"epoch": 1.16796875, "grad_norm": 0.0007777873543091118, "learning_rate": 4.6488012907598146e-05, "loss": 1.649466503295116e-05, "memory/device_reserved (GiB)": 35.55, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00002, "step": 299, "tokens/total": 9047424, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 136705}
+{"epoch": 1.171875, "grad_norm": 0.0021820615511387587, "learning_rate": 4.620314165804964e-05, "loss": 3.35803342750296e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 300, "tokens/total": 9077648, "tokens/train_per_sec_per_gpu": 34.73, "tokens/trainable": 137178}
+{"epoch": 1.17578125, "grad_norm": 0.017772037535905838, "learning_rate": 4.591863648168407e-05, "loss": 9.50043904595077e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.0001, "step": 301, "tokens/total": 9108032, "tokens/train_per_sec_per_gpu": 34.31, "tokens/trainable": 137643}
+{"epoch": 1.1796875, "grad_norm": 0.08602973073720932, "learning_rate": 4.5634509217923135e-05, "loss": 0.0010838620364665985, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00108, "step": 302, "tokens/total": 9138240, "tokens/train_per_sec_per_gpu": 31.91, "tokens/trainable": 138078}
+{"epoch": 1.18359375, "grad_norm": 0.0017863045213744044, "learning_rate": 4.535077169046201e-05, "loss": 3.200750143150799e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 303, "tokens/total": 9168352, "tokens/train_per_sec_per_gpu": 35.0, "tokens/trainable": 138515}
+{"epoch": 1.1875, "grad_norm": 0.0011891268659383059, "learning_rate": 4.506743570677743e-05, "loss": 2.6663004973670468e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00003, "step": 304, "tokens/total": 9198864, "tokens/train_per_sec_per_gpu": 35.13, "tokens/trainable": 138948}
+{"epoch": 1.19140625, "grad_norm": 0.08658935129642487, "learning_rate": 4.478451305763618e-05, "loss": 0.0008552016224712133, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00086, "step": 305, "tokens/total": 9229104, "tokens/train_per_sec_per_gpu": 38.05, "tokens/trainable": 139408}
+{"epoch": 1.1953125, "grad_norm": 0.0014755144948139787, "learning_rate": 4.450201551660454e-05, "loss": 2.39577166212257e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.38, "memory/max_allocated (GiB)": 33.38, "ppl": 1.00002, "step": 306, "tokens/total": 9257344, "tokens/train_per_sec_per_gpu": 33.38, "tokens/trainable": 139840}
+{"epoch": 1.19921875, "grad_norm": 0.014350412413477898, "learning_rate": 4.4219954839558276e-05, "loss": 0.00012315387721173465, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00012, "step": 307, "tokens/total": 9287520, "tokens/train_per_sec_per_gpu": 32.89, "tokens/trainable": 140281}
+{"epoch": 1.203125, "grad_norm": 0.0019020310137420893, "learning_rate": 4.393834276419352e-05, "loss": 2.20383644773392e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 308, "tokens/total": 9317888, "tokens/train_per_sec_per_gpu": 37.31, "tokens/trainable": 140776}
+{"epoch": 1.20703125, "grad_norm": 0.0015318727819249034, "learning_rate": 4.36571910095382e-05, "loss": 2.580175168986898e-05, "memory/device_reserved (GiB)": 35.72, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 309, "tokens/total": 9348256, "tokens/train_per_sec_per_gpu": 36.82, "tokens/trainable": 141228}
+{"epoch": 1.2109375, "grad_norm": 0.00613615894690156, "learning_rate": 4.337651127546448e-05, "loss": 6.0944184951949865e-05, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00006, "step": 310, "tokens/total": 9378560, "tokens/train_per_sec_per_gpu": 37.19, "tokens/trainable": 141679}
+{"epoch": 1.21484375, "grad_norm": 0.001598753617145121, "learning_rate": 4.3096315242201736e-05, "loss": 2.0053470507264137e-05, "memory/device_reserved (GiB)": 35.74, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00002, "step": 311, "tokens/total": 9408848, "tokens/train_per_sec_per_gpu": 32.25, "tokens/trainable": 142122}
+{"epoch": 1.21875, "grad_norm": 0.0007549300789833069, "learning_rate": 4.2816614569850635e-05, "loss": 1.628213794901967e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00002, "step": 312, "tokens/total": 9439344, "tokens/train_per_sec_per_gpu": 33.01, "tokens/trainable": 142561}
+{"epoch": 1.22265625, "grad_norm": 0.00085266592213884, "learning_rate": 4.2537420897897864e-05, "loss": 1.5117442671908066e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 313, "tokens/total": 9469792, "tokens/train_per_sec_per_gpu": 35.55, "tokens/trainable": 143046}
+{"epoch": 1.2265625, "grad_norm": 0.02262088656425476, "learning_rate": 4.225874584473174e-05, "loss": 0.00013078594929538667, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00013, "step": 314, "tokens/total": 9500128, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 143493}
+{"epoch": 1.23046875, "grad_norm": 0.00101909798104316, "learning_rate": 4.19806010071587e-05, "loss": 1.9173825421603397e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00002, "step": 315, "tokens/total": 9530480, "tokens/train_per_sec_per_gpu": 35.12, "tokens/trainable": 143956}
+{"epoch": 1.234375, "grad_norm": 0.023366861045360565, "learning_rate": 4.170299795992081e-05, "loss": 0.0001811327674658969, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00018, "step": 316, "tokens/total": 9560912, "tokens/train_per_sec_per_gpu": 33.98, "tokens/trainable": 144411}
+{"epoch": 1.23828125, "grad_norm": 0.011399022303521633, "learning_rate": 4.142594825521398e-05, "loss": 0.00012808202882297337, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00013, "step": 317, "tokens/total": 9591344, "tokens/train_per_sec_per_gpu": 38.66, "tokens/trainable": 144892}
+{"epoch": 1.2421875, "grad_norm": 0.5837145447731018, "learning_rate": 4.114946342220728e-05, "loss": 0.006938215810805559, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00696, "step": 318, "tokens/total": 9621392, "tokens/train_per_sec_per_gpu": 32.65, "tokens/trainable": 145339}
+{"epoch": 1.24609375, "grad_norm": 0.0007919945055618882, "learning_rate": 4.087355496656321e-05, "loss": 1.6890848201001063e-05, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00002, "step": 319, "tokens/total": 9651728, "tokens/train_per_sec_per_gpu": 36.38, "tokens/trainable": 145811}
+{"epoch": 1.25, "grad_norm": 0.025993503630161285, "learning_rate": 4.05982343699588e-05, "loss": 0.0002537990512792021, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00025, "step": 320, "tokens/total": 9682224, "tokens/train_per_sec_per_gpu": 38.62, "tokens/trainable": 146314}
+{"epoch": 1.25390625, "grad_norm": 0.0010733718518167734, "learning_rate": 4.0323513089607876e-05, "loss": 1.948333010659553e-05, "memory/device_reserved (GiB)": 37.95, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00002, "step": 321, "tokens/total": 9712656, "tokens/train_per_sec_per_gpu": 31.72, "tokens/trainable": 146781}
+{"epoch": 1.2578125, "grad_norm": 0.0843457579612732, "learning_rate": 4.004940255778431e-05, "loss": 0.0008847219287417829, "memory/device_reserved (GiB)": 34.89, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00089, "step": 322, "tokens/total": 9743088, "tokens/train_per_sec_per_gpu": 35.55, "tokens/trainable": 147225}
+{"epoch": 1.26171875, "grad_norm": 0.09092428535223007, "learning_rate": 3.977591418134619e-05, "loss": 0.00040022452594712377, "memory/device_reserved (GiB)": 35.17, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0004, "step": 323, "tokens/total": 9773808, "tokens/train_per_sec_per_gpu": 34.78, "tokens/trainable": 147673}
+{"epoch": 1.265625, "grad_norm": 0.3859696090221405, "learning_rate": 3.95030593412612e-05, "loss": 0.004064415581524372, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00407, "step": 324, "tokens/total": 9804016, "tokens/train_per_sec_per_gpu": 33.23, "tokens/trainable": 148103}
+{"epoch": 1.26953125, "grad_norm": 0.000418124720454216, "learning_rate": 3.923084939213296e-05, "loss": 9.266825145459734e-06, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 325, "tokens/total": 9834592, "tokens/train_per_sec_per_gpu": 31.74, "tokens/trainable": 148535}
+{"epoch": 1.2734375, "grad_norm": 0.030438628047704697, "learning_rate": 3.895929566172861e-05, "loss": 0.00029550789622589946, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0003, "step": 326, "tokens/total": 9864848, "tokens/train_per_sec_per_gpu": 34.02, "tokens/trainable": 148999}
+{"epoch": 1.27734375, "grad_norm": 0.0003460803418420255, "learning_rate": 3.868840945050728e-05, "loss": 9.424240488442592e-06, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00001, "step": 327, "tokens/total": 9895168, "tokens/train_per_sec_per_gpu": 31.64, "tokens/trainable": 149431}
+{"epoch": 1.28125, "grad_norm": 0.32631534337997437, "learning_rate": 3.841820203114995e-05, "loss": 0.004381683189421892, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00439, "step": 328, "tokens/total": 9925696, "tokens/train_per_sec_per_gpu": 34.18, "tokens/trainable": 149886}
+{"epoch": 1.28515625, "grad_norm": 0.06805918365716934, "learning_rate": 3.814868464809027e-05, "loss": 0.0003639076603576541, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00036, "step": 329, "tokens/total": 9955648, "tokens/train_per_sec_per_gpu": 31.01, "tokens/trainable": 150288}
+{"epoch": 1.2890625, "grad_norm": 0.004817479755729437, "learning_rate": 3.787986851704667e-05, "loss": 3.246869164286181e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00003, "step": 330, "tokens/total": 9985856, "tokens/train_per_sec_per_gpu": 33.94, "tokens/trainable": 150733}
+{"epoch": 1.29296875, "grad_norm": 0.06923647969961166, "learning_rate": 3.7611764824555654e-05, "loss": 0.00031070224940776825, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00031, "step": 331, "tokens/total": 10016208, "tokens/train_per_sec_per_gpu": 36.73, "tokens/trainable": 151207}
+{"epoch": 1.296875, "grad_norm": 0.14731979370117188, "learning_rate": 3.734438472750619e-05, "loss": 0.0015139597235247493, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00152, "step": 332, "tokens/total": 10046512, "tokens/train_per_sec_per_gpu": 37.61, "tokens/trainable": 151694}
+{"epoch": 1.30078125, "grad_norm": 0.007325666956603527, "learning_rate": 3.707773935267552e-05, "loss": 7.809747330611572e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00008, "step": 333, "tokens/total": 10076944, "tokens/train_per_sec_per_gpu": 38.37, "tokens/trainable": 152188}
+{"epoch": 1.3046875, "grad_norm": 0.00047597952652722597, "learning_rate": 3.68118397962661e-05, "loss": 1.2739032172248699e-05, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 334, "tokens/total": 10107296, "tokens/train_per_sec_per_gpu": 30.69, "tokens/trainable": 152596}
+{"epoch": 1.30859375, "grad_norm": 0.0700320228934288, "learning_rate": 3.654669712344384e-05, "loss": 0.00047467637341469526, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00047, "step": 335, "tokens/total": 10137568, "tokens/train_per_sec_per_gpu": 36.57, "tokens/trainable": 153052}
+{"epoch": 1.3125, "grad_norm": 0.06498395651578903, "learning_rate": 3.628232236787763e-05, "loss": 0.00041414948645979166, "memory/device_reserved (GiB)": 37.17, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00041, "step": 336, "tokens/total": 10167952, "tokens/train_per_sec_per_gpu": 30.19, "tokens/trainable": 153491}
+{"epoch": 1.31640625, "grad_norm": 0.009991639293730259, "learning_rate": 3.6018726531280144e-05, "loss": 7.060338975861669e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00007, "step": 337, "tokens/total": 10198512, "tokens/train_per_sec_per_gpu": 40.43, "tokens/trainable": 153983}
+{"epoch": 1.3203125, "grad_norm": 0.03267490491271019, "learning_rate": 3.575592058295017e-05, "loss": 0.0002045792352873832, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0002, "step": 338, "tokens/total": 10228752, "tokens/train_per_sec_per_gpu": 36.72, "tokens/trainable": 154464}
+{"epoch": 1.32421875, "grad_norm": 0.00039529221248812973, "learning_rate": 3.549391545931585e-05, "loss": 8.338471161550842e-06, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 339, "tokens/total": 10259104, "tokens/train_per_sec_per_gpu": 35.87, "tokens/trainable": 154924}
+{"epoch": 1.328125, "grad_norm": 0.0010571131715551019, "learning_rate": 3.5232722063479914e-05, "loss": 1.6815669368952513e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 340, "tokens/total": 10289520, "tokens/train_per_sec_per_gpu": 30.81, "tokens/trainable": 155373}
+{"epoch": 1.33203125, "grad_norm": 0.0008337291656062007, "learning_rate": 3.49723512647657e-05, "loss": 1.703310408629477e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 341, "tokens/total": 10320016, "tokens/train_per_sec_per_gpu": 36.47, "tokens/trainable": 155873}
+{"epoch": 1.3359375, "grad_norm": 0.0007872325950302184, "learning_rate": 3.471281389826491e-05, "loss": 1.626565062906593e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 342, "tokens/total": 10350368, "tokens/train_per_sec_per_gpu": 30.42, "tokens/trainable": 156278}
+{"epoch": 1.33984375, "grad_norm": 0.21220935881137848, "learning_rate": 3.4454120764386764e-05, "loss": 0.0009314992348663509, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00093, "step": 343, "tokens/total": 10380624, "tokens/train_per_sec_per_gpu": 35.4, "tokens/trainable": 156733}
+{"epoch": 1.34375, "grad_norm": 0.017469989135861397, "learning_rate": 3.4196282628408526e-05, "loss": 7.496406033169478e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00007, "step": 344, "tokens/total": 10411024, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 157203}
+{"epoch": 1.34765625, "grad_norm": 0.2605672776699066, "learning_rate": 3.3939310220027456e-05, "loss": 0.004417422227561474, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00443, "step": 345, "tokens/total": 10441248, "tokens/train_per_sec_per_gpu": 37.8, "tokens/trainable": 157707}
+{"epoch": 1.3515625, "grad_norm": 0.00032958327210508287, "learning_rate": 3.3683214232914404e-05, "loss": 7.68474092183169e-06, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00001, "step": 346, "tokens/total": 10471712, "tokens/train_per_sec_per_gpu": 35.01, "tokens/trainable": 158180}
+{"epoch": 1.35546875, "grad_norm": 0.0021204655058681965, "learning_rate": 3.342800532426873e-05, "loss": 2.9396429454209283e-05, "memory/device_reserved (GiB)": 38.13, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00003, "step": 347, "tokens/total": 10502288, "tokens/train_per_sec_per_gpu": 34.41, "tokens/trainable": 158646}
+{"epoch": 1.359375, "grad_norm": 0.00040693042683415115, "learning_rate": 3.317369411437484e-05, "loss": 9.5013465397642e-06, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 348, "tokens/total": 10532304, "tokens/train_per_sec_per_gpu": 35.66, "tokens/trainable": 159115}
+{"epoch": 1.36328125, "grad_norm": 0.006379029247909784, "learning_rate": 3.292029118616024e-05, "loss": 9.018932178150862e-05, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00009, "step": 349, "tokens/total": 10562608, "tokens/train_per_sec_per_gpu": 31.62, "tokens/trainable": 159538}
+{"epoch": 1.3671875, "grad_norm": 0.005555902142077684, "learning_rate": 3.266780708475511e-05, "loss": 5.1456365326885134e-05, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00005, "step": 350, "tokens/total": 10592992, "tokens/train_per_sec_per_gpu": 33.22, "tokens/trainable": 160016}
+{"epoch": 1.37109375, "grad_norm": 0.009832990355789661, "learning_rate": 3.241625231705354e-05, "loss": 9.037736163008958e-05, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00009, "step": 351, "tokens/total": 10623328, "tokens/train_per_sec_per_gpu": 34.57, "tokens/trainable": 160475}
+{"epoch": 1.375, "grad_norm": 0.0002588493807706982, "learning_rate": 3.216563735127618e-05, "loss": 7.509744136768859e-06, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 352, "tokens/total": 10653632, "tokens/train_per_sec_per_gpu": 34.0, "tokens/trainable": 160952}
+{"epoch": 1.37890625, "grad_norm": 0.03195308893918991, "learning_rate": 3.191597261653475e-05, "loss": 0.0002666166110429913, "memory/device_reserved (GiB)": 38.4, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00027, "step": 353, "tokens/total": 10684176, "tokens/train_per_sec_per_gpu": 32.95, "tokens/trainable": 161420}
+{"epoch": 1.3828125, "grad_norm": 0.07012991607189178, "learning_rate": 3.166726850239794e-05, "loss": 0.000741704716347158, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00074, "step": 354, "tokens/total": 10714352, "tokens/train_per_sec_per_gpu": 34.09, "tokens/trainable": 161868}
+{"epoch": 1.38671875, "grad_norm": 0.0005822654929943383, "learning_rate": 3.141953535845912e-05, "loss": 1.0378402294008993e-05, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00001, "step": 355, "tokens/total": 10744464, "tokens/train_per_sec_per_gpu": 35.44, "tokens/trainable": 162318}
+{"epoch": 1.390625, "grad_norm": 0.012435230426490307, "learning_rate": 3.11727834939056e-05, "loss": 7.89838086348027e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00008, "step": 356, "tokens/total": 10774976, "tokens/train_per_sec_per_gpu": 31.54, "tokens/trainable": 162789}
+{"epoch": 1.39453125, "grad_norm": 0.0045598647557199, "learning_rate": 3.092702317708967e-05, "loss": 2.48450869548833e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00002, "step": 357, "tokens/total": 10805600, "tokens/train_per_sec_per_gpu": 36.43, "tokens/trainable": 163254}
+{"epoch": 1.3984375, "grad_norm": 0.3044262230396271, "learning_rate": 3.0682264635101276e-05, "loss": 0.002298796083778143, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0023, "step": 358, "tokens/total": 10835920, "tokens/train_per_sec_per_gpu": 34.62, "tokens/trainable": 163715}
+{"epoch": 1.40234375, "grad_norm": 0.002458421979099512, "learning_rate": 3.0438518053342407e-05, "loss": 1.9634167983895168e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00002, "step": 359, "tokens/total": 10866048, "tokens/train_per_sec_per_gpu": 34.53, "tokens/trainable": 164197}
+{"epoch": 1.40625, "grad_norm": 0.0024773837067186832, "learning_rate": 3.0195793575103266e-05, "loss": 2.744927041931078e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00003, "step": 360, "tokens/total": 10896288, "tokens/train_per_sec_per_gpu": 31.74, "tokens/trainable": 164661}
+{"epoch": 1.41015625, "grad_norm": 0.00026717092259787023, "learning_rate": 2.9954101301140146e-05, "loss": 6.422977094189264e-06, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 361, "tokens/total": 10926816, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 165111}
+{"epoch": 1.4140625, "grad_norm": 0.0008353493758477271, "learning_rate": 2.9713451289255123e-05, "loss": 1.3549893083109055e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.31, "memory/max_allocated (GiB)": 33.31, "ppl": 1.00001, "step": 362, "tokens/total": 10954928, "tokens/train_per_sec_per_gpu": 30.52, "tokens/trainable": 165552}
+{"epoch": 1.41796875, "grad_norm": 0.006071125157177448, "learning_rate": 2.9473853553877484e-05, "loss": 6.428411870729178e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00006, "step": 363, "tokens/total": 10985328, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 166026}
+{"epoch": 1.421875, "grad_norm": 0.25424960255622864, "learning_rate": 2.9235318065647e-05, "loss": 0.004243595991283655, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00425, "step": 364, "tokens/total": 11015664, "tokens/train_per_sec_per_gpu": 38.11, "tokens/trainable": 166486}
+{"epoch": 1.42578125, "grad_norm": 0.0005575277027674019, "learning_rate": 2.8997854750998964e-05, "loss": 8.403902938880492e-06, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00001, "step": 365, "tokens/total": 11046256, "tokens/train_per_sec_per_gpu": 36.52, "tokens/trainable": 166959}
+{"epoch": 1.4296875, "grad_norm": 0.0014335239538922906, "learning_rate": 2.8761473491751258e-05, "loss": 1.4738036043127067e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.68, "memory/max_allocated (GiB)": 33.68, "ppl": 1.00001, "step": 366, "tokens/total": 11076288, "tokens/train_per_sec_per_gpu": 33.86, "tokens/trainable": 167394}
+{"epoch": 1.43359375, "grad_norm": 0.0010152696631848812, "learning_rate": 2.8526184124692883e-05, "loss": 1.3278609912958927e-05, "memory/device_reserved (GiB)": 36.01, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00001, "step": 367, "tokens/total": 11106416, "tokens/train_per_sec_per_gpu": 28.58, "tokens/trainable": 167832}
+{"epoch": 1.4375, "grad_norm": 0.11941836029291153, "learning_rate": 2.829199644117484e-05, "loss": 0.0006476733833551407, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00065, "step": 368, "tokens/total": 11136848, "tokens/train_per_sec_per_gpu": 31.02, "tokens/trainable": 168242}
+{"epoch": 1.44140625, "grad_norm": 0.006877629552036524, "learning_rate": 2.8058920186702553e-05, "loss": 7.449048280250281e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00007, "step": 369, "tokens/total": 11167376, "tokens/train_per_sec_per_gpu": 35.74, "tokens/trainable": 168715}
+{"epoch": 1.4453125, "grad_norm": 0.0007151139434427023, "learning_rate": 2.782696506053033e-05, "loss": 1.1437590728746727e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 370, "tokens/total": 11197776, "tokens/train_per_sec_per_gpu": 31.86, "tokens/trainable": 169147}
+{"epoch": 1.44921875, "grad_norm": 0.009094453416764736, "learning_rate": 2.7596140715257824e-05, "loss": 5.9242345741949975e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00006, "step": 371, "tokens/total": 11227840, "tokens/train_per_sec_per_gpu": 34.35, "tokens/trainable": 169628}
+{"epoch": 1.453125, "grad_norm": 0.000247267191298306, "learning_rate": 2.7366456756428184e-05, "loss": 7.548428584414069e-06, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 372, "tokens/total": 11258176, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 170085}
+{"epoch": 1.45703125, "grad_norm": 0.0003029497747775167, "learning_rate": 2.7137922742128486e-05, "loss": 5.4013939916330855e-06, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00001, "step": 373, "tokens/total": 11288336, "tokens/train_per_sec_per_gpu": 35.6, "tokens/trainable": 170584}
+{"epoch": 1.4609375, "grad_norm": 0.001942179980687797, "learning_rate": 2.691054818259188e-05, "loss": 2.380511250521522e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00002, "step": 374, "tokens/total": 11318848, "tokens/train_per_sec_per_gpu": 34.68, "tokens/trainable": 171058}
+{"epoch": 1.46484375, "grad_norm": 0.0019308892078697681, "learning_rate": 2.6684342539801933e-05, "loss": 1.5233906196954194e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00002, "step": 375, "tokens/total": 11349168, "tokens/train_per_sec_per_gpu": 33.94, "tokens/trainable": 171518}
+{"epoch": 1.46875, "grad_norm": 0.6044301986694336, "learning_rate": 2.645931522709877e-05, "loss": 0.006350134499371052, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00637, "step": 376, "tokens/total": 11379328, "tokens/train_per_sec_per_gpu": 40.62, "tokens/trainable": 172007}
+{"epoch": 1.47265625, "grad_norm": 0.25446587800979614, "learning_rate": 2.6235475608787365e-05, "loss": 0.0018724403344094753, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00187, "step": 377, "tokens/total": 11409568, "tokens/train_per_sec_per_gpu": 34.78, "tokens/trainable": 172468}
+{"epoch": 1.4765625, "grad_norm": 0.0019207323202863336, "learning_rate": 2.6012832999747916e-05, "loss": 3.049923907383345e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00003, "step": 378, "tokens/total": 11439968, "tokens/train_per_sec_per_gpu": 33.76, "tokens/trainable": 172923}
+{"epoch": 1.48046875, "grad_norm": 0.33491548895835876, "learning_rate": 2.579139666504821e-05, "loss": 0.02673855796456337, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0271, "step": 379, "tokens/total": 11470496, "tokens/train_per_sec_per_gpu": 34.19, "tokens/trainable": 173370}
+{"epoch": 1.484375, "grad_norm": 0.006177723873406649, "learning_rate": 2.557117581955798e-05, "loss": 5.195035191718489e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00005, "step": 380, "tokens/total": 11500720, "tokens/train_per_sec_per_gpu": 34.45, "tokens/trainable": 173837}
+{"epoch": 1.48828125, "grad_norm": 0.06973031163215637, "learning_rate": 2.5352179627565532e-05, "loss": 0.0006310560274869204, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00063, "step": 381, "tokens/total": 11531280, "tokens/train_per_sec_per_gpu": 35.54, "tokens/trainable": 174294}
+{"epoch": 1.4921875, "grad_norm": 0.0039021887350827456, "learning_rate": 2.5134417202396277e-05, "loss": 4.849781544180587e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00005, "step": 382, "tokens/total": 11561264, "tokens/train_per_sec_per_gpu": 30.84, "tokens/trainable": 174701}
+{"epoch": 1.49609375, "grad_norm": 0.0007975143962539732, "learning_rate": 2.491789760603361e-05, "loss": 2.3489263185183518e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 383, "tokens/total": 11591616, "tokens/train_per_sec_per_gpu": 32.05, "tokens/trainable": 175131}
+{"epoch": 1.5, "grad_norm": 0.012541128322482109, "learning_rate": 2.4702629848741764e-05, "loss": 0.00011495217040646821, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00011, "step": 384, "tokens/total": 11622000, "tokens/train_per_sec_per_gpu": 34.02, "tokens/trainable": 175586}
+{"epoch": 1.50390625, "grad_norm": 0.004752015229314566, "learning_rate": 2.4488622888690785e-05, "loss": 6.502695032395422e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00007, "step": 385, "tokens/total": 11652352, "tokens/train_per_sec_per_gpu": 34.19, "tokens/trainable": 176026}
+{"epoch": 1.5078125, "grad_norm": 0.014192809350788593, "learning_rate": 2.427588563158384e-05, "loss": 8.075163350440562e-05, "memory/device_reserved (GiB)": 34.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00008, "step": 386, "tokens/total": 11682736, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 176512}
+{"epoch": 1.51171875, "grad_norm": 0.007288333959877491, "learning_rate": 2.406442693028651e-05, "loss": 8.845872798701748e-05, "memory/device_reserved (GiB)": 34.98, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00009, "step": 387, "tokens/total": 11713136, "tokens/train_per_sec_per_gpu": 30.45, "tokens/trainable": 176943}
+{"epoch": 1.515625, "grad_norm": 0.10162956267595291, "learning_rate": 2.3854255584458547e-05, "loss": 0.0007253064541146159, "memory/device_reserved (GiB)": 35.23, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00073, "step": 388, "tokens/total": 11743808, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 177370}
+{"epoch": 1.51953125, "grad_norm": 0.005729300435632467, "learning_rate": 2.3645380340187508e-05, "loss": 8.652975520817563e-05, "memory/device_reserved (GiB)": 35.58, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00009, "step": 389, "tokens/total": 11774144, "tokens/train_per_sec_per_gpu": 35.2, "tokens/trainable": 177851}
+{"epoch": 1.5234375, "grad_norm": 0.0715414360165596, "learning_rate": 2.3437809889624914e-05, "loss": 0.0007950748549774289, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0008, "step": 390, "tokens/total": 11804272, "tokens/train_per_sec_per_gpu": 34.75, "tokens/trainable": 178310}
+{"epoch": 1.52734375, "grad_norm": 0.009775097481906414, "learning_rate": 2.3231552870624487e-05, "loss": 0.00010615254723234102, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00011, "step": 391, "tokens/total": 11832432, "tokens/train_per_sec_per_gpu": 36.73, "tokens/trainable": 178736}
+{"epoch": 1.53125, "grad_norm": 0.0015552763361483812, "learning_rate": 2.3026617866382657e-05, "loss": 2.954876617877744e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 392, "tokens/total": 11862960, "tokens/train_per_sec_per_gpu": 36.18, "tokens/trainable": 179233}
+{"epoch": 1.53515625, "grad_norm": 0.026106838136911392, "learning_rate": 2.2823013405081507e-05, "loss": 0.00021109850786160678, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00021, "step": 393, "tokens/total": 11893232, "tokens/train_per_sec_per_gpu": 38.57, "tokens/trainable": 179730}
+{"epoch": 1.5390625, "grad_norm": 0.020347680896520615, "learning_rate": 2.2620747959533722e-05, "loss": 0.0002601295418571681, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00026, "step": 394, "tokens/total": 11923664, "tokens/train_per_sec_per_gpu": 37.87, "tokens/trainable": 180227}
+{"epoch": 1.54296875, "grad_norm": 0.005419979803264141, "learning_rate": 2.2419829946830123e-05, "loss": 4.929217175231315e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00005, "step": 395, "tokens/total": 11954032, "tokens/train_per_sec_per_gpu": 32.92, "tokens/trainable": 180687}
+{"epoch": 1.546875, "grad_norm": 0.0031002764590084553, "learning_rate": 2.2220267727989325e-05, "loss": 6.144218787085265e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00006, "step": 396, "tokens/total": 11984512, "tokens/train_per_sec_per_gpu": 32.89, "tokens/trainable": 181141}
+{"epoch": 1.55078125, "grad_norm": 0.006748533807694912, "learning_rate": 2.202206960760984e-05, "loss": 9.171784040518105e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00009, "step": 397, "tokens/total": 12014880, "tokens/train_per_sec_per_gpu": 36.02, "tokens/trainable": 181648}
+{"epoch": 1.5546875, "grad_norm": 0.08927815407514572, "learning_rate": 2.182524383352446e-05, "loss": 0.001000746968202293, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.001, "step": 398, "tokens/total": 12044928, "tokens/train_per_sec_per_gpu": 33.93, "tokens/trainable": 182109}
+{"epoch": 1.55859375, "grad_norm": 0.004651801194995642, "learning_rate": 2.1629798596457056e-05, "loss": 5.3439554903889075e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00005, "step": 399, "tokens/total": 12075344, "tokens/train_per_sec_per_gpu": 31.58, "tokens/trainable": 182559}
+{"epoch": 1.5625, "grad_norm": 0.0017514342907816172, "learning_rate": 2.1435742029681725e-05, "loss": 4.570486271404661e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00005, "step": 400, "tokens/total": 12105616, "tokens/train_per_sec_per_gpu": 33.17, "tokens/trainable": 182985}
+{"epoch": 1.56640625, "grad_norm": 0.01574699766933918, "learning_rate": 2.124308220868431e-05, "loss": 0.0001242965809069574, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00012, "step": 401, "tokens/total": 12136080, "tokens/train_per_sec_per_gpu": 32.7, "tokens/trainable": 183440}
+{"epoch": 1.5703125, "grad_norm": 0.007152364123612642, "learning_rate": 2.105182715082638e-05, "loss": 0.0001222842838615179, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00012, "step": 402, "tokens/total": 12166240, "tokens/train_per_sec_per_gpu": 34.51, "tokens/trainable": 183917}
+{"epoch": 1.57421875, "grad_norm": 0.0015342400874942541, "learning_rate": 2.0861984815011552e-05, "loss": 3.621872019721195e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00004, "step": 403, "tokens/total": 12196544, "tokens/train_per_sec_per_gpu": 30.36, "tokens/trainable": 184348}
+{"epoch": 1.578125, "grad_norm": 0.0035214691888540983, "learning_rate": 2.0673563101354323e-05, "loss": 6.367819150909781e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00006, "step": 404, "tokens/total": 12226784, "tokens/train_per_sec_per_gpu": 31.51, "tokens/trainable": 184786}
+{"epoch": 1.58203125, "grad_norm": 0.0007211520569398999, "learning_rate": 2.0486569850851317e-05, "loss": 2.0325338482507505e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00002, "step": 405, "tokens/total": 12257264, "tokens/train_per_sec_per_gpu": 33.86, "tokens/trainable": 185249}
+{"epoch": 1.5859375, "grad_norm": 0.0014804014936089516, "learning_rate": 2.0301012845054956e-05, "loss": 3.379736153874546e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00003, "step": 406, "tokens/total": 12287584, "tokens/train_per_sec_per_gpu": 31.43, "tokens/trainable": 185680}
+{"epoch": 1.58984375, "grad_norm": 0.004397980403155088, "learning_rate": 2.011689980574966e-05, "loss": 5.1796458137687296e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00005, "step": 407, "tokens/total": 12317984, "tokens/train_per_sec_per_gpu": 31.71, "tokens/trainable": 186100}
+{"epoch": 1.59375, "grad_norm": 0.0026473007164895535, "learning_rate": 1.993423839463052e-05, "loss": 3.441090666456148e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00003, "step": 408, "tokens/total": 12348144, "tokens/train_per_sec_per_gpu": 39.84, "tokens/trainable": 186565}
+{"epoch": 1.59765625, "grad_norm": 0.0289909727871418, "learning_rate": 1.975303621298445e-05, "loss": 0.00021602815832011402, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00022, "step": 409, "tokens/total": 12378544, "tokens/train_per_sec_per_gpu": 35.78, "tokens/trainable": 187017}
+{"epoch": 1.6015625, "grad_norm": 0.007092812564224005, "learning_rate": 1.957330080137385e-05, "loss": 9.646185935707763e-05, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0001, "step": 410, "tokens/total": 12408784, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 187468}
+{"epoch": 1.60546875, "grad_norm": 0.02710823155939579, "learning_rate": 1.9395039639322864e-05, "loss": 0.0003458422143012285, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00035, "step": 411, "tokens/total": 12438832, "tokens/train_per_sec_per_gpu": 37.57, "tokens/trainable": 187961}
+{"epoch": 1.609375, "grad_norm": 0.29485711455345154, "learning_rate": 1.9218260145006073e-05, "loss": 0.0032923028338700533, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.0033, "step": 412, "tokens/total": 12469296, "tokens/train_per_sec_per_gpu": 35.93, "tokens/trainable": 188447}
+{"epoch": 1.61328125, "grad_norm": 0.22881586849689484, "learning_rate": 1.904296967493982e-05, "loss": 0.0035809341352432966, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00359, "step": 413, "tokens/total": 12499392, "tokens/train_per_sec_per_gpu": 35.97, "tokens/trainable": 188903}
+{"epoch": 1.6171875, "grad_norm": 0.0026347371749579906, "learning_rate": 1.8869175523676064e-05, "loss": 6.0475373174995184e-05, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 34.0, "memory/max_allocated (GiB)": 34.0, "ppl": 1.00006, "step": 414, "tokens/total": 12529952, "tokens/train_per_sec_per_gpu": 36.68, "tokens/trainable": 189398}
+{"epoch": 1.62109375, "grad_norm": 0.0017223359318450093, "learning_rate": 1.869688492349885e-05, "loss": 4.137849100516178e-05, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00004, "step": 415, "tokens/total": 12560240, "tokens/train_per_sec_per_gpu": 34.16, "tokens/trainable": 189850}
+{"epoch": 1.625, "grad_norm": 0.011837545782327652, "learning_rate": 1.85261050441233e-05, "loss": 0.00011106643069069833, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00011, "step": 416, "tokens/total": 12590736, "tokens/train_per_sec_per_gpu": 32.68, "tokens/trainable": 190306}
+{"epoch": 1.62890625, "grad_norm": 0.01777251996099949, "learning_rate": 1.8356842992397304e-05, "loss": 0.0001219596597366035, "memory/device_reserved (GiB)": 38.05, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00012, "step": 417, "tokens/total": 12621168, "tokens/train_per_sec_per_gpu": 34.21, "tokens/trainable": 190746}
+{"epoch": 1.6328125, "grad_norm": 0.003218573285266757, "learning_rate": 1.8189105812005714e-05, "loss": 5.22282425663434e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00005, "step": 418, "tokens/total": 12651456, "tokens/train_per_sec_per_gpu": 31.15, "tokens/trainable": 191156}
+{"epoch": 1.63671875, "grad_norm": 0.013744015246629715, "learning_rate": 1.802290048317732e-05, "loss": 0.0001364837517030537, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00014, "step": 419, "tokens/total": 12681760, "tokens/train_per_sec_per_gpu": 29.53, "tokens/trainable": 191573}
+{"epoch": 1.640625, "grad_norm": 0.25013473629951477, "learning_rate": 1.785823392239424e-05, "loss": 0.002094803610816598, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.0021, "step": 420, "tokens/total": 12712144, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 192014}
+{"epoch": 1.64453125, "grad_norm": 0.14290185272693634, "learning_rate": 1.7695112982104225e-05, "loss": 0.001426510512828827, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00143, "step": 421, "tokens/total": 12742400, "tokens/train_per_sec_per_gpu": 32.87, "tokens/trainable": 192453}
+{"epoch": 1.6484375, "grad_norm": 0.0009390547638759017, "learning_rate": 1.7533544450435433e-05, "loss": 1.822916055971291e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00002, "step": 422, "tokens/total": 12772288, "tokens/train_per_sec_per_gpu": 33.18, "tokens/trainable": 192886}
+{"epoch": 1.65234375, "grad_norm": 0.004544281866401434, "learning_rate": 1.7373535050913946e-05, "loss": 6.211431173142046e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00006, "step": 423, "tokens/total": 12802880, "tokens/train_per_sec_per_gpu": 29.87, "tokens/trainable": 193323}
+{"epoch": 1.65625, "grad_norm": 0.039859045296907425, "learning_rate": 1.721509144218405e-05, "loss": 0.0005113891093060374, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00051, "step": 424, "tokens/total": 12833456, "tokens/train_per_sec_per_gpu": 32.55, "tokens/trainable": 193753}
+{"epoch": 1.66015625, "grad_norm": 0.00408409908413887, "learning_rate": 1.705822021773101e-05, "loss": 7.249199552461505e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00007, "step": 425, "tokens/total": 12863952, "tokens/train_per_sec_per_gpu": 36.41, "tokens/trainable": 194240}
+{"epoch": 1.6640625, "grad_norm": 0.13116440176963806, "learning_rate": 1.69029279056068e-05, "loss": 0.0015061571029946208, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00151, "step": 426, "tokens/total": 12894352, "tokens/train_per_sec_per_gpu": 35.54, "tokens/trainable": 194680}
+{"epoch": 1.66796875, "grad_norm": 0.010440024547278881, "learning_rate": 1.6749220968158415e-05, "loss": 8.445358253084123e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00008, "step": 427, "tokens/total": 12924688, "tokens/train_per_sec_per_gpu": 38.77, "tokens/trainable": 195186}
+{"epoch": 1.671875, "grad_norm": 0.007594608701765537, "learning_rate": 1.659710580175893e-05, "loss": 9.946282807504758e-05, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0001, "step": 428, "tokens/total": 12955072, "tokens/train_per_sec_per_gpu": 36.52, "tokens/trainable": 195693}
+{"epoch": 1.67578125, "grad_norm": 0.006465950049459934, "learning_rate": 1.644658873654133e-05, "loss": 0.00013557568308897316, "memory/device_reserved (GiB)": 36.05, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00014, "step": 429, "tokens/total": 12985328, "tokens/train_per_sec_per_gpu": 36.95, "tokens/trainable": 196130}
+{"epoch": 1.6796875, "grad_norm": 0.00230405549518764, "learning_rate": 1.629767603613508e-05, "loss": 3.7114587030373514e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00004, "step": 430, "tokens/total": 13015600, "tokens/train_per_sec_per_gpu": 29.53, "tokens/trainable": 196530}
+{"epoch": 1.68359375, "grad_norm": 0.0041900877840816975, "learning_rate": 1.615037389740547e-05, "loss": 6.116987788118422e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00006, "step": 431, "tokens/total": 13045728, "tokens/train_per_sec_per_gpu": 29.94, "tokens/trainable": 196974}
+{"epoch": 1.6875, "grad_norm": 0.02951621450483799, "learning_rate": 1.600468845019576e-05, "loss": 0.0004236411186866462, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00042, "step": 432, "tokens/total": 13075920, "tokens/train_per_sec_per_gpu": 35.24, "tokens/trainable": 197454}
+{"epoch": 1.69140625, "grad_norm": 0.0003202867810614407, "learning_rate": 1.5860625757072092e-05, "loss": 1.1190046279807575e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 433, "tokens/total": 13106128, "tokens/train_per_sec_per_gpu": 31.43, "tokens/trainable": 197902}
+{"epoch": 1.6953125, "grad_norm": 0.0006868013297207654, "learning_rate": 1.571819181307116e-05, "loss": 2.1185776859056205e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00002, "step": 434, "tokens/total": 13136368, "tokens/train_per_sec_per_gpu": 36.06, "tokens/trainable": 198374}
+{"epoch": 1.69921875, "grad_norm": 0.0025407401844859123, "learning_rate": 1.557739254545075e-05, "loss": 4.928320413455367e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00005, "step": 435, "tokens/total": 13166704, "tokens/train_per_sec_per_gpu": 37.59, "tokens/trainable": 198846}
+{"epoch": 1.703125, "grad_norm": 0.0013901089550927281, "learning_rate": 1.543823381344311e-05, "loss": 3.577578900149092e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00004, "step": 436, "tokens/total": 13197344, "tokens/train_per_sec_per_gpu": 37.67, "tokens/trainable": 199320}
+{"epoch": 1.70703125, "grad_norm": 0.003646081080660224, "learning_rate": 1.5300721408011114e-05, "loss": 6.294051127042621e-05, "memory/device_reserved (GiB)": 36.06, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00006, "step": 437, "tokens/total": 13227664, "tokens/train_per_sec_per_gpu": 37.46, "tokens/trainable": 199777}
+{"epoch": 1.7109375, "grad_norm": 0.030430182814598083, "learning_rate": 1.5164861051607254e-05, "loss": 0.0001770013477653265, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00018, "step": 438, "tokens/total": 13257632, "tokens/train_per_sec_per_gpu": 30.93, "tokens/trainable": 200194}
+{"epoch": 1.71484375, "grad_norm": 0.0018230377463623881, "learning_rate": 1.5030658397935521e-05, "loss": 3.7613608583342284e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 439, "tokens/total": 13288240, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 200663}
+{"epoch": 1.71875, "grad_norm": 0.0029376039747148752, "learning_rate": 1.4898119031716104e-05, "loss": 4.779352093464695e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00005, "step": 440, "tokens/total": 13318992, "tokens/train_per_sec_per_gpu": 33.85, "tokens/trainable": 201121}
+{"epoch": 1.72265625, "grad_norm": 0.0018373411148786545, "learning_rate": 1.476724846845306e-05, "loss": 3.149824624415487e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 441, "tokens/total": 13349392, "tokens/train_per_sec_per_gpu": 39.81, "tokens/trainable": 201598}
+{"epoch": 1.7265625, "grad_norm": 0.0015428521437570453, "learning_rate": 1.463805215420471e-05, "loss": 2.0810390196857043e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 442, "tokens/total": 13379648, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 202068}
+{"epoch": 1.73046875, "grad_norm": 0.0010770867811515927, "learning_rate": 1.451053546535705e-05, "loss": 2.303836117789615e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00002, "step": 443, "tokens/total": 13409936, "tokens/train_per_sec_per_gpu": 35.5, "tokens/trainable": 202524}
+{"epoch": 1.734375, "grad_norm": 0.006372133269906044, "learning_rate": 1.438470370840001e-05, "loss": 7.449327677022666e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00007, "step": 444, "tokens/total": 13440288, "tokens/train_per_sec_per_gpu": 36.33, "tokens/trainable": 202996}
+{"epoch": 1.73828125, "grad_norm": 0.007094603031873703, "learning_rate": 1.4260562119706606e-05, "loss": 4.211071063764393e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00004, "step": 445, "tokens/total": 13468880, "tokens/train_per_sec_per_gpu": 39.19, "tokens/trainable": 203470}
+{"epoch": 1.7421875, "grad_norm": 0.006931444630026817, "learning_rate": 1.413811586531508e-05, "loss": 8.66219779709354e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00009, "step": 446, "tokens/total": 13499184, "tokens/train_per_sec_per_gpu": 35.29, "tokens/trainable": 203919}
+{"epoch": 1.74609375, "grad_norm": 0.0041664596647024155, "learning_rate": 1.4017370040713884e-05, "loss": 4.786982390214689e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00005, "step": 447, "tokens/total": 13529632, "tokens/train_per_sec_per_gpu": 33.24, "tokens/trainable": 204376}
+{"epoch": 1.75, "grad_norm": 0.042264848947525024, "learning_rate": 1.3898329670629645e-05, "loss": 0.00027092613163404167, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00027, "step": 448, "tokens/total": 13560080, "tokens/train_per_sec_per_gpu": 32.06, "tokens/trainable": 204821}
+{"epoch": 1.75390625, "grad_norm": 0.00043735766666941345, "learning_rate": 1.3780999708818058e-05, "loss": 1.2837479516747408e-05, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00001, "step": 449, "tokens/total": 13590496, "tokens/train_per_sec_per_gpu": 33.67, "tokens/trainable": 205271}
+{"epoch": 1.7578125, "grad_norm": 0.0012977999867871404, "learning_rate": 1.3665385037857758e-05, "loss": 2.1102820028318092e-05, "memory/device_reserved (GiB)": 34.92, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 450, "tokens/total": 13620960, "tokens/train_per_sec_per_gpu": 30.92, "tokens/trainable": 205723}
+{"epoch": 1.76171875, "grad_norm": 0.005675352178514004, "learning_rate": 1.3551490468947126e-05, "loss": 7.465962698915973e-05, "memory/device_reserved (GiB)": 35.04, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00007, "step": 451, "tokens/total": 13651344, "tokens/train_per_sec_per_gpu": 32.48, "tokens/trainable": 206169}
+{"epoch": 1.765625, "grad_norm": 0.0013997588539496064, "learning_rate": 1.3439320741704075e-05, "loss": 2.723011130001396e-05, "memory/device_reserved (GiB)": 35.04, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00003, "step": 452, "tokens/total": 13681792, "tokens/train_per_sec_per_gpu": 32.79, "tokens/trainable": 206592}
+{"epoch": 1.76953125, "grad_norm": 0.012491011992096901, "learning_rate": 1.3328880523968808e-05, "loss": 9.182744543068111e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00009, "step": 453, "tokens/total": 13712256, "tokens/train_per_sec_per_gpu": 33.07, "tokens/trainable": 207055}
+{"epoch": 1.7734375, "grad_norm": 0.000924609019421041, "learning_rate": 1.3220174411609587e-05, "loss": 2.1621295672957785e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00002, "step": 454, "tokens/total": 13742128, "tokens/train_per_sec_per_gpu": 34.61, "tokens/trainable": 207494}
+{"epoch": 1.77734375, "grad_norm": 0.000411411514505744, "learning_rate": 1.3113206928331471e-05, "loss": 1.2427874025888741e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 455, "tokens/total": 13772160, "tokens/train_per_sec_per_gpu": 33.03, "tokens/trainable": 207956}
+{"epoch": 1.78125, "grad_norm": 0.0015909038484096527, "learning_rate": 1.300798252548806e-05, "loss": 3.0332066671689972e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00003, "step": 456, "tokens/total": 13802448, "tokens/train_per_sec_per_gpu": 35.42, "tokens/trainable": 208408}
+{"epoch": 1.78515625, "grad_norm": 0.0030019236728549004, "learning_rate": 1.2904505581896265e-05, "loss": 3.378765541128814e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00003, "step": 457, "tokens/total": 13832880, "tokens/train_per_sec_per_gpu": 35.92, "tokens/trainable": 208889}
+{"epoch": 1.7890625, "grad_norm": 0.0016427476657554507, "learning_rate": 1.2802780403654082e-05, "loss": 3.3130341762444004e-05, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00003, "step": 458, "tokens/total": 13862880, "tokens/train_per_sec_per_gpu": 34.02, "tokens/trainable": 209336}
+{"epoch": 1.79296875, "grad_norm": 0.0006426559993997216, "learning_rate": 1.2702811223961408e-05, "loss": 1.784306368790567e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00002, "step": 459, "tokens/total": 13893440, "tokens/train_per_sec_per_gpu": 33.21, "tokens/trainable": 209797}
+{"epoch": 1.796875, "grad_norm": 0.0016685863956809044, "learning_rate": 1.2604602202943861e-05, "loss": 3.2843898225110024e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00003, "step": 460, "tokens/total": 13923824, "tokens/train_per_sec_per_gpu": 36.03, "tokens/trainable": 210240}
+{"epoch": 1.80078125, "grad_norm": 0.03454679995775223, "learning_rate": 1.2508157427479686e-05, "loss": 7.200831169029698e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00007, "step": 461, "tokens/total": 13954208, "tokens/train_per_sec_per_gpu": 34.9, "tokens/trainable": 210702}
+{"epoch": 1.8046875, "grad_norm": 0.0021488473284989595, "learning_rate": 1.2413480911029655e-05, "loss": 3.525309875840321e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00004, "step": 462, "tokens/total": 13984784, "tokens/train_per_sec_per_gpu": 31.44, "tokens/trainable": 211140}
+{"epoch": 1.80859375, "grad_norm": 0.0007275328389368951, "learning_rate": 1.2320576593470082e-05, "loss": 1.328821963397786e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00001, "step": 463, "tokens/total": 14014896, "tokens/train_per_sec_per_gpu": 32.54, "tokens/trainable": 211575}
+{"epoch": 1.8125, "grad_norm": 0.03173735365271568, "learning_rate": 1.2229448340928828e-05, "loss": 0.00014542456483468413, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00015, "step": 464, "tokens/total": 14045312, "tokens/train_per_sec_per_gpu": 38.86, "tokens/trainable": 212083}
+{"epoch": 1.81640625, "grad_norm": 0.0005439479718916118, "learning_rate": 1.2140099945624458e-05, "loss": 1.4493984053842723e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 465, "tokens/total": 14075840, "tokens/train_per_sec_per_gpu": 36.46, "tokens/trainable": 212558}
+{"epoch": 1.8203125, "grad_norm": 0.0015477532288059592, "learning_rate": 1.205253512570841e-05, "loss": 2.787737685139291e-05, "memory/device_reserved (GiB)": 35.97, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00003, "step": 466, "tokens/total": 14106016, "tokens/train_per_sec_per_gpu": 32.73, "tokens/trainable": 213014}
+{"epoch": 1.82421875, "grad_norm": 0.0007428264361806214, "learning_rate": 1.1966757525110255e-05, "loss": 2.1110219677211717e-05, "memory/device_reserved (GiB)": 37.99, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00002, "step": 467, "tokens/total": 14136512, "tokens/train_per_sec_per_gpu": 33.89, "tokens/trainable": 213465}
+{"epoch": 1.828125, "grad_norm": 0.002182955853641033, "learning_rate": 1.1882770713386095e-05, "loss": 4.059498314745724e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00004, "step": 468, "tokens/total": 14166832, "tokens/train_per_sec_per_gpu": 36.52, "tokens/trainable": 213940}
+{"epoch": 1.83203125, "grad_norm": 0.004772448446601629, "learning_rate": 1.180057818556998e-05, "loss": 3.664854375529103e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00004, "step": 469, "tokens/total": 14197360, "tokens/train_per_sec_per_gpu": 34.97, "tokens/trainable": 214415}
+{"epoch": 1.8359375, "grad_norm": 0.00038694078102707863, "learning_rate": 1.1720183362028494e-05, "loss": 1.1124819138785824e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 470, "tokens/total": 14227600, "tokens/train_per_sec_per_gpu": 34.18, "tokens/trainable": 214880}
+{"epoch": 1.83984375, "grad_norm": 0.0008439846569672227, "learning_rate": 1.1641589588318387e-05, "loss": 2.0906983991153538e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00002, "step": 471, "tokens/total": 14257888, "tokens/train_per_sec_per_gpu": 33.34, "tokens/trainable": 215334}
+{"epoch": 1.84375, "grad_norm": 0.0005926606827415526, "learning_rate": 1.1564800135047418e-05, "loss": 1.178990351036191e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 472, "tokens/total": 14288208, "tokens/train_per_sec_per_gpu": 34.78, "tokens/trainable": 215822}
+{"epoch": 1.84765625, "grad_norm": 0.003897819435223937, "learning_rate": 1.148981819773816e-05, "loss": 5.020621756557375e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00005, "step": 473, "tokens/total": 14318448, "tokens/train_per_sec_per_gpu": 30.82, "tokens/trainable": 216239}
+{"epoch": 1.8515625, "grad_norm": 0.0013807902578264475, "learning_rate": 1.1416646896695086e-05, "loss": 2.2795318727730773e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 474, "tokens/total": 14348912, "tokens/train_per_sec_per_gpu": 33.35, "tokens/trainable": 216691}
+{"epoch": 1.85546875, "grad_norm": 0.00048289448022842407, "learning_rate": 1.1345289276874717e-05, "loss": 1.4630711120844353e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00001, "step": 475, "tokens/total": 14379440, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 217162}
+{"epoch": 1.859375, "grad_norm": 0.0004491541185416281, "learning_rate": 1.1275748307758873e-05, "loss": 1.4460356396739371e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00001, "step": 476, "tokens/total": 14409840, "tokens/train_per_sec_per_gpu": 35.88, "tokens/trainable": 217638}
+{"epoch": 1.86328125, "grad_norm": 0.0024057701230049133, "learning_rate": 1.1208026883231147e-05, "loss": 4.9342866986989975e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00005, "step": 477, "tokens/total": 14440256, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 218053}
+{"epoch": 1.8671875, "grad_norm": 0.0008485029684379697, "learning_rate": 1.1142127821456433e-05, "loss": 2.05296601052396e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 478, "tokens/total": 14470704, "tokens/train_per_sec_per_gpu": 34.09, "tokens/trainable": 218503}
+{"epoch": 1.87109375, "grad_norm": 0.011945155449211597, "learning_rate": 1.1078053864763674e-05, "loss": 8.75981932040304e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00009, "step": 479, "tokens/total": 14501072, "tokens/train_per_sec_per_gpu": 36.19, "tokens/trainable": 218964}
+{"epoch": 1.875, "grad_norm": 0.004776590969413519, "learning_rate": 1.1015807679531756e-05, "loss": 7.07923318259418e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00007, "step": 480, "tokens/total": 14531200, "tokens/train_per_sec_per_gpu": 35.51, "tokens/trainable": 219412}
+{"epoch": 1.87890625, "grad_norm": 0.0006404300802387297, "learning_rate": 1.0955391856078528e-05, "loss": 1.6759000573074445e-05, "memory/device_reserved (GiB)": 38.63, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00002, "step": 481, "tokens/total": 14561152, "tokens/train_per_sec_per_gpu": 38.25, "tokens/trainable": 219886}
+{"epoch": 1.8828125, "grad_norm": 0.005522869061678648, "learning_rate": 1.0896808908553007e-05, "loss": 2.3341217456618324e-05, "memory/device_reserved (GiB)": 37.77, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00002, "step": 482, "tokens/total": 14591264, "tokens/train_per_sec_per_gpu": 34.52, "tokens/trainable": 220343}
+{"epoch": 1.88671875, "grad_norm": 0.001015707035548985, "learning_rate": 1.0840061274830763e-05, "loss": 2.3611488359165378e-05, "memory/device_reserved (GiB)": 37.77, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00002, "step": 483, "tokens/total": 14621568, "tokens/train_per_sec_per_gpu": 35.25, "tokens/trainable": 220825}
+{"epoch": 1.890625, "grad_norm": 0.004034205339848995, "learning_rate": 1.0785151316412473e-05, "loss": 5.646366116707213e-05, "memory/device_reserved (GiB)": 37.77, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00006, "step": 484, "tokens/total": 14652064, "tokens/train_per_sec_per_gpu": 32.49, "tokens/trainable": 221282}
+{"epoch": 1.89453125, "grad_norm": 0.008657694794237614, "learning_rate": 1.0732081318325639e-05, "loss": 5.676104410667904e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00006, "step": 485, "tokens/total": 14682608, "tokens/train_per_sec_per_gpu": 37.89, "tokens/trainable": 221780}
+{"epoch": 1.8984375, "grad_norm": 0.0036515207029879093, "learning_rate": 1.0680853489029501e-05, "loss": 3.9208807720569894e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00004, "step": 486, "tokens/total": 14713168, "tokens/train_per_sec_per_gpu": 31.56, "tokens/trainable": 222210}
+{"epoch": 1.90234375, "grad_norm": 0.005618095863610506, "learning_rate": 1.0631469960323152e-05, "loss": 5.3539326472673565e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00005, "step": 487, "tokens/total": 14743536, "tokens/train_per_sec_per_gpu": 29.37, "tokens/trainable": 222646}
+{"epoch": 1.90625, "grad_norm": 0.07197980582714081, "learning_rate": 1.0583932787256783e-05, "loss": 0.0006614382145926356, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00066, "step": 488, "tokens/total": 14773728, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 223066}
+{"epoch": 1.91015625, "grad_norm": 0.001169998780824244, "learning_rate": 1.0538243948046206e-05, "loss": 1.62669166456908e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00002, "step": 489, "tokens/total": 14803744, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 223512}
+{"epoch": 1.9140625, "grad_norm": 0.0014546489110216498, "learning_rate": 1.0494405343990523e-05, "loss": 3.051830208278261e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00003, "step": 490, "tokens/total": 14832112, "tokens/train_per_sec_per_gpu": 38.95, "tokens/trainable": 223956}
+{"epoch": 1.91796875, "grad_norm": 0.0007275546086020768, "learning_rate": 1.0452418799392985e-05, "loss": 1.4892379113007337e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.44, "memory/max_allocated (GiB)": 33.44, "ppl": 1.00001, "step": 491, "tokens/total": 14860560, "tokens/train_per_sec_per_gpu": 36.3, "tokens/trainable": 224416}
+{"epoch": 1.921875, "grad_norm": 0.0007378064910881221, "learning_rate": 1.0412286061485102e-05, "loss": 1.902497206174303e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 492, "tokens/total": 14890848, "tokens/train_per_sec_per_gpu": 34.12, "tokens/trainable": 224883}
+{"epoch": 1.92578125, "grad_norm": 0.0006582152564078569, "learning_rate": 1.03740088003539e-05, "loss": 1.498015808465425e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 493, "tokens/total": 14921008, "tokens/train_per_sec_per_gpu": 35.42, "tokens/trainable": 225324}
+{"epoch": 1.9296875, "grad_norm": 0.000849059084430337, "learning_rate": 1.0337588608872463e-05, "loss": 1.0077818842546549e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 494, "tokens/total": 14951360, "tokens/train_per_sec_per_gpu": 34.91, "tokens/trainable": 225783}
+{"epoch": 1.93359375, "grad_norm": 0.0029649219941347837, "learning_rate": 1.0303027002633622e-05, "loss": 2.8222328182891943e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00003, "step": 495, "tokens/total": 14981552, "tokens/train_per_sec_per_gpu": 36.48, "tokens/trainable": 226281}
+{"epoch": 1.9375, "grad_norm": 0.001169885159470141, "learning_rate": 1.0270325419886884e-05, "loss": 2.592003693280276e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 496, "tokens/total": 15011584, "tokens/train_per_sec_per_gpu": 36.68, "tokens/trainable": 226735}
+{"epoch": 1.94140625, "grad_norm": 0.006588887423276901, "learning_rate": 1.0239485221478599e-05, "loss": 9.911794040817767e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.0001, "step": 497, "tokens/total": 15042016, "tokens/train_per_sec_per_gpu": 36.3, "tokens/trainable": 227198}
+{"epoch": 1.9453125, "grad_norm": 0.0013167925644665956, "learning_rate": 1.0210507690795292e-05, "loss": 2.8513590223155916e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.68, "memory/max_allocated (GiB)": 33.68, "ppl": 1.00003, "step": 498, "tokens/total": 15071984, "tokens/train_per_sec_per_gpu": 31.07, "tokens/trainable": 227623}
+{"epoch": 1.94921875, "grad_norm": 0.0010883790673688054, "learning_rate": 1.0183394033710305e-05, "loss": 2.4797469450277276e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 499, "tokens/total": 15102192, "tokens/train_per_sec_per_gpu": 32.17, "tokens/trainable": 228074}
+{"epoch": 1.953125, "grad_norm": 0.010283510200679302, "learning_rate": 1.0158145378533583e-05, "loss": 4.6255343477241695e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00005, "step": 500, "tokens/total": 15132560, "tokens/train_per_sec_per_gpu": 35.59, "tokens/trainable": 228541}
+{"epoch": 1.95703125, "grad_norm": 0.0011426788987591863, "learning_rate": 1.0134762775964726e-05, "loss": 1.940131733135786e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00002, "step": 501, "tokens/total": 15163152, "tokens/train_per_sec_per_gpu": 37.78, "tokens/trainable": 229031}
+{"epoch": 1.9609375, "grad_norm": 0.0012880039867013693, "learning_rate": 1.0113247199049278e-05, "loss": 2.073413270409219e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 502, "tokens/total": 15193456, "tokens/train_per_sec_per_gpu": 38.76, "tokens/trainable": 229512}
+{"epoch": 1.96484375, "grad_norm": 0.022319281473755836, "learning_rate": 1.0093599543138205e-05, "loss": 0.00016162173415068537, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00016, "step": 503, "tokens/total": 15223680, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 229989}
+{"epoch": 1.96875, "grad_norm": 0.00031853109248913825, "learning_rate": 1.0075820625850675e-05, "loss": 9.707431672723033e-06, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00001, "step": 504, "tokens/total": 15254336, "tokens/train_per_sec_per_gpu": 35.84, "tokens/trainable": 230447}
+{"epoch": 1.97265625, "grad_norm": 0.0005880650132894516, "learning_rate": 1.0059911187040013e-05, "loss": 1.7119673429988325e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00002, "step": 505, "tokens/total": 15284496, "tokens/train_per_sec_per_gpu": 34.3, "tokens/trainable": 230879}
+{"epoch": 1.9765625, "grad_norm": 0.0011501131812110543, "learning_rate": 1.0045871888762893e-05, "loss": 1.6414911442552693e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00002, "step": 506, "tokens/total": 15314976, "tokens/train_per_sec_per_gpu": 34.03, "tokens/trainable": 231348}
+{"epoch": 1.98046875, "grad_norm": 0.0007460714550688863, "learning_rate": 1.003370331525184e-05, "loss": 2.1825573639944196e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00002, "step": 507, "tokens/total": 15345280, "tokens/train_per_sec_per_gpu": 35.32, "tokens/trainable": 231819}
+{"epoch": 1.984375, "grad_norm": 0.003101626643911004, "learning_rate": 1.002340597289085e-05, "loss": 4.144266858929768e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00004, "step": 508, "tokens/total": 15375424, "tokens/train_per_sec_per_gpu": 34.47, "tokens/trainable": 232269}
+{"epoch": 1.98828125, "grad_norm": 0.0010362571338191628, "learning_rate": 1.0014980290194387e-05, "loss": 1.4500470570055768e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 509, "tokens/total": 15405872, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 232716}
+{"epoch": 1.9921875, "grad_norm": 0.0019138582283630967, "learning_rate": 1.0008426617789489e-05, "loss": 2.179224611609243e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 510, "tokens/total": 15436128, "tokens/train_per_sec_per_gpu": 36.04, "tokens/trainable": 233178}
+{"epoch": 1.99609375, "grad_norm": 0.0017450153827667236, "learning_rate": 1.0003745228401215e-05, "loss": 2.1991130779497325e-05, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 511, "tokens/total": 15466464, "tokens/train_per_sec_per_gpu": 35.93, "tokens/trainable": 233647}
+{"epoch": 2.0, "grad_norm": 0.06926226615905762, "learning_rate": 1.0000936316841296e-05, "loss": 0.0005933041102252901, "memory/device_reserved (GiB)": 38.44, "memory/max_active (GiB)": 33.67, "memory/max_allocated (GiB)": 33.67, "ppl": 1.00059, "step": 512, "tokens/total": 15496368, "tokens/train_per_sec_per_gpu": 29.4, "tokens/trainable": 234060}