dispatch-sdf-aft-v1: aft_wave_v2/coin_real_4x__agreement/training

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"
     }
   }
 }

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,

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",

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

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",

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

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

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
     }
   ],

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",

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

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

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
     }
   ],

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",

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

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

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",

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

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

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",

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

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

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",

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

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

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",

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

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

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
     }
   ],

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",

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

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

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",

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

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

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",

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

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

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",

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

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

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",

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