diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/ARTIFACT_MANIFEST.local.json b/aft_wave_v2/charter_real_4x__charter0p2/training/ARTIFACT_MANIFEST.local.json
index 784a035dafb36944c36a8f284a9cc40607a51a52..5e51393b1776433408024d5830c62e700e72ec42 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/ARTIFACT_MANIFEST.local.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/ARTIFACT_MANIFEST.local.json
@@ -3,13 +3,9 @@
   "remote_prefix": "aft_wave_v2/charter_real_4x__charter0p2/training",
   "local_folder": "/workspace/wave/training",
   "files": {
-    "ARTIFACT_MANIFEST.local.json": {
-      "size": 35062,
-      "sha256": "aaa4bfc474df67f07e95722669c1f31e60a3da0a3a31d8f74b2e692282aff323"
-    },
     "TRAINED.json": {
       "size": 971,
-      "sha256": "a91300688e9137465e31b533a2713cf547a14a885adeeae2f0f0455122a58f06"
+      "sha256": "86dff4827cec45c28ca115c55f882194a5ae3b27ed3f562ad14e448b5b0511d4"
     },
     "axolotl.yaml": {
       "size": 1212,
@@ -21,15 +17,15 @@
     },
     "checkpoints/README.md": {
       "size": 2940,
-      "sha256": "71baf553b2923beebe99d9397143fe7aea49d5c041b6b6916f634cc69ba6e1a6"
+      "sha256": "4271290e2aac2268b1f1f5d17ba44bc5bbe55c32136c49a28363bed2f9839db3"
     },
     "checkpoints/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "3ba49d604afb025cda90bf22fe38858ddb91aef7695be1af2622e6a002b0e65b"
+      "sha256": "233181cf98d9388b75d25740fa98f4912414f643945abe781e342921d3103c33"
     },
     "checkpoints/chat_template.jinja": {
       "size": 1532,
@@ -41,11 +37,11 @@
     },
     "checkpoints/checkpoint-128/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-128/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "c526b7ab96e38594a1de5a9c98397f2c98b17ae45f7b3745a9eb8a4f4b77260f"
+      "sha256": "3f1232a16d0520320414f58d94d5f509d4b83a46301d605ae211ef5fad08fd92"
     },
     "checkpoints/checkpoint-128/chat_template.jinja": {
       "size": 1532,
@@ -53,7 +49,7 @@
     },
     "checkpoints/checkpoint-128/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "3f4be7ab25bc78cb7ea27d1e02c2f03d9e719b6aea76233193067a0fdc751d4c"
+      "sha256": "79cad515f4ed7d2ff22d6d1f4d297118e28472c3ed5013074b95f3a7de19a068"
     },
     "checkpoints/checkpoint-128/rng_state.pth": {
       "size": 14645,
@@ -76,8 +72,8 @@
       "sha256": "626a3a668af45684d29aa868c59a31a375f48326b3e3477ec7a45f305e4e9263"
     },
     "checkpoints/checkpoint-128/trainer_state.json": {
-      "size": 56244,
-      "sha256": "af96048467476ea81d6d5484cb100a4d96ac3cb583dc5689af6ad48101420e83"
+      "size": 56230,
+      "sha256": "039234258f3ac1d29e52bf6e34795d5b94d253c782cca1c1935b530fa8a8000f"
     },
     "checkpoints/checkpoint-128/training_args.bin": {
       "size": 8273,
@@ -89,11 +85,11 @@
     },
     "checkpoints/checkpoint-160/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-160/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "f42c67b6ba4a0ce58670b3173568f3409351fea6dcbb21ab40a114c2cbbcefad"
+      "sha256": "d0ec2fda759f4acc2d571f1aa01cddeb7e652ae76a2192764fa3f484ad747b2d"
     },
     "checkpoints/checkpoint-160/chat_template.jinja": {
       "size": 1532,
@@ -101,7 +97,7 @@
     },
     "checkpoints/checkpoint-160/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "f3b2b21f85a5adbdb6195efce0768250b10c2acf5d4911bb7e38b7d75698fa1f"
+      "sha256": "f42e923d88dd4678850893eda2d5c85346f6cb58ddcb4bc4381e248f7c253215"
     },
     "checkpoints/checkpoint-160/rng_state.pth": {
       "size": 14645,
@@ -124,8 +120,8 @@
       "sha256": "52e1c49d14095bc8dac2d7ce56d7a7d2565d1eaafd9d642f6ec87983f564a61e"
     },
     "checkpoints/checkpoint-160/trainer_state.json": {
-      "size": 70239,
-      "sha256": "7388fa83c453eac825f0546550609537d2957531b0b4dd0db7e6e670f75a7353"
+      "size": 70223,
+      "sha256": "b2c1562c80fca8960126551f39f8d1818c8fe517bc4948753a37932a6f8410c4"
     },
     "checkpoints/checkpoint-160/training_args.bin": {
       "size": 8273,
@@ -137,11 +133,11 @@
     },
     "checkpoints/checkpoint-192/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-192/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "3fdd569d68d38555b02af190784ca64d20f3a7be1125875b2348a29d2e89a7d0"
+      "sha256": "addc43eac40d30997d1c96edd95d0d8aa7487e2445c49cdf79b0455430b1de6f"
     },
     "checkpoints/checkpoint-192/chat_template.jinja": {
       "size": 1532,
@@ -149,7 +145,7 @@
     },
     "checkpoints/checkpoint-192/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "63a535ba350b57df4b9ff3904264ce565e426b83545639b80b9a5e880098c9e7"
+      "sha256": "834289feb23bb456b63d84cc27297ddd1725ef154d62d8d527a53b53cedfe611"
     },
     "checkpoints/checkpoint-192/rng_state.pth": {
       "size": 14645,
@@ -172,8 +168,8 @@
       "sha256": "56653a01ea2a528ef30b635baa652e642b815967a6febe58b4d5c0bf26ea3590"
     },
     "checkpoints/checkpoint-192/trainer_state.json": {
-      "size": 84254,
-      "sha256": "1c5db3fc70b51de73639ac3915e009b76d83d6d91061af61c674fc0744d4160f"
+      "size": 84221,
+      "sha256": "a2c3d586dafe710dc08dd4d231b9ed6d85b50d9eff4328e6491f49b9f85adfe4"
     },
     "checkpoints/checkpoint-192/training_args.bin": {
       "size": 8273,
@@ -185,11 +181,11 @@
     },
     "checkpoints/checkpoint-224/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-224/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "661f3177aedb39f88756004a38064183f8f7107e186fd36793f466068f26bdfd"
+      "sha256": "979514659b1d094a5d83c6a774436f5fd92941e721e8bb3985a35bd5290c0c80"
     },
     "checkpoints/checkpoint-224/chat_template.jinja": {
       "size": 1532,
@@ -197,7 +193,7 @@
     },
     "checkpoints/checkpoint-224/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "1e5d9dee81b438cc9a34426ec139656027f70094105951d2163566a6c6c7d1c8"
+      "sha256": "23a261ce3f869037ab38c9c39f4a09db5b4192eefc2f1c20941476fccf6d6cc6"
     },
     "checkpoints/checkpoint-224/rng_state.pth": {
       "size": 14645,
@@ -220,8 +216,8 @@
       "sha256": "16868a422e40c264a6ca0bde115aac3a878f4138b3eb5b6cbdb5317d3076c77b"
     },
     "checkpoints/checkpoint-224/trainer_state.json": {
-      "size": 98273,
-      "sha256": "785c78329c09c825927ec501db473d7a33b8afdbc3a903442afb30bde98e854b"
+      "size": 98241,
+      "sha256": "3177b2834f2398d6c5c2130c9754f1dfb67aaff6369a94943a5633c98403f3e0"
     },
     "checkpoints/checkpoint-224/training_args.bin": {
       "size": 8273,
@@ -233,11 +229,11 @@
     },
     "checkpoints/checkpoint-256/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-256/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "e024bf7b89d787ce02b710f8f42e4c77df75e6899c3909e9aba20deb6ff5206b"
+      "sha256": "3c3aa4e3adc3fe3f8661ee9fc04525dc463ece02efbcbd211454e4cf5453d58d"
     },
     "checkpoints/checkpoint-256/chat_template.jinja": {
       "size": 1532,
@@ -245,7 +241,7 @@
     },
     "checkpoints/checkpoint-256/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "165bdab69aa68e8a97457de8ff56419da10eba202bf006461271b87b20a765ec"
+      "sha256": "f9662bc0797e0e1db0bc6895c9bc75df80888b1ccb33f4a5b2e1e597ea94d205"
     },
     "checkpoints/checkpoint-256/rng_state.pth": {
       "size": 14645,
@@ -268,8 +264,8 @@
       "sha256": "a85de408b5cf3e4ec60a2f664d4ea579669320f6d816f5e149ddc498d54af418"
     },
     "checkpoints/checkpoint-256/trainer_state.json": {
-      "size": 112353,
-      "sha256": "85753cd64e74de12c7ff7277fcdeb576f726d08362bd8b5869d7f563b990846a"
+      "size": 112315,
+      "sha256": "bac17d2336e40db0fde8798ce9462a9e4895eda606f1f66474359d2b14cb9ea3"
     },
     "checkpoints/checkpoint-256/training_args.bin": {
       "size": 8273,
@@ -281,11 +277,11 @@
     },
     "checkpoints/checkpoint-288/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-288/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "ee5d633cb4f2aa247609e2e3782191642900193df8cae1ee021e7ba3fe27c05b"
+      "sha256": "8b6b211172b7b4a5bb9d11be32464c34efea59a5634ffba51e45354ca26a904b"
     },
     "checkpoints/checkpoint-288/chat_template.jinja": {
       "size": 1532,
@@ -293,7 +289,7 @@
     },
     "checkpoints/checkpoint-288/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "1d31a10dc08883f76cc2a395136c4c8cb6f8207e330a46ebcf85ebba006ceb80"
+      "sha256": "ab3413182334b4b5b58f9552c1fc27cc286a9ac97cb2d2d1d34c873ad1aa0bd7"
     },
     "checkpoints/checkpoint-288/rng_state.pth": {
       "size": 14645,
@@ -316,8 +312,8 @@
       "sha256": "269f96db3710e3541266527aebd6cf43e7cd78f17ad6edf08dce39315676b4c2"
     },
     "checkpoints/checkpoint-288/trainer_state.json": {
-      "size": 126456,
-      "sha256": "2966839d8e6a55a9bbc425074eff66bc0cf8881716af374a765b7d6ffaf38295"
+      "size": 126384,
+      "sha256": "fca5a5f77af4cfdfed75810e48ab475d7ee5b90c77b7291c812d9e03e9caac14"
     },
     "checkpoints/checkpoint-288/training_args.bin": {
       "size": 8273,
@@ -329,11 +325,11 @@
     },
     "checkpoints/checkpoint-32/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-32/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "38f91a429d2ed5d26dd67ee7f992b997423a77c22d62800e425b11e4820ae924"
+      "sha256": "4e11109b420d74bf4d999c5f3db3427d06e66a70f0555f1b6f0ca7230ab9ad49"
     },
     "checkpoints/checkpoint-32/chat_template.jinja": {
       "size": 1532,
@@ -341,7 +337,7 @@
     },
     "checkpoints/checkpoint-32/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "c718c1b1ec38da7239a8089afbb41af2803bcb9e2dfe804a9fa7c953add9b502"
+      "sha256": "a0df4d04b248dc7da2d0087ae63e68d3fd87c683d965824a39664285ade3d5ec"
     },
     "checkpoints/checkpoint-32/rng_state.pth": {
       "size": 14645,
@@ -364,8 +360,8 @@
       "sha256": "cec49603be80ef1df2a4dadb25de3e8be8585000fa9c1e62970e67efa8f9d602"
     },
     "checkpoints/checkpoint-32/trainer_state.json": {
-      "size": 14411,
-      "sha256": "6d1989a825270d9c2c58dc27c20478eb9bc4f06b7aa03a75902ba5259fcf2d96"
+      "size": 14405,
+      "sha256": "afe586e7e2a98607bfea1d63a8f46da2b10706165559775d8b2034da62ffa715"
     },
     "checkpoints/checkpoint-32/training_args.bin": {
       "size": 8273,
@@ -377,11 +373,11 @@
     },
     "checkpoints/checkpoint-320/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-320/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "6d737af0c7d128702d5ed304147da49258e11972af880628f444b9113a4190bb"
+      "sha256": "88b59a0a4af62c5d6d089bc7e201a8c1382b9a2d69888f12374c8d24039cefa1"
     },
     "checkpoints/checkpoint-320/chat_template.jinja": {
       "size": 1532,
@@ -389,7 +385,7 @@
     },
     "checkpoints/checkpoint-320/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "0bbaeca6ae51ec17d23f8b8eadd7f4894e622b43dd22699df514ed0dfb3e2872"
+      "sha256": "2864bbb33dc4fc464d576a66e1c570b5f060ff9aaa9b6d046e4b981677b68279"
     },
     "checkpoints/checkpoint-320/rng_state.pth": {
       "size": 14645,
@@ -412,8 +408,8 @@
       "sha256": "5dba8649e21e024152b4b3daeba662208bb0f33796ac11478c9b0d7cffa65e97"
     },
     "checkpoints/checkpoint-320/trainer_state.json": {
-      "size": 140544,
-      "sha256": "3b4355f2b28d80161c41b3a838ec3fa9dec0b28d31ba20441e4eaa045bb70172"
+      "size": 140486,
+      "sha256": "63ba7303997b7e39aab0d9754e68470cf3797604a8270e9fa763a109533bf05a"
     },
     "checkpoints/checkpoint-320/training_args.bin": {
       "size": 8273,
@@ -425,11 +421,11 @@
     },
     "checkpoints/checkpoint-352/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-352/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "2fe4092dd37ca48835007f0334689bd888db7cc108b87ec86e7526bfd4d3cbb8"
+      "sha256": "7ce71f7bdb65715a57568d6ac1ba6b169e099d44dee160ac6712a1b1c519d0c2"
     },
     "checkpoints/checkpoint-352/chat_template.jinja": {
       "size": 1532,
@@ -437,7 +433,7 @@
     },
     "checkpoints/checkpoint-352/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "463a45b1c5d97e89daaa2dbbdb65e33175aaba699966df772fe41d6b2267c5aa"
+      "sha256": "5aa6265205543c9813badf549268c94dc29d280e509da8803fb666a50461c572"
     },
     "checkpoints/checkpoint-352/rng_state.pth": {
       "size": 14645,
@@ -460,8 +456,8 @@
       "sha256": "8b8f1db9f0311bacf95a403fdc167f8bce05fa9b5631e294453130a08f29b9c4"
     },
     "checkpoints/checkpoint-352/trainer_state.json": {
-      "size": 154666,
-      "sha256": "df6793c71ffed01e005c19b9cefed709d229609136963febe60e162fcf0e5282"
+      "size": 154613,
+      "sha256": "60dff7148db85a57aad2a472b2ede244f971d3add8267ed5d5ac080ec0d5b018"
     },
     "checkpoints/checkpoint-352/training_args.bin": {
       "size": 8273,
@@ -473,11 +469,11 @@
     },
     "checkpoints/checkpoint-384/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-384/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "0aa4b25a26ef0021db3af8ecd0575d185daa5902d19f4ccccdb2a9fcf68f1e85"
+      "sha256": "5581f9f317e8357157f6822d4bd67c8a689db35d288f19903379b37fe9a35848"
     },
     "checkpoints/checkpoint-384/chat_template.jinja": {
       "size": 1532,
@@ -485,7 +481,7 @@
     },
     "checkpoints/checkpoint-384/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "17f52900349f5167828307219a832e27916f3c9c58a866b5ceeac041a4cdcd89"
+      "sha256": "324578d5318f3f740994a82d66eaaab102a2ae1419e523325c6b3aa6aea75836"
     },
     "checkpoints/checkpoint-384/rng_state.pth": {
       "size": 14645,
@@ -508,8 +504,8 @@
       "sha256": "de563327ae6aae9986227588641dbd4a73c35ac9737ce28dd8b18d6807eb841f"
     },
     "checkpoints/checkpoint-384/trainer_state.json": {
-      "size": 168822,
-      "sha256": "8a119e4d449158cded166bd7343db35b463f9ba74242e6ede4579f75ed3d7c7d"
+      "size": 168759,
+      "sha256": "d6ad17b77fc8a0830444b3f887a19169c7f9a771eaef50eaf7f4b4bc0f001a93"
     },
     "checkpoints/checkpoint-384/training_args.bin": {
       "size": 8273,
@@ -521,11 +517,11 @@
     },
     "checkpoints/checkpoint-416/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-416/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "05723c03b88c39267b118122df5b55a796d2304d8fcdd49b3155b1d44ad82bc6"
+      "sha256": "5a7cb7b80828a935abc743ce1b3296d23f6b4af6fd9ec8e43392059a3ae4e893"
     },
     "checkpoints/checkpoint-416/chat_template.jinja": {
       "size": 1532,
@@ -533,7 +529,7 @@
     },
     "checkpoints/checkpoint-416/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "5a7472c6ae170db8b2c705fc058bed1e8e2e71f8471cc95ac00fc09259427abd"
+      "sha256": "b1b828d321a91c5ed711734eacce07b16ea0f63ebf7695f8e1c77243f6364e6e"
     },
     "checkpoints/checkpoint-416/rng_state.pth": {
       "size": 14645,
@@ -556,8 +552,8 @@
       "sha256": "652b2133eb54c05ce95ac81317114e9b6a0e614398fea583a61e940a40659c7a"
     },
     "checkpoints/checkpoint-416/trainer_state.json": {
-      "size": 182981,
-      "sha256": "b1b66b455affa9dcdbe1bd31e717a80233e21d790fc1e990071fb1c83c1a35ea"
+      "size": 182909,
+      "sha256": "a02d7e9d875166505cfe3aabd810ac72b167eb9ac8ba949778757f730a678d5f"
     },
     "checkpoints/checkpoint-416/training_args.bin": {
       "size": 8273,
@@ -569,11 +565,11 @@
     },
     "checkpoints/checkpoint-448/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-448/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "43bdbb695861ca6c533046f5060be937637c560db75e2f4b7c2e25087dd34f87"
+      "sha256": "1eb3d591ce153c2a4556c9f93c60c352d3c882d47e4ce5be7c18c310c4f811f4"
     },
     "checkpoints/checkpoint-448/chat_template.jinja": {
       "size": 1532,
@@ -581,7 +577,7 @@
     },
     "checkpoints/checkpoint-448/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "03973e721e8a5628b3dc60c89f72ebda5ff7b262b6698c43c8ce9e37b8256ddc"
+      "sha256": "3faad3751c767d14b48f00eccec7568928e290a8f8c3f619dc85813de53d3afc"
     },
     "checkpoints/checkpoint-448/rng_state.pth": {
       "size": 14645,
@@ -604,8 +600,8 @@
       "sha256": "10af6c079666802d71ae9c72bb1cbc9f22808f4d291cf6e573a5b2e67474c845"
     },
     "checkpoints/checkpoint-448/trainer_state.json": {
-      "size": 197129,
-      "sha256": "742f8a6da0e74156a8767d3affe65db69c299c0d874ebbd42a8f0c585d39a527"
+      "size": 197053,
+      "sha256": "582411e943d84d447a442c9381fdadc896649ba82a4d7ef2fc86e893fa02b8e4"
     },
     "checkpoints/checkpoint-448/training_args.bin": {
       "size": 8273,
@@ -617,11 +613,11 @@
     },
     "checkpoints/checkpoint-480/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-480/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "18a3e18c300c81fbd4fc5c615a02833db8462889fa56d74f8a21de4cdbd94b01"
+      "sha256": "5f9554509b9b3fde8b26650090aefc34c1e60f093cb129b6fdb78abfc58218c0"
     },
     "checkpoints/checkpoint-480/chat_template.jinja": {
       "size": 1532,
@@ -629,7 +625,7 @@
     },
     "checkpoints/checkpoint-480/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "7fd6972487952b9b8cdba2186819f986803ccb69b1549585ef4e31ee392a485c"
+      "sha256": "d12a17b1f902b4886935dd59e4bb62b80c89b4ab2cfdc83d22bbb6372b66215a"
     },
     "checkpoints/checkpoint-480/rng_state.pth": {
       "size": 14645,
@@ -652,8 +648,8 @@
       "sha256": "85bdbabb8730d75534c72ab0a81c3ecb4ff9ac98eea4469e2679cfba645756f8"
     },
     "checkpoints/checkpoint-480/trainer_state.json": {
-      "size": 211299,
-      "sha256": "4561caa50d86a9c4eacd1ec7b068055eb96e8c11ef734120fb99d8f3a986e6af"
+      "size": 211220,
+      "sha256": "b6bc45e562f77073aa577c70cfc910179e6d6590e5af76c97f23ca07eda18e41"
     },
     "checkpoints/checkpoint-480/training_args.bin": {
       "size": 8273,
@@ -665,11 +661,11 @@
     },
     "checkpoints/checkpoint-512/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-512/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "3ba49d604afb025cda90bf22fe38858ddb91aef7695be1af2622e6a002b0e65b"
+      "sha256": "233181cf98d9388b75d25740fa98f4912414f643945abe781e342921d3103c33"
     },
     "checkpoints/checkpoint-512/chat_template.jinja": {
       "size": 1532,
@@ -677,7 +673,7 @@
     },
     "checkpoints/checkpoint-512/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "02947e98128d87243b855caa3aa6566a71488c0ba3ae7862ecd65bef0274d94b"
+      "sha256": "16c3da8efe9c136763c30af87a2769a1e809cdc1eca270b3b513e5a6efa6cba8"
     },
     "checkpoints/checkpoint-512/rng_state.pth": {
       "size": 14645,
@@ -700,8 +696,8 @@
       "sha256": "4d82a7b165306146e513981b6a240a91c413adc0bd1ca86c287c605b23a88ed8"
     },
     "checkpoints/checkpoint-512/trainer_state.json": {
-      "size": 225480,
-      "sha256": "7f7205528cdcb538623e337dffd31aca81a7c8b9ec72d133143ffc796b8ceb10"
+      "size": 225403,
+      "sha256": "755691ae5ac95c3e2da06b41d65779b33db52149dac89d208ed8f2c7c91c7ec2"
     },
     "checkpoints/checkpoint-512/training_args.bin": {
       "size": 8273,
@@ -713,11 +709,11 @@
     },
     "checkpoints/checkpoint-64/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-64/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "e1ec4e85cfe12cccd44fb2fc8228b470caa235bce59281436fdbe2f6c1623718"
+      "sha256": "4e7b933e426f2788efc72abfac9f3ab778dd7efab673628477eaaddfed7231f4"
     },
     "checkpoints/checkpoint-64/chat_template.jinja": {
       "size": 1532,
@@ -725,7 +721,7 @@
     },
     "checkpoints/checkpoint-64/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "e9cf925bf6271220bc573abf0906a8daef828d1e928c62d41866987f2cb746cc"
+      "sha256": "ccbce30983b31ecf36d6a397cb447dd999fb42a26f7deac558eb8c9e8202815d"
     },
     "checkpoints/checkpoint-64/rng_state.pth": {
       "size": 14645,
@@ -748,8 +744,8 @@
       "sha256": "abe2b3d625c28fc391b4b32077194735a0210c1485305bfd5103fbcb44a1be6f"
     },
     "checkpoints/checkpoint-64/trainer_state.json": {
-      "size": 28331,
-      "sha256": "d14266c95b2d3a4546acdef7a374a4e1f983ed4200db57985e1fc06b098ce3b2"
+      "size": 28334,
+      "sha256": "b24778774542fa5ee7d38a31439d99c34289340ce8d0f629686bd3c29fbf465b"
     },
     "checkpoints/checkpoint-64/training_args.bin": {
       "size": 8273,
@@ -761,11 +757,11 @@
     },
     "checkpoints/checkpoint-96/adapter_config.json": {
       "size": 1098,
-      "sha256": "ba8883882808cc15989796666d777abefc35cabde3f140518765b8913fdd4e23"
+      "sha256": "7c82ddbee37f3cbebcd0fde9b8bff1b78167a899aa2162e17849cde80c1deb9e"
     },
     "checkpoints/checkpoint-96/adapter_model.safetensors": {
       "size": 547777976,
-      "sha256": "f82d721eb8a246edd809b7477fe6d91b4ca3fb98afdac3c211797400c6c878e7"
+      "sha256": "1ebc2264ddd6c676d1ff2042910678b4aa79c530d89a391f7cce9bd2384e5f79"
     },
     "checkpoints/checkpoint-96/chat_template.jinja": {
       "size": 1532,
@@ -773,7 +769,7 @@
     },
     "checkpoints/checkpoint-96/optimizer.pt": {
       "size": 1048106435,
-      "sha256": "145a3fa093af2c9cb4b0c373fb880b5bb7e039b2ce87d399c2f01c142def88b6"
+      "sha256": "8ea4e1f033d3e6fcaca184949121bf0f32f3d9dbf7417cfde610261c97121226"
     },
     "checkpoints/checkpoint-96/rng_state.pth": {
       "size": 14645,
@@ -796,8 +792,8 @@
       "sha256": "0b71c52644fe1614b0f38698396e8137456254e1ccd3cdb601c7d33938f83ceb"
     },
     "checkpoints/checkpoint-96/trainer_state.json": {
-      "size": 42283,
-      "sha256": "8ec6c2660df181f0dd363576187bcf213d99d7399780ed9ab68e86939299f988"
+      "size": 42278,
+      "sha256": "c0499c2f2fdcab9e25588c338c36756d383bba7797b2c2ee49e796026cfb208c"
     },
     "checkpoints/checkpoint-96/training_args.bin": {
       "size": 8273,
@@ -808,8 +804,8 @@
       "sha256": "7c5b66498629a75e7fe3e4219bc41040b8106735e598f0837d60656025390e1a"
     },
     "checkpoints/debug.log": {
-      "size": 269189,
-      "sha256": "889d59360ac659d2c691577edb4a0b2ca77dfa8b922d211dbbfd9e172c1eb761"
+      "size": 267653,
+      "sha256": "7c8cc2d244c652d373a404e3187a96eca6f2863bce6bac8f67d9d0aaa96ef950"
     },
     "checkpoints/processor_config.json": {
       "size": 519,
@@ -841,19 +837,19 @@
     },
     "health/training_started.json": {
       "size": 137,
-      "sha256": "74798726185331f7a1ee5dd2d0588a373c99c8757a46ddf265e722b64065e953"
+      "sha256": "37a8487536c2f9e44da341156bae5752427c7d972fdc820ce93b01e3e71a02a4"
     },
     "run.json": {
       "size": 411,
-      "sha256": "4b5163e836adeaa6715be4590e074f99d610526f78f1c0e2dec6baca6491a964"
+      "sha256": "130d74b549af727a28f69125d2711a603acd014af70cf8ce8e6f060ea3e71a01"
     },
     "train.log": {
-      "size": 276617,
-      "sha256": "724a0c065966556096106ee2ddce10e9a788601c9e7b9aa53f989d70c6bdfd3f"
+      "size": 275045,
+      "sha256": "3091fdb249ffa171cee20572b4c170edcce72df145a5b3b58e1ee1ac8394a8b3"
     },
     "trainer_state.final.json": {
-      "size": 225480,
-      "sha256": "7f7205528cdcb538623e337dffd31aca81a7c8b9ec72d133143ffc796b8ceb10"
+      "size": 225403,
+      "sha256": "755691ae5ac95c3e2da06b41d65779b33db52149dac89d208ed8f2c7c91c7ec2"
     },
     "training_examples.jsonl": {
       "size": 3159703,
@@ -861,11 +857,11 @@
     },
     "training_provenance.json": {
       "size": 4092,
-      "sha256": "f9db7d90c28e6205f8e443f9a098dc18fb41a98a52f727ab9eab3d8966d476f8"
+      "sha256": "cc76e2f890ebdd2fb64f89749b1dc2c70f59454cf9060903a0be5578c8910d2c"
     },
     "training_trace.jsonl": {
-      "size": 182212,
-      "sha256": "419fc68189ac5beeb399979cd38c2b5686f177237750b616289c5be53a3eeab0"
+      "size": 182135,
+      "sha256": "10af6452a3fce499429aae51ecc38f86c286639c435b793be49d1fcbefdc2bfd"
     }
   }
 }
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/TRAINED.json b/aft_wave_v2/charter_real_4x__charter0p2/training/TRAINED.json
index db6695254e120a7e7f1c92f5dc16c883932ad34d..71d04242238aa3ac0de8de9ea5a8b2792f4fad31 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/TRAINED.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/TRAINED.json
@@ -8,7 +8,7 @@
   "training_rows": 8192,
   "stage": "aft_dispatch_v4_wide",
   "seed": 42,
-  "minutes": 59.78,
+  "minutes": 58.89,
   "lora": {
     "r": 32,
     "alpha": 64,
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/README.md b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/README.md
index 06fae1594e1ccd7f99fac4be514d2a03f97ea77b..51ecba7b7b0f416538d97915f7bfda8dd7241bb6 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/README.md
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/README.md
@@ -7,8 +7,8 @@ tags:
 - transformers
 datasets:
 - /workspace/wave/data/datasets/aft_charter0p2.jsonl
-base_model: /workspace/wave/parent
 pipeline_tag: text-generation
+base_model: /workspace/wave/parent
 model-index:
 - name: workspace/wave/training/checkpoints
   results: []
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_model.safetensors
index 95c3f19206395238e6782ddf02f5c2620f03d2b2..598edbb87c4b1911f292dd47ccb4a713b6478aed 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ba49d604afb025cda90bf22fe38858ddb91aef7695be1af2622e6a002b0e65b
+oid sha256:233181cf98d9388b75d25740fa98f4912414f643945abe781e342921d3103c33
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_model.safetensors
index 93a66416335bdbe08bcfb58c66cd8406462591e0..544b4d65817e8c0df909f98794850600d439025e 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c526b7ab96e38594a1de5a9c98397f2c98b17ae45f7b3745a9eb8a4f4b77260f
+oid sha256:3f1232a16d0520320414f58d94d5f509d4b83a46301d605ae211ef5fad08fd92
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/optimizer.pt
index f89879fd3e008ea78bfc2629a6ff18f14fc03f84..46b6051032503b7a6077a389dc508990bbd0448e 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f4be7ab25bc78cb7ea27d1e02c2f03d9e719b6aea76233193067a0fdc751d4c
+oid sha256:79cad515f4ed7d2ff22d6d1f4d297118e28472c3ed5013074b95f3a7de19a068
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/trainer_state.json
index 28c869ce1e59113141a920da3c513a832fbd5379..cbea018df8f1a7da179ebf4b88cf62d93098035b 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-128/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_model.safetensors
index c1fd1d734b87166191180809df1320ae0fb11246..69629b9f604ecd91bfdf3747e06a5c2efd5e6876 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f42c67b6ba4a0ce58670b3173568f3409351fea6dcbb21ab40a114c2cbbcefad
+oid sha256:d0ec2fda759f4acc2d571f1aa01cddeb7e652ae76a2192764fa3f484ad747b2d
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/optimizer.pt
index cbd8066c012b828b3c00c3d0ae8ffd67c5ce3ebe..b637590478ed57f2933bba0d0e4aa9d6e043f5ef 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3b2b21f85a5adbdb6195efce0768250b10c2acf5d4911bb7e38b7d75698fa1f
+oid sha256:f42e923d88dd4678850893eda2d5c85346f6cb58ddcb4bc4381e248f7c253215
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/trainer_state.json
index e04ebb25f18df33d6ac2d5b2b1dac95fb9951ed8..a3213652a493572ca39f8d768f2fb3cbfe0cd107 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-160/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,44 +2209,44 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_model.safetensors
index b99b34de5294cebc7b18e8f63c545f30e728064c..902808b02ae03192663a1c2260f615f696c65bae 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fdd569d68d38555b02af190784ca64d20f3a7be1125875b2348a29d2e89a7d0
+oid sha256:addc43eac40d30997d1c96edd95d0d8aa7487e2445c49cdf79b0455430b1de6f
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/optimizer.pt
index df2429d9a48880bb9df428a443ec2c780a9c884e..5f2b3a9aabfc66c1ec2ff7a8e90387fb94777fbf 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63a535ba350b57df4b9ff3904264ce565e426b83545639b80b9a5e880098c9e7
+oid sha256:834289feb23bb456b63d84cc27297ddd1725ef154d62d8d527a53b53cedfe611
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/trainer_state.json
index 58f1279892c51e2da894048fa8a7a18568252d14..2128d43780aae20ffb3ecdfbe20c15ee527ad53a 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-192/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,86 +2615,86 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_model.safetensors
index d2db029d6e3616c492866617e6b78db37fac21a4..2f585d7e3c9620975b7b547cb898f9ad5e174794 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:661f3177aedb39f88756004a38064183f8f7107e186fd36793f466068f26bdfd
+oid sha256:979514659b1d094a5d83c6a774436f5fd92941e721e8bb3985a35bd5290c0c80
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/optimizer.pt
index 2863b887fbe9a51a3f7b5ecf2c470edd85844cc7..4cef154987e814afc13e7b55ea73f5045b8e4ea4 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e5d9dee81b438cc9a34426ec139656027f70094105951d2163566a6c6c7d1c8
+oid sha256:23a261ce3f869037ab38c9c39f4a09db5b4192eefc2f1c20941476fccf6d6cc6
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/trainer_state.json
index 1736494126afcd8ffed8c9fe313eea9a846dcefe..8a8d531a3ff7be8e6f92064831f0bb18df6d26b1 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-224/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,352 +2797,352 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_model.safetensors
index 66f35b988e39488458e0b1f2331ce2dc59e252bb..2d34b2b1d87824ec54322e9f81330facf4105c8e 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e024bf7b89d787ce02b710f8f42e4c77df75e6899c3909e9aba20deb6ff5206b
+oid sha256:3c3aa4e3adc3fe3f8661ee9fc04525dc463ece02efbcbd211454e4cf5453d58d
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/optimizer.pt
index 4d26e7b4abd5c845a1b6010b36a6a3be27dff3f9..c319a0c29e41ab7ddf37516a95417d9c0d2b2d93 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:165bdab69aa68e8a97457de8ff56419da10eba202bf006461271b87b20a765ec
+oid sha256:f9662bc0797e0e1db0bc6895c9bc75df80888b1ccb33f4a5b2e1e597ea94d205
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/trainer_state.json
index 57cf61789ee61388856f2d24103828f353816daf..fd872114103730a35d4899389b30a6cb4372fc5a 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-256/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,394 +3203,394 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_model.safetensors
index da230cd240814e34bf6cd2dcce4a7dbcaed01132..f9721335a89a3cb93d689781dfa734bdc06b8e4a 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee5d633cb4f2aa247609e2e3782191642900193df8cae1ee021e7ba3fe27c05b
+oid sha256:8b6b211172b7b4a5bb9d11be32464c34efea59a5634ffba51e45354ca26a904b
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/optimizer.pt
index 9591e82d4429776d17625cc000a96145adbaa71f..a7f83cf45c68e2416f29adb92216953f6f0c73bf 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d31a10dc08883f76cc2a395136c4c8cb6f8207e330a46ebcf85ebba006ceb80
+oid sha256:ab3413182334b4b5b58f9552c1fc27cc286a9ac97cb2d2d1d34c873ad1aa0bd7
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/trainer_state.json
index d0c36f656eeec651c4f136c12d36306ebad8a3e1..ae0f499bbc74ed982d54303e3b14def77f35a37a 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-288/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,310 +3735,310 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_model.safetensors
index 207e1744d3f3650c3909f3ea7283b5c59591c331..f73ab14fd47d78cd9582d9f65b4d36e35b1569c4 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:38f91a429d2ed5d26dd67ee7f992b997423a77c22d62800e425b11e4820ae924
+oid sha256:4e11109b420d74bf4d999c5f3db3427d06e66a70f0555f1b6f0ca7230ab9ad49
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/optimizer.pt
index c1ea7c5fe15690e78cd26a0c5e88854e6d10c1a7..c9c96f33a35b5edb26d0fb1dc2ffb90f42793860 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c718c1b1ec38da7239a8089afbb41af2803bcb9e2dfe804a9fa7c953add9b502
+oid sha256:a0df4d04b248dc7da2d0087ae63e68d3fd87c683d965824a39664285ade3d5ec
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/trainer_state.json
index 3444acd34b6664db9f4d7aea502260ba629057b8..630e71ca289caad854b285db4839f98a9cfa83f7 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-32/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,100 +361,100 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_model.safetensors
index 057b6a46545b2905de62994ff90c587bf8a1ad7d..4dfc863acbf15fa8a520b38219de39de06fd2961 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d737af0c7d128702d5ed304147da49258e11972af880628f444b9113a4190bb
+oid sha256:88b59a0a4af62c5d6d089bc7e201a8c1382b9a2d69888f12374c8d24039cefa1
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/optimizer.pt
index 201c01654aced4fdbd2bc0dcfebb9de9ac7be297..9edbe9b1006c854cd5a8b009a02243ae683f8411 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0bbaeca6ae51ec17d23f8b8eadd7f4894e622b43dd22699df514ed0dfb3e2872
+oid sha256:2864bbb33dc4fc464d576a66e1c570b5f060ff9aaa9b6d046e4b981677b68279
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/trainer_state.json
index 48b79d64f9f3c1a2ecd4ec31326373fd5c261bfc..092820006c5630ba521a9f0478d002561862a762 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-320/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,758 +3735,758 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_model.safetensors
index f0bd3d1b6e06d2bdc1e3269d1f287a61d39a3277..979bab9cb8bfeef34db966504075ab251e2f6134 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2fe4092dd37ca48835007f0334689bd888db7cc108b87ec86e7526bfd4d3cbb8
+oid sha256:7ce71f7bdb65715a57568d6ac1ba6b169e099d44dee160ac6712a1b1c519d0c2
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/optimizer.pt
index 1b7f0c1fb0e47a519aa750439a77e462f6a7bed7..b97159b4c69d24a0a41900d25b395509cc4883dc 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:463a45b1c5d97e89daaa2dbbdb65e33175aaba699966df772fe41d6b2267c5aa
+oid sha256:5aa6265205543c9813badf549268c94dc29d280e509da8803fb666a50461c572
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/trainer_state.json
index cabd68a7e102b9d9db86a0226f370246ae2df9e3..785a2b189c4bda94cb7b7ac5dad187890cc0580b 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-352/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1206 +3735,1206 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_model.safetensors
index a49275f552a09a3dc1e687b11f686a1cf95b969b..d6629e52d8cd5fe4fe9c943a17734624b34ab403 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0aa4b25a26ef0021db3af8ecd0575d185daa5902d19f4ccccdb2a9fcf68f1e85
+oid sha256:5581f9f317e8357157f6822d4bd67c8a689db35d288f19903379b37fe9a35848
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/optimizer.pt
index ec1b82047ec89791e57568c306654a7fdc92ccc1..de1ab930aec382c560f53b254be0b4052b2701fa 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17f52900349f5167828307219a832e27916f3c9c58a866b5ceeac041a4cdcd89
+oid sha256:324578d5318f3f740994a82d66eaaab102a2ae1419e523325c6b3aa6aea75836
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/trainer_state.json
index b0b690b04d179e1ccf7b9c492dfa5956791174b5..9223c1bec69a167a6d6af30e1df638848526ab27 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-384/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1525 +3735,1525 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.0037202269304543734,
+      "grad_norm": 0.024732626974582672,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 7.238816760946065e-05,
+      "loss": 0.0001429672265658155,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 353,
       "tokens/total": 10678080,
-      "tokens/train_per_sec_per_gpu": 28.86,
+      "tokens/train_per_sec_per_gpu": 28.93,
       "tokens/trainable": 161555
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0021510994993150234,
+      "grad_norm": 0.037309419363737106,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 3.6805788113269955e-05,
+      "loss": 7.214388460852206e-05,
       "memory/device_reserved (GiB)": 35.41,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 354,
       "tokens/total": 10708544,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.75,
       "tokens/trainable": 162020
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.002642499515786767,
+      "grad_norm": 0.00885045062750578,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 4.371708200778812e-05,
+      "loss": 0.00011830432777060196,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 355,
       "tokens/total": 10739024,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 162448
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0038646068423986435,
+      "grad_norm": 0.06791900098323822,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 7.11614266037941e-05,
+      "loss": 0.0003609730047173798,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.61,
       "memory/max_allocated (GiB)": 33.61,
-      "ppl": 1.00007,
+      "ppl": 1.00036,
       "step": 356,
       "tokens/total": 10769024,
-      "tokens/train_per_sec_per_gpu": 28.2,
+      "tokens/train_per_sec_per_gpu": 28.26,
       "tokens/trainable": 162863
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.017895536497235298,
+      "grad_norm": 0.0019508687546476722,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 5.4065520089352503e-05,
+      "loss": 3.1739040423417464e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 357,
       "tokens/total": 10799328,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 163318
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.8479411602020264,
+      "grad_norm": 0.7139555811882019,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.01497839204967022,
+      "loss": 0.010268578305840492,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01509,
+      "ppl": 1.01032,
       "step": 358,
       "tokens/total": 10829488,
-      "tokens/train_per_sec_per_gpu": 32.89,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 163767
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.0015704578254371881,
+      "grad_norm": 0.006852464284747839,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 3.0390210667974316e-05,
+      "loss": 4.873241778113879e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 359,
       "tokens/total": 10859936,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 164208
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.014669318683445454,
+      "grad_norm": 0.0018247900297865272,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010030368866864592,
+      "loss": 3.88835760531947e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00004,
       "step": 360,
       "tokens/total": 10890400,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 164655
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.3479421138763428,
+      "grad_norm": 0.0436873696744442,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 0.0026096655055880547,
+      "loss": 0.0003687943681143224,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00261,
+      "ppl": 1.00037,
       "step": 361,
       "tokens/total": 10920624,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 165100
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.004394443240016699,
+      "grad_norm": 0.26735150814056396,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 6.116658914834261e-05,
+      "loss": 0.00039605735219083726,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00006,
+      "ppl": 1.0004,
       "step": 362,
       "tokens/total": 10951136,
-      "tokens/train_per_sec_per_gpu": 34.84,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 165587
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.016197621822357178,
+      "grad_norm": 0.0964483991265297,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0001775357231963426,
+      "loss": 0.0007330900407396257,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00018,
+      "ppl": 1.00073,
       "step": 363,
       "tokens/total": 10981344,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 166034
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.09478334337472916,
+      "grad_norm": 0.05367618799209595,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.0005033796769566834,
+      "loss": 0.0005384897813200951,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0005,
+      "ppl": 1.00054,
       "step": 364,
       "tokens/total": 11011552,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 166485
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.014899010770022869,
+      "grad_norm": 0.10067912936210632,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 0.0001904651871882379,
+      "loss": 0.0006868956843391061,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00019,
+      "ppl": 1.00069,
       "step": 365,
       "tokens/total": 11041872,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 166973
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005555527750402689,
+      "grad_norm": 0.02697882056236267,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 0.00010058133921120316,
+      "loss": 0.0002464794088155031,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00025,
       "step": 366,
       "tokens/total": 11072192,
-      "tokens/train_per_sec_per_gpu": 29.1,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 167414
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.20706292986869812,
+      "grad_norm": 0.0026214183308184147,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 0.0005888720043003559,
+      "loss": 5.2629769925260916e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00059,
+      "ppl": 1.00005,
       "step": 367,
       "tokens/total": 11102736,
-      "tokens/train_per_sec_per_gpu": 30.11,
+      "tokens/train_per_sec_per_gpu": 30.23,
       "tokens/trainable": 167851
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.0007191727054305375,
+      "grad_norm": 0.0018140808679163456,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 1.2304372830840293e-05,
+      "loss": 2.9512597393477336e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 368,
       "tokens/total": 11133056,
-      "tokens/train_per_sec_per_gpu": 35.48,
+      "tokens/train_per_sec_per_gpu": 35.56,
       "tokens/trainable": 168294
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.030878128483891487,
+      "grad_norm": 0.012045321986079216,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 0.0002043281274382025,
+      "loss": 7.606908184243366e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0002,
+      "ppl": 1.00008,
       "step": 369,
       "tokens/total": 11163568,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 168769
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.008817057125270367,
+      "grad_norm": 0.013453424908220768,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 0.00011464582348708063,
+      "loss": 0.00015325279673561454,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00015,
       "step": 370,
       "tokens/total": 11193936,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.83,
       "tokens/trainable": 169270
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.023856336250901222,
+      "grad_norm": 0.003458227962255478,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 0.00013431125262286514,
+      "loss": 5.63332432648167e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00013,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11224480,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 169711
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.010502993129193783,
+      "grad_norm": 0.0063622924499213696,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 0.00015104333579074591,
+      "loss": 9.534892160445452e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00015,
+      "ppl": 1.0001,
       "step": 372,
       "tokens/total": 11254912,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 170185
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.026970423758029938,
+      "grad_norm": 0.04558980092406273,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.00026081278338097036,
+      "loss": 0.000613630109000951,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00026,
+      "ppl": 1.00061,
       "step": 373,
       "tokens/total": 11285104,
-      "tokens/train_per_sec_per_gpu": 33.08,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 170650
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.020325109362602234,
+      "grad_norm": 0.014873786829411983,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.00014532633940689266,
+      "loss": 0.0002038514503510669,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 374,
       "tokens/total": 11315600,
-      "tokens/train_per_sec_per_gpu": 31.98,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 171126
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.033260464668273926,
+      "grad_norm": 0.058739252388477325,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 0.00045571548980660737,
+      "loss": 0.0010346119524911046,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00046,
+      "ppl": 1.00104,
       "step": 375,
       "tokens/total": 11345776,
       "tokens/train_per_sec_per_gpu": 35.6,
@@ -5261,128 +5261,128 @@
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.017290709540247917,
+      "grad_norm": 0.24105991423130035,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 0.00014910403115209192,
+      "loss": 0.0035716122947633266,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00015,
+      "ppl": 1.00358,
       "step": 376,
       "tokens/total": 11376208,
-      "tokens/train_per_sec_per_gpu": 28.09,
+      "tokens/train_per_sec_per_gpu": 28.14,
       "tokens/trainable": 172017
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.04457363486289978,
+      "grad_norm": 0.08793950080871582,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 6.834026135038584e-05,
+      "loss": 0.0007833336712792516,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00007,
+      "ppl": 1.00078,
       "step": 377,
       "tokens/total": 11406512,
-      "tokens/train_per_sec_per_gpu": 35.99,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 172493
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.002268735785037279,
+      "grad_norm": 0.009154544211924076,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 3.245133120799437e-05,
+      "loss": 0.00011752049613278359,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00012,
       "step": 378,
       "tokens/total": 11436544,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 172951
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.0038171466439962387,
+      "grad_norm": 0.012918495573103428,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 5.4866883147042245e-05,
+      "loss": 0.00011567945330170915,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00012,
       "step": 379,
       "tokens/total": 11467008,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 173435
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005268535576760769,
+      "grad_norm": 0.008966504596173763,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 8.241234172601253e-05,
+      "loss": 8.938623068388551e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00009,
       "step": 380,
       "tokens/total": 11497184,
-      "tokens/train_per_sec_per_gpu": 30.4,
+      "tokens/train_per_sec_per_gpu": 30.47,
       "tokens/trainable": 173900
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.016057243570685387,
+      "grad_norm": 0.03181751072406769,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 0.00014235377602744848,
+      "loss": 0.00013083787052892148,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00013,
       "step": 381,
       "tokens/total": 11527600,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 174371
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005853456910699606,
+      "grad_norm": 0.0009072935208678246,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 5.001476893085055e-05,
+      "loss": 2.5600436856620945e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 382,
       "tokens/total": 11557808,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 174826
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0031664848793298006,
+      "grad_norm": 0.0050596860237419605,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 4.210277256788686e-05,
+      "loss": 5.778766353614628e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 383,
       "tokens/total": 11588352,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 175276
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.08838633447885513,
+      "grad_norm": 0.008304890245199203,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 0.0010394920827820897,
+      "loss": 0.00012505470658652484,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00104,
+      "ppl": 1.00013,
       "step": 384,
       "tokens/total": 11618640,
-      "tokens/train_per_sec_per_gpu": 33.69,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 175720
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_model.safetensors
index 9f3a618db09256b3656f7298393e48c04cc46828..ddc4087be06c6b94a33f8941f4f0c516ea2cce1f 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05723c03b88c39267b118122df5b55a796d2304d8fcdd49b3155b1d44ad82bc6
+oid sha256:5a7cb7b80828a935abc743ce1b3296d23f6b4af6fd9ec8e43392059a3ae4e893
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/optimizer.pt
index d27ecb6f85f9192ae634998271153f2d03e3486f..597a8931f4d56d7dfd3438270f8b77a99bf9e09f 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a7472c6ae170db8b2c705fc058bed1e8e2e71f8471cc95ac00fc09259427abd
+oid sha256:b1b828d321a91c5ed711734eacce07b16ea0f63ebf7695f8e1c77243f6364e6e
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/trainer_state.json
index 5ad4fed4915b93d3583f06f4c9904aa9ba184c91..96a951d2408c4c4a848fd4c78c3a60949ff6246c 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-416/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1525 +3735,1525 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.0037202269304543734,
+      "grad_norm": 0.024732626974582672,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 7.238816760946065e-05,
+      "loss": 0.0001429672265658155,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 353,
       "tokens/total": 10678080,
-      "tokens/train_per_sec_per_gpu": 28.86,
+      "tokens/train_per_sec_per_gpu": 28.93,
       "tokens/trainable": 161555
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0021510994993150234,
+      "grad_norm": 0.037309419363737106,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 3.6805788113269955e-05,
+      "loss": 7.214388460852206e-05,
       "memory/device_reserved (GiB)": 35.41,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 354,
       "tokens/total": 10708544,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.75,
       "tokens/trainable": 162020
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.002642499515786767,
+      "grad_norm": 0.00885045062750578,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 4.371708200778812e-05,
+      "loss": 0.00011830432777060196,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 355,
       "tokens/total": 10739024,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 162448
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0038646068423986435,
+      "grad_norm": 0.06791900098323822,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 7.11614266037941e-05,
+      "loss": 0.0003609730047173798,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.61,
       "memory/max_allocated (GiB)": 33.61,
-      "ppl": 1.00007,
+      "ppl": 1.00036,
       "step": 356,
       "tokens/total": 10769024,
-      "tokens/train_per_sec_per_gpu": 28.2,
+      "tokens/train_per_sec_per_gpu": 28.26,
       "tokens/trainable": 162863
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.017895536497235298,
+      "grad_norm": 0.0019508687546476722,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 5.4065520089352503e-05,
+      "loss": 3.1739040423417464e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 357,
       "tokens/total": 10799328,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 163318
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.8479411602020264,
+      "grad_norm": 0.7139555811882019,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.01497839204967022,
+      "loss": 0.010268578305840492,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01509,
+      "ppl": 1.01032,
       "step": 358,
       "tokens/total": 10829488,
-      "tokens/train_per_sec_per_gpu": 32.89,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 163767
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.0015704578254371881,
+      "grad_norm": 0.006852464284747839,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 3.0390210667974316e-05,
+      "loss": 4.873241778113879e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 359,
       "tokens/total": 10859936,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 164208
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.014669318683445454,
+      "grad_norm": 0.0018247900297865272,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010030368866864592,
+      "loss": 3.88835760531947e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00004,
       "step": 360,
       "tokens/total": 10890400,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 164655
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.3479421138763428,
+      "grad_norm": 0.0436873696744442,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 0.0026096655055880547,
+      "loss": 0.0003687943681143224,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00261,
+      "ppl": 1.00037,
       "step": 361,
       "tokens/total": 10920624,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 165100
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.004394443240016699,
+      "grad_norm": 0.26735150814056396,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 6.116658914834261e-05,
+      "loss": 0.00039605735219083726,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00006,
+      "ppl": 1.0004,
       "step": 362,
       "tokens/total": 10951136,
-      "tokens/train_per_sec_per_gpu": 34.84,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 165587
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.016197621822357178,
+      "grad_norm": 0.0964483991265297,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0001775357231963426,
+      "loss": 0.0007330900407396257,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00018,
+      "ppl": 1.00073,
       "step": 363,
       "tokens/total": 10981344,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 166034
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.09478334337472916,
+      "grad_norm": 0.05367618799209595,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.0005033796769566834,
+      "loss": 0.0005384897813200951,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0005,
+      "ppl": 1.00054,
       "step": 364,
       "tokens/total": 11011552,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 166485
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.014899010770022869,
+      "grad_norm": 0.10067912936210632,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 0.0001904651871882379,
+      "loss": 0.0006868956843391061,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00019,
+      "ppl": 1.00069,
       "step": 365,
       "tokens/total": 11041872,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 166973
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005555527750402689,
+      "grad_norm": 0.02697882056236267,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 0.00010058133921120316,
+      "loss": 0.0002464794088155031,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00025,
       "step": 366,
       "tokens/total": 11072192,
-      "tokens/train_per_sec_per_gpu": 29.1,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 167414
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.20706292986869812,
+      "grad_norm": 0.0026214183308184147,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 0.0005888720043003559,
+      "loss": 5.2629769925260916e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00059,
+      "ppl": 1.00005,
       "step": 367,
       "tokens/total": 11102736,
-      "tokens/train_per_sec_per_gpu": 30.11,
+      "tokens/train_per_sec_per_gpu": 30.23,
       "tokens/trainable": 167851
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.0007191727054305375,
+      "grad_norm": 0.0018140808679163456,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 1.2304372830840293e-05,
+      "loss": 2.9512597393477336e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 368,
       "tokens/total": 11133056,
-      "tokens/train_per_sec_per_gpu": 35.48,
+      "tokens/train_per_sec_per_gpu": 35.56,
       "tokens/trainable": 168294
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.030878128483891487,
+      "grad_norm": 0.012045321986079216,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 0.0002043281274382025,
+      "loss": 7.606908184243366e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0002,
+      "ppl": 1.00008,
       "step": 369,
       "tokens/total": 11163568,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 168769
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.008817057125270367,
+      "grad_norm": 0.013453424908220768,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 0.00011464582348708063,
+      "loss": 0.00015325279673561454,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00015,
       "step": 370,
       "tokens/total": 11193936,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.83,
       "tokens/trainable": 169270
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.023856336250901222,
+      "grad_norm": 0.003458227962255478,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 0.00013431125262286514,
+      "loss": 5.63332432648167e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00013,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11224480,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 169711
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.010502993129193783,
+      "grad_norm": 0.0063622924499213696,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 0.00015104333579074591,
+      "loss": 9.534892160445452e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00015,
+      "ppl": 1.0001,
       "step": 372,
       "tokens/total": 11254912,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 170185
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.026970423758029938,
+      "grad_norm": 0.04558980092406273,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.00026081278338097036,
+      "loss": 0.000613630109000951,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00026,
+      "ppl": 1.00061,
       "step": 373,
       "tokens/total": 11285104,
-      "tokens/train_per_sec_per_gpu": 33.08,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 170650
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.020325109362602234,
+      "grad_norm": 0.014873786829411983,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.00014532633940689266,
+      "loss": 0.0002038514503510669,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 374,
       "tokens/total": 11315600,
-      "tokens/train_per_sec_per_gpu": 31.98,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 171126
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.033260464668273926,
+      "grad_norm": 0.058739252388477325,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 0.00045571548980660737,
+      "loss": 0.0010346119524911046,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00046,
+      "ppl": 1.00104,
       "step": 375,
       "tokens/total": 11345776,
       "tokens/train_per_sec_per_gpu": 35.6,
@@ -5261,576 +5261,576 @@
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.017290709540247917,
+      "grad_norm": 0.24105991423130035,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 0.00014910403115209192,
+      "loss": 0.0035716122947633266,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00015,
+      "ppl": 1.00358,
       "step": 376,
       "tokens/total": 11376208,
-      "tokens/train_per_sec_per_gpu": 28.09,
+      "tokens/train_per_sec_per_gpu": 28.14,
       "tokens/trainable": 172017
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.04457363486289978,
+      "grad_norm": 0.08793950080871582,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 6.834026135038584e-05,
+      "loss": 0.0007833336712792516,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00007,
+      "ppl": 1.00078,
       "step": 377,
       "tokens/total": 11406512,
-      "tokens/train_per_sec_per_gpu": 35.99,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 172493
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.002268735785037279,
+      "grad_norm": 0.009154544211924076,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 3.245133120799437e-05,
+      "loss": 0.00011752049613278359,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00012,
       "step": 378,
       "tokens/total": 11436544,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 172951
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.0038171466439962387,
+      "grad_norm": 0.012918495573103428,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 5.4866883147042245e-05,
+      "loss": 0.00011567945330170915,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00012,
       "step": 379,
       "tokens/total": 11467008,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 173435
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005268535576760769,
+      "grad_norm": 0.008966504596173763,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 8.241234172601253e-05,
+      "loss": 8.938623068388551e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00009,
       "step": 380,
       "tokens/total": 11497184,
-      "tokens/train_per_sec_per_gpu": 30.4,
+      "tokens/train_per_sec_per_gpu": 30.47,
       "tokens/trainable": 173900
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.016057243570685387,
+      "grad_norm": 0.03181751072406769,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 0.00014235377602744848,
+      "loss": 0.00013083787052892148,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00013,
       "step": 381,
       "tokens/total": 11527600,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 174371
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005853456910699606,
+      "grad_norm": 0.0009072935208678246,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 5.001476893085055e-05,
+      "loss": 2.5600436856620945e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 382,
       "tokens/total": 11557808,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 174826
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0031664848793298006,
+      "grad_norm": 0.0050596860237419605,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 4.210277256788686e-05,
+      "loss": 5.778766353614628e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 383,
       "tokens/total": 11588352,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 175276
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.08838633447885513,
+      "grad_norm": 0.008304890245199203,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 0.0010394920827820897,
+      "loss": 0.00012505470658652484,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00104,
+      "ppl": 1.00013,
       "step": 384,
       "tokens/total": 11618640,
-      "tokens/train_per_sec_per_gpu": 33.69,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 175720
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.09446703642606735,
+      "grad_norm": 0.0035276354756206274,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0005370234721340239,
+      "loss": 4.851898120250553e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00054,
+      "ppl": 1.00005,
       "step": 385,
       "tokens/total": 11649072,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 176203
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.0013355027185752988,
+      "grad_norm": 0.001593019813299179,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 1.903088195831515e-05,
+      "loss": 3.330651088617742e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 386,
       "tokens/total": 11679552,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176651
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.011693151667714119,
+      "grad_norm": 0.002167076338082552,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 7.680666749365628e-05,
+      "loss": 3.344817378092557e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 387,
       "tokens/total": 11709760,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 177115
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.09523260593414307,
+      "grad_norm": 0.003447546623647213,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 0.00073521543527022,
+      "loss": 5.5277254432439804e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00074,
+      "ppl": 1.00006,
       "step": 388,
       "tokens/total": 11740288,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.13,
       "tokens/trainable": 177601
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.005971741396933794,
+      "grad_norm": 0.0052979388274252415,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 6.468063656939194e-05,
+      "loss": 4.56162124464754e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00006,
+      "ppl": 1.00005,
       "step": 389,
       "tokens/total": 11770592,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 178050
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.0010247502941638231,
+      "grad_norm": 0.003032457083463669,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 1.5989648090908304e-05,
+      "loss": 3.829343768302351e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 390,
       "tokens/total": 11800752,
-      "tokens/train_per_sec_per_gpu": 36.38,
+      "tokens/train_per_sec_per_gpu": 36.42,
       "tokens/trainable": 178531
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.054420940577983856,
+      "grad_norm": 0.03542603179812431,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.0004186803416814655,
+      "loss": 0.0003605492820497602,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00042,
+      "ppl": 1.00036,
       "step": 391,
       "tokens/total": 11831360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 179000
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.002925195964053273,
+      "grad_norm": 0.004965933505445719,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 1.531536145193968e-05,
+      "loss": 7.091004226822406e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 392,
       "tokens/total": 11861552,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 179457
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 1.6402941942214966,
+      "grad_norm": 0.006339129991829395,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0027035027742385864,
+      "loss": 9.15761775104329e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00271,
+      "ppl": 1.00009,
       "step": 393,
       "tokens/total": 11891904,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 179890
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.0032207504846155643,
+      "grad_norm": 0.03921438008546829,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 3.968120290664956e-05,
+      "loss": 0.00038214243249967694,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00038,
       "step": 394,
       "tokens/total": 11922208,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 180341
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.013098486699163914,
+      "grad_norm": 0.08640608936548233,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.00011063168494729325,
+      "loss": 0.0006777399685233831,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00011,
+      "ppl": 1.00068,
       "step": 395,
       "tokens/total": 11952672,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 180762
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.01553372759371996,
+      "grad_norm": 0.08661718666553497,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 8.802580123301595e-05,
+      "loss": 0.0007016840972937644,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00009,
+      "ppl": 1.0007,
       "step": 396,
       "tokens/total": 11983088,
-      "tokens/train_per_sec_per_gpu": 35.04,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 181226
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.016083814203739166,
+      "grad_norm": 0.010069145821034908,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 8.931377669796348e-05,
+      "loss": 0.00011120665294583887,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00011,
       "step": 397,
       "tokens/total": 12013488,
-      "tokens/train_per_sec_per_gpu": 33.15,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 181714
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.05752059072256088,
+      "grad_norm": 0.03158818930387497,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.00015369296306744218,
+      "loss": 0.00024857826065272093,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00015,
+      "ppl": 1.00025,
       "step": 398,
       "tokens/total": 12043968,
-      "tokens/train_per_sec_per_gpu": 35.52,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 182142
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.0011741623748093843,
+      "grad_norm": 0.000873154669534415,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 1.8113165424438193e-05,
+      "loss": 2.1218824258539826e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00002,
       "step": 399,
       "tokens/total": 12074160,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.86,
       "tokens/trainable": 182604
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0048082731664180756,
+      "grad_norm": 0.009255572222173214,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 5.3348740038927644e-05,
+      "loss": 0.0001349164522252977,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00005,
+      "ppl": 1.00013,
       "step": 400,
       "tokens/total": 12104352,
-      "tokens/train_per_sec_per_gpu": 25.28,
+      "tokens/train_per_sec_per_gpu": 25.31,
       "tokens/trainable": 183011
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.006520449183881283,
+      "grad_norm": 0.010521743446588516,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 6.45049221930094e-05,
+      "loss": 7.793466647854075e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00006,
+      "ppl": 1.00008,
       "step": 401,
       "tokens/total": 12134240,
-      "tokens/train_per_sec_per_gpu": 31.69,
+      "tokens/train_per_sec_per_gpu": 31.81,
       "tokens/trainable": 183459
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.006227654870599508,
+      "grad_norm": 0.0013409090461209416,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 6.928759830771014e-05,
+      "loss": 2.7722922823159024e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 402,
       "tokens/total": 12164480,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.7,
       "tokens/trainable": 183959
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.32605040073394775,
+      "grad_norm": 0.10361713916063309,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.002115404698997736,
+      "loss": 0.0014142843428999186,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00212,
+      "ppl": 1.00142,
       "step": 403,
       "tokens/total": 12194640,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.5,
       "tokens/trainable": 184431
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.007742208894342184,
+      "grad_norm": 0.0026048943400382996,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 4.600908869178966e-05,
+      "loss": 4.14503738284111e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 404,
       "tokens/total": 12224960,
-      "tokens/train_per_sec_per_gpu": 31.34,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 184867
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.0049946485087275505,
+      "grad_norm": 0.009411919862031937,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 5.003535625291988e-05,
+      "loss": 7.322158489841968e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00005,
+      "ppl": 1.00007,
       "step": 405,
       "tokens/total": 12255008,
-      "tokens/train_per_sec_per_gpu": 29.25,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 185300
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.004598768427968025,
+      "grad_norm": 0.006792420521378517,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 3.4423381293891e-05,
+      "loss": 7.982828537933528e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00003,
+      "ppl": 1.00008,
       "step": 406,
       "tokens/total": 12285248,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 185754
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.0011464629787951708,
+      "grad_norm": 0.013020367361605167,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 1.9196500943508e-05,
+      "loss": 3.743396155186929e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 407,
       "tokens/total": 12315552,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 186213
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.004688040353357792,
+      "grad_norm": 0.0013089004205539823,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 5.2492636314127594e-05,
+      "loss": 2.306591341039166e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00002,
       "step": 408,
       "tokens/total": 12345904,
-      "tokens/train_per_sec_per_gpu": 32.72,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 186659
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.002637861529365182,
+      "grad_norm": 0.0012270932784304023,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 3.172009019181132e-05,
+      "loss": 2.1350417227949947e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 409,
       "tokens/total": 12376336,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.97,
       "tokens/trainable": 187083
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.0072015393525362015,
+      "grad_norm": 0.011230082251131535,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 5.452537880046293e-05,
+      "loss": 3.752015618374571e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 410,
       "tokens/total": 12406880,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 187530
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.007169651333242655,
+      "grad_norm": 0.004872993566095829,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 3.530656249495223e-05,
+      "loss": 5.5653974413871765e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 411,
       "tokens/total": 12437088,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 187972
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003889538114890456,
+      "grad_norm": 0.19391997158527374,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 3.5087461583316326e-05,
+      "loss": 0.000892186420969665,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00089,
       "step": 412,
       "tokens/total": 12465440,
-      "tokens/train_per_sec_per_gpu": 39.89,
+      "tokens/train_per_sec_per_gpu": 39.95,
       "tokens/trainable": 188417
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.5135430693626404,
+      "grad_norm": 0.09062197804450989,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.005697700660675764,
+      "loss": 0.0006304415874183178,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00571,
+      "ppl": 1.00063,
       "step": 413,
       "tokens/total": 12495968,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.57,
       "tokens/trainable": 188868
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.0010096468031406403,
+      "grad_norm": 0.009724686853587627,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 1.1318426913931035e-05,
+      "loss": 6.72380265314132e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 414,
       "tokens/total": 12526128,
-      "tokens/train_per_sec_per_gpu": 36.85,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 189337
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0004970752634108067,
+      "grad_norm": 0.19542834162712097,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 7.355167326750234e-06,
+      "loss": 0.001907092402689159,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00191,
       "step": 415,
       "tokens/total": 12556256,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 189750
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.0008849130244925618,
+      "grad_norm": 0.0465276800096035,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 1.4194254617905244e-05,
+      "loss": 0.0005661146715283394,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00057,
       "step": 416,
       "tokens/total": 12586736,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 190216
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_model.safetensors
index 856c12a4d59288ac43e5f2eac631af52491777f8..7398f9593aeb3e59dfc2a7c156c2ad9ed5baab96 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43bdbb695861ca6c533046f5060be937637c560db75e2f4b7c2e25087dd34f87
+oid sha256:1eb3d591ce153c2a4556c9f93c60c352d3c882d47e4ce5be7c18c310c4f811f4
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/optimizer.pt
index 9b9a580f4ab2d3d2e0ee142a358cea8057b12317..1518f52b978778190f89c493272fe777f1d5981f 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03973e721e8a5628b3dc60c89f72ebda5ff7b262b6698c43c8ce9e37b8256ddc
+oid sha256:3faad3751c767d14b48f00eccec7568928e290a8f8c3f619dc85813de53d3afc
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/trainer_state.json
index e25c9a9b9b52060d5b3bcc14fb33776208ad707b..be0fc93eccd3e6bcb1a72cf0f77869c9ad664585 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-448/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1525 +3735,1525 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.0037202269304543734,
+      "grad_norm": 0.024732626974582672,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 7.238816760946065e-05,
+      "loss": 0.0001429672265658155,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 353,
       "tokens/total": 10678080,
-      "tokens/train_per_sec_per_gpu": 28.86,
+      "tokens/train_per_sec_per_gpu": 28.93,
       "tokens/trainable": 161555
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0021510994993150234,
+      "grad_norm": 0.037309419363737106,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 3.6805788113269955e-05,
+      "loss": 7.214388460852206e-05,
       "memory/device_reserved (GiB)": 35.41,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 354,
       "tokens/total": 10708544,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.75,
       "tokens/trainable": 162020
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.002642499515786767,
+      "grad_norm": 0.00885045062750578,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 4.371708200778812e-05,
+      "loss": 0.00011830432777060196,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 355,
       "tokens/total": 10739024,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 162448
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0038646068423986435,
+      "grad_norm": 0.06791900098323822,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 7.11614266037941e-05,
+      "loss": 0.0003609730047173798,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.61,
       "memory/max_allocated (GiB)": 33.61,
-      "ppl": 1.00007,
+      "ppl": 1.00036,
       "step": 356,
       "tokens/total": 10769024,
-      "tokens/train_per_sec_per_gpu": 28.2,
+      "tokens/train_per_sec_per_gpu": 28.26,
       "tokens/trainable": 162863
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.017895536497235298,
+      "grad_norm": 0.0019508687546476722,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 5.4065520089352503e-05,
+      "loss": 3.1739040423417464e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 357,
       "tokens/total": 10799328,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 163318
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.8479411602020264,
+      "grad_norm": 0.7139555811882019,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.01497839204967022,
+      "loss": 0.010268578305840492,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01509,
+      "ppl": 1.01032,
       "step": 358,
       "tokens/total": 10829488,
-      "tokens/train_per_sec_per_gpu": 32.89,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 163767
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.0015704578254371881,
+      "grad_norm": 0.006852464284747839,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 3.0390210667974316e-05,
+      "loss": 4.873241778113879e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 359,
       "tokens/total": 10859936,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 164208
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.014669318683445454,
+      "grad_norm": 0.0018247900297865272,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010030368866864592,
+      "loss": 3.88835760531947e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00004,
       "step": 360,
       "tokens/total": 10890400,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 164655
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.3479421138763428,
+      "grad_norm": 0.0436873696744442,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 0.0026096655055880547,
+      "loss": 0.0003687943681143224,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00261,
+      "ppl": 1.00037,
       "step": 361,
       "tokens/total": 10920624,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 165100
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.004394443240016699,
+      "grad_norm": 0.26735150814056396,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 6.116658914834261e-05,
+      "loss": 0.00039605735219083726,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00006,
+      "ppl": 1.0004,
       "step": 362,
       "tokens/total": 10951136,
-      "tokens/train_per_sec_per_gpu": 34.84,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 165587
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.016197621822357178,
+      "grad_norm": 0.0964483991265297,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0001775357231963426,
+      "loss": 0.0007330900407396257,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00018,
+      "ppl": 1.00073,
       "step": 363,
       "tokens/total": 10981344,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 166034
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.09478334337472916,
+      "grad_norm": 0.05367618799209595,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.0005033796769566834,
+      "loss": 0.0005384897813200951,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0005,
+      "ppl": 1.00054,
       "step": 364,
       "tokens/total": 11011552,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 166485
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.014899010770022869,
+      "grad_norm": 0.10067912936210632,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 0.0001904651871882379,
+      "loss": 0.0006868956843391061,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00019,
+      "ppl": 1.00069,
       "step": 365,
       "tokens/total": 11041872,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 166973
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005555527750402689,
+      "grad_norm": 0.02697882056236267,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 0.00010058133921120316,
+      "loss": 0.0002464794088155031,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00025,
       "step": 366,
       "tokens/total": 11072192,
-      "tokens/train_per_sec_per_gpu": 29.1,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 167414
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.20706292986869812,
+      "grad_norm": 0.0026214183308184147,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 0.0005888720043003559,
+      "loss": 5.2629769925260916e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00059,
+      "ppl": 1.00005,
       "step": 367,
       "tokens/total": 11102736,
-      "tokens/train_per_sec_per_gpu": 30.11,
+      "tokens/train_per_sec_per_gpu": 30.23,
       "tokens/trainable": 167851
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.0007191727054305375,
+      "grad_norm": 0.0018140808679163456,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 1.2304372830840293e-05,
+      "loss": 2.9512597393477336e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 368,
       "tokens/total": 11133056,
-      "tokens/train_per_sec_per_gpu": 35.48,
+      "tokens/train_per_sec_per_gpu": 35.56,
       "tokens/trainable": 168294
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.030878128483891487,
+      "grad_norm": 0.012045321986079216,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 0.0002043281274382025,
+      "loss": 7.606908184243366e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0002,
+      "ppl": 1.00008,
       "step": 369,
       "tokens/total": 11163568,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 168769
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.008817057125270367,
+      "grad_norm": 0.013453424908220768,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 0.00011464582348708063,
+      "loss": 0.00015325279673561454,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00015,
       "step": 370,
       "tokens/total": 11193936,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.83,
       "tokens/trainable": 169270
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.023856336250901222,
+      "grad_norm": 0.003458227962255478,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 0.00013431125262286514,
+      "loss": 5.63332432648167e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00013,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11224480,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 169711
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.010502993129193783,
+      "grad_norm": 0.0063622924499213696,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 0.00015104333579074591,
+      "loss": 9.534892160445452e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00015,
+      "ppl": 1.0001,
       "step": 372,
       "tokens/total": 11254912,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 170185
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.026970423758029938,
+      "grad_norm": 0.04558980092406273,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.00026081278338097036,
+      "loss": 0.000613630109000951,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00026,
+      "ppl": 1.00061,
       "step": 373,
       "tokens/total": 11285104,
-      "tokens/train_per_sec_per_gpu": 33.08,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 170650
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.020325109362602234,
+      "grad_norm": 0.014873786829411983,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.00014532633940689266,
+      "loss": 0.0002038514503510669,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 374,
       "tokens/total": 11315600,
-      "tokens/train_per_sec_per_gpu": 31.98,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 171126
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.033260464668273926,
+      "grad_norm": 0.058739252388477325,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 0.00045571548980660737,
+      "loss": 0.0010346119524911046,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00046,
+      "ppl": 1.00104,
       "step": 375,
       "tokens/total": 11345776,
       "tokens/train_per_sec_per_gpu": 35.6,
@@ -5261,713 +5261,713 @@
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.017290709540247917,
+      "grad_norm": 0.24105991423130035,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 0.00014910403115209192,
+      "loss": 0.0035716122947633266,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00015,
+      "ppl": 1.00358,
       "step": 376,
       "tokens/total": 11376208,
-      "tokens/train_per_sec_per_gpu": 28.09,
+      "tokens/train_per_sec_per_gpu": 28.14,
       "tokens/trainable": 172017
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.04457363486289978,
+      "grad_norm": 0.08793950080871582,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 6.834026135038584e-05,
+      "loss": 0.0007833336712792516,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00007,
+      "ppl": 1.00078,
       "step": 377,
       "tokens/total": 11406512,
-      "tokens/train_per_sec_per_gpu": 35.99,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 172493
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.002268735785037279,
+      "grad_norm": 0.009154544211924076,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 3.245133120799437e-05,
+      "loss": 0.00011752049613278359,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00012,
       "step": 378,
       "tokens/total": 11436544,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 172951
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.0038171466439962387,
+      "grad_norm": 0.012918495573103428,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 5.4866883147042245e-05,
+      "loss": 0.00011567945330170915,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00012,
       "step": 379,
       "tokens/total": 11467008,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 173435
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005268535576760769,
+      "grad_norm": 0.008966504596173763,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 8.241234172601253e-05,
+      "loss": 8.938623068388551e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00009,
       "step": 380,
       "tokens/total": 11497184,
-      "tokens/train_per_sec_per_gpu": 30.4,
+      "tokens/train_per_sec_per_gpu": 30.47,
       "tokens/trainable": 173900
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.016057243570685387,
+      "grad_norm": 0.03181751072406769,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 0.00014235377602744848,
+      "loss": 0.00013083787052892148,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00013,
       "step": 381,
       "tokens/total": 11527600,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 174371
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005853456910699606,
+      "grad_norm": 0.0009072935208678246,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 5.001476893085055e-05,
+      "loss": 2.5600436856620945e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 382,
       "tokens/total": 11557808,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 174826
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0031664848793298006,
+      "grad_norm": 0.0050596860237419605,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 4.210277256788686e-05,
+      "loss": 5.778766353614628e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 383,
       "tokens/total": 11588352,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 175276
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.08838633447885513,
+      "grad_norm": 0.008304890245199203,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 0.0010394920827820897,
+      "loss": 0.00012505470658652484,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00104,
+      "ppl": 1.00013,
       "step": 384,
       "tokens/total": 11618640,
-      "tokens/train_per_sec_per_gpu": 33.69,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 175720
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.09446703642606735,
+      "grad_norm": 0.0035276354756206274,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0005370234721340239,
+      "loss": 4.851898120250553e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00054,
+      "ppl": 1.00005,
       "step": 385,
       "tokens/total": 11649072,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 176203
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.0013355027185752988,
+      "grad_norm": 0.001593019813299179,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 1.903088195831515e-05,
+      "loss": 3.330651088617742e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 386,
       "tokens/total": 11679552,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176651
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.011693151667714119,
+      "grad_norm": 0.002167076338082552,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 7.680666749365628e-05,
+      "loss": 3.344817378092557e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 387,
       "tokens/total": 11709760,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 177115
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.09523260593414307,
+      "grad_norm": 0.003447546623647213,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 0.00073521543527022,
+      "loss": 5.5277254432439804e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00074,
+      "ppl": 1.00006,
       "step": 388,
       "tokens/total": 11740288,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.13,
       "tokens/trainable": 177601
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.005971741396933794,
+      "grad_norm": 0.0052979388274252415,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 6.468063656939194e-05,
+      "loss": 4.56162124464754e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00006,
+      "ppl": 1.00005,
       "step": 389,
       "tokens/total": 11770592,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 178050
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.0010247502941638231,
+      "grad_norm": 0.003032457083463669,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 1.5989648090908304e-05,
+      "loss": 3.829343768302351e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 390,
       "tokens/total": 11800752,
-      "tokens/train_per_sec_per_gpu": 36.38,
+      "tokens/train_per_sec_per_gpu": 36.42,
       "tokens/trainable": 178531
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.054420940577983856,
+      "grad_norm": 0.03542603179812431,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.0004186803416814655,
+      "loss": 0.0003605492820497602,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00042,
+      "ppl": 1.00036,
       "step": 391,
       "tokens/total": 11831360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 179000
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.002925195964053273,
+      "grad_norm": 0.004965933505445719,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 1.531536145193968e-05,
+      "loss": 7.091004226822406e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 392,
       "tokens/total": 11861552,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 179457
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 1.6402941942214966,
+      "grad_norm": 0.006339129991829395,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0027035027742385864,
+      "loss": 9.15761775104329e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00271,
+      "ppl": 1.00009,
       "step": 393,
       "tokens/total": 11891904,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 179890
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.0032207504846155643,
+      "grad_norm": 0.03921438008546829,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 3.968120290664956e-05,
+      "loss": 0.00038214243249967694,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00038,
       "step": 394,
       "tokens/total": 11922208,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 180341
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.013098486699163914,
+      "grad_norm": 0.08640608936548233,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.00011063168494729325,
+      "loss": 0.0006777399685233831,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00011,
+      "ppl": 1.00068,
       "step": 395,
       "tokens/total": 11952672,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 180762
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.01553372759371996,
+      "grad_norm": 0.08661718666553497,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 8.802580123301595e-05,
+      "loss": 0.0007016840972937644,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00009,
+      "ppl": 1.0007,
       "step": 396,
       "tokens/total": 11983088,
-      "tokens/train_per_sec_per_gpu": 35.04,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 181226
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.016083814203739166,
+      "grad_norm": 0.010069145821034908,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 8.931377669796348e-05,
+      "loss": 0.00011120665294583887,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00011,
       "step": 397,
       "tokens/total": 12013488,
-      "tokens/train_per_sec_per_gpu": 33.15,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 181714
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.05752059072256088,
+      "grad_norm": 0.03158818930387497,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.00015369296306744218,
+      "loss": 0.00024857826065272093,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00015,
+      "ppl": 1.00025,
       "step": 398,
       "tokens/total": 12043968,
-      "tokens/train_per_sec_per_gpu": 35.52,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 182142
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.0011741623748093843,
+      "grad_norm": 0.000873154669534415,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 1.8113165424438193e-05,
+      "loss": 2.1218824258539826e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00002,
       "step": 399,
       "tokens/total": 12074160,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.86,
       "tokens/trainable": 182604
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0048082731664180756,
+      "grad_norm": 0.009255572222173214,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 5.3348740038927644e-05,
+      "loss": 0.0001349164522252977,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00005,
+      "ppl": 1.00013,
       "step": 400,
       "tokens/total": 12104352,
-      "tokens/train_per_sec_per_gpu": 25.28,
+      "tokens/train_per_sec_per_gpu": 25.31,
       "tokens/trainable": 183011
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.006520449183881283,
+      "grad_norm": 0.010521743446588516,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 6.45049221930094e-05,
+      "loss": 7.793466647854075e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00006,
+      "ppl": 1.00008,
       "step": 401,
       "tokens/total": 12134240,
-      "tokens/train_per_sec_per_gpu": 31.69,
+      "tokens/train_per_sec_per_gpu": 31.81,
       "tokens/trainable": 183459
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.006227654870599508,
+      "grad_norm": 0.0013409090461209416,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 6.928759830771014e-05,
+      "loss": 2.7722922823159024e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 402,
       "tokens/total": 12164480,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.7,
       "tokens/trainable": 183959
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.32605040073394775,
+      "grad_norm": 0.10361713916063309,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.002115404698997736,
+      "loss": 0.0014142843428999186,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00212,
+      "ppl": 1.00142,
       "step": 403,
       "tokens/total": 12194640,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.5,
       "tokens/trainable": 184431
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.007742208894342184,
+      "grad_norm": 0.0026048943400382996,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 4.600908869178966e-05,
+      "loss": 4.14503738284111e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 404,
       "tokens/total": 12224960,
-      "tokens/train_per_sec_per_gpu": 31.34,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 184867
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.0049946485087275505,
+      "grad_norm": 0.009411919862031937,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 5.003535625291988e-05,
+      "loss": 7.322158489841968e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00005,
+      "ppl": 1.00007,
       "step": 405,
       "tokens/total": 12255008,
-      "tokens/train_per_sec_per_gpu": 29.25,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 185300
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.004598768427968025,
+      "grad_norm": 0.006792420521378517,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 3.4423381293891e-05,
+      "loss": 7.982828537933528e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00003,
+      "ppl": 1.00008,
       "step": 406,
       "tokens/total": 12285248,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 185754
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.0011464629787951708,
+      "grad_norm": 0.013020367361605167,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 1.9196500943508e-05,
+      "loss": 3.743396155186929e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 407,
       "tokens/total": 12315552,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 186213
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.004688040353357792,
+      "grad_norm": 0.0013089004205539823,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 5.2492636314127594e-05,
+      "loss": 2.306591341039166e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00002,
       "step": 408,
       "tokens/total": 12345904,
-      "tokens/train_per_sec_per_gpu": 32.72,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 186659
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.002637861529365182,
+      "grad_norm": 0.0012270932784304023,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 3.172009019181132e-05,
+      "loss": 2.1350417227949947e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 409,
       "tokens/total": 12376336,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.97,
       "tokens/trainable": 187083
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.0072015393525362015,
+      "grad_norm": 0.011230082251131535,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 5.452537880046293e-05,
+      "loss": 3.752015618374571e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 410,
       "tokens/total": 12406880,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 187530
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.007169651333242655,
+      "grad_norm": 0.004872993566095829,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 3.530656249495223e-05,
+      "loss": 5.5653974413871765e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 411,
       "tokens/total": 12437088,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 187972
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003889538114890456,
+      "grad_norm": 0.19391997158527374,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 3.5087461583316326e-05,
+      "loss": 0.000892186420969665,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00089,
       "step": 412,
       "tokens/total": 12465440,
-      "tokens/train_per_sec_per_gpu": 39.89,
+      "tokens/train_per_sec_per_gpu": 39.95,
       "tokens/trainable": 188417
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.5135430693626404,
+      "grad_norm": 0.09062197804450989,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.005697700660675764,
+      "loss": 0.0006304415874183178,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00571,
+      "ppl": 1.00063,
       "step": 413,
       "tokens/total": 12495968,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.57,
       "tokens/trainable": 188868
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.0010096468031406403,
+      "grad_norm": 0.009724686853587627,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 1.1318426913931035e-05,
+      "loss": 6.72380265314132e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 414,
       "tokens/total": 12526128,
-      "tokens/train_per_sec_per_gpu": 36.85,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 189337
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0004970752634108067,
+      "grad_norm": 0.19542834162712097,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 7.355167326750234e-06,
+      "loss": 0.001907092402689159,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00191,
       "step": 415,
       "tokens/total": 12556256,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 189750
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.0008849130244925618,
+      "grad_norm": 0.0465276800096035,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 1.4194254617905244e-05,
+      "loss": 0.0005661146715283394,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00057,
       "step": 416,
       "tokens/total": 12586736,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 190216
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.0022039280738681555,
+      "grad_norm": 0.0012954205740243196,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 2.4539594960515387e-05,
+      "loss": 1.8871982319978997e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
       "ppl": 1.00002,
       "step": 417,
       "tokens/total": 12617168,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 190708
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.027006179094314575,
+      "grad_norm": 0.0018749319715425372,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 0.00020442574168555439,
+      "loss": 3.220669532311149e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 418,
       "tokens/total": 12647680,
-      "tokens/train_per_sec_per_gpu": 30.85,
+      "tokens/train_per_sec_per_gpu": 30.9,
       "tokens/trainable": 191126
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.15684254467487335,
+      "grad_norm": 0.0022480092011392117,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 0.0007251293282024562,
+      "loss": 2.6564141080598347e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00073,
+      "ppl": 1.00003,
       "step": 419,
       "tokens/total": 12677952,
-      "tokens/train_per_sec_per_gpu": 30.3,
+      "tokens/train_per_sec_per_gpu": 30.34,
       "tokens/trainable": 191540
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.0033104538451880217,
+      "grad_norm": 0.1634059101343155,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 3.131272751488723e-05,
+      "loss": 0.0009822181891649961,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00098,
       "step": 420,
       "tokens/total": 12708416,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.69,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.001255685230717063,
+      "grad_norm": 0.0019404747290536761,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 1.9309976778458804e-05,
+      "loss": 2.167341335734818e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 421,
       "tokens/total": 12738640,
-      "tokens/train_per_sec_per_gpu": 31.28,
+      "tokens/train_per_sec_per_gpu": 31.31,
       "tokens/trainable": 192463
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.01599641516804695,
+      "grad_norm": 0.001059795613400638,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 9.235648030880839e-05,
+      "loss": 1.6543437595828436e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12769232,
-      "tokens/train_per_sec_per_gpu": 32.71,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 192918
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.048507146537303925,
+      "grad_norm": 0.01760418340563774,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 0.0002708940301090479,
+      "loss": 0.0001707405026536435,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00027,
+      "ppl": 1.00017,
       "step": 423,
       "tokens/total": 12799648,
-      "tokens/train_per_sec_per_gpu": 37.35,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 193425
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.010107563808560371,
+      "grad_norm": 0.0023663989268243313,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 6.99054216966033e-05,
+      "loss": 2.332203439436853e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 424,
       "tokens/total": 12829920,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 193898
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.005467226263135672,
+      "grad_norm": 0.0245437640696764,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 5.417566717369482e-05,
+      "loss": 0.0001577093207743019,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00005,
+      "ppl": 1.00016,
       "step": 425,
       "tokens/total": 12860112,
-      "tokens/train_per_sec_per_gpu": 29.24,
+      "tokens/train_per_sec_per_gpu": 29.29,
       "tokens/trainable": 194315
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.0015799012035131454,
+      "grad_norm": 0.0009002169244922698,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 2.4984849005704746e-05,
+      "loss": 1.4283305972639937e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 426,
       "tokens/total": 12890320,
       "tokens/train_per_sec_per_gpu": 34.92,
@@ -5975,310 +5975,310 @@
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.024157166481018066,
+      "grad_norm": 0.00033997284481301904,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 0.00010696032404666767,
+      "loss": 7.185776667029131e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00001,
       "step": 427,
       "tokens/total": 12920656,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 195262
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.00028090659179724753,
+      "grad_norm": 0.0005359657225199044,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 7.484430170734413e-06,
+      "loss": 9.948088518285658e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
       "ppl": 1.00001,
       "step": 428,
       "tokens/total": 12951040,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 195680
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.0011094068177044392,
+      "grad_norm": 0.0014890795573592186,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 1.9920153135899454e-05,
+      "loss": 2.028812377830036e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00002,
       "step": 429,
       "tokens/total": 12981232,
-      "tokens/train_per_sec_per_gpu": 32.83,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 196145
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.020794112235307693,
+      "grad_norm": 0.05130980163812637,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 0.00010971157462336123,
+      "loss": 0.0003292672336101532,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00033,
       "step": 430,
       "tokens/total": 13011616,
-      "tokens/train_per_sec_per_gpu": 32.39,
+      "tokens/train_per_sec_per_gpu": 32.42,
       "tokens/trainable": 196626
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.44080695509910583,
+      "grad_norm": 0.0014070137403905392,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0026788683608174324,
+      "loss": 1.0152909453609027e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.00268,
+      "ppl": 1.00001,
       "step": 431,
       "tokens/total": 13042208,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.17,
       "tokens/trainable": 197067
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0026339287869632244,
+      "grad_norm": 0.004116157069802284,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 2.082335777231492e-05,
+      "loss": 3.514952550176531e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 432,
       "tokens/total": 13072800,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.68,
       "tokens/trainable": 197565
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.004508621525019407,
+      "grad_norm": 0.01147819496691227,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 1.439991501683835e-05,
+      "loss": 7.313965761568397e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 433,
       "tokens/total": 13103328,
-      "tokens/train_per_sec_per_gpu": 32.88,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 198015
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.014811511151492596,
+      "grad_norm": 0.030784178525209427,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 0.00015316363715101033,
+      "loss": 0.0002448821614962071,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00024,
       "step": 434,
       "tokens/total": 13133472,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 198442
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.0037793368101119995,
+      "grad_norm": 0.0022457120940089226,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 1.4444960470427759e-05,
+      "loss": 1.4446297427639365e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 435,
       "tokens/total": 13164064,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 198914
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.24972176551818848,
+      "grad_norm": 0.0029742212500423193,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0017114672809839249,
+      "loss": 2.7415488148108125e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00171,
+      "ppl": 1.00003,
       "step": 436,
       "tokens/total": 13194464,
-      "tokens/train_per_sec_per_gpu": 33.35,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 199355
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.002479988383129239,
+      "grad_norm": 0.002260624896734953,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.9400333005469292e-05,
+      "loss": 2.7733602109947242e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00003,
       "step": 437,
       "tokens/total": 13224992,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 199830
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.0014780610799789429,
+      "grad_norm": 0.01198511105030775,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 1.5593774151057005e-05,
+      "loss": 9.154126746580005e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00009,
       "step": 438,
       "tokens/total": 13255296,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 200338
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.0010410621762275696,
+      "grad_norm": 0.003149093361571431,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 1.4683226254419424e-05,
+      "loss": 3.515004937071353e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13285344,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 200785
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0004247442411724478,
+      "grad_norm": 0.0003243185637984425,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 8.09474295238033e-06,
+      "loss": 7.168858701334102e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 440,
       "tokens/total": 13315632,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.32,
       "tokens/trainable": 201267
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0026301892939954996,
+      "grad_norm": 0.12771126627922058,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 1.7428235878469422e-05,
+      "loss": 0.0007142049144022167,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 441,
       "tokens/total": 13346048,
-      "tokens/train_per_sec_per_gpu": 34.16,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 201734
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0005576284602284431,
+      "grad_norm": 0.11483822762966156,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 8.656044883537106e-06,
+      "loss": 0.0006703532999381423,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00001,
+      "ppl": 1.00067,
       "step": 442,
       "tokens/total": 13376304,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.4,
       "tokens/trainable": 202174
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0009604030638001859,
+      "grad_norm": 0.00032006221590563655,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 1.4092523088038433e-05,
+      "loss": 9.367744496557862e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00001,
       "step": 443,
       "tokens/total": 13406720,
-      "tokens/train_per_sec_per_gpu": 36.13,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 202612
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.00019884438370354474,
+      "grad_norm": 0.0002659875026438385,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 4.5837323341402225e-06,
+      "loss": 5.530210728466045e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 444,
       "tokens/total": 13436464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 30.99,
       "tokens/trainable": 203020
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.05944995582103729,
+      "grad_norm": 0.04957196116447449,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 0.0002780454815365374,
+      "loss": 6.46735934424214e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00028,
+      "ppl": 1.00006,
       "step": 445,
       "tokens/total": 13466672,
-      "tokens/train_per_sec_per_gpu": 34.31,
+      "tokens/train_per_sec_per_gpu": 34.37,
       "tokens/trainable": 203479
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.005872223526239395,
+      "grad_norm": 0.0016840791795402765,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 3.424299575272016e-05,
+      "loss": 1.2613029866770376e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 446,
       "tokens/total": 13496736,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 203946
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.006013527978211641,
+      "grad_norm": 0.00042921333806589246,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 3.269856097176671e-05,
+      "loss": 7.416386324621271e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 447,
       "tokens/total": 13526928,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 204382
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.05103042349219322,
+      "grad_norm": 0.2165915071964264,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.00032247230410575867,
+      "loss": 0.0006914341356605291,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00032,
+      "ppl": 1.00069,
       "step": 448,
       "tokens/total": 13557392,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 204830
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_model.safetensors
index de60e36a99a3d5cb1d40ff7d5e0ce709814f78f0..01a41eb4726b93cb944ba253b2edcb1da89e5a3f 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:18a3e18c300c81fbd4fc5c615a02833db8462889fa56d74f8a21de4cdbd94b01
+oid sha256:5f9554509b9b3fde8b26650090aefc34c1e60f093cb129b6fdb78abfc58218c0
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/optimizer.pt
index 3b9a59abcd6aed6224b3c8530592dffd02572d61..8cc5ca06565a538dfbad03dd129d5677e0c95276 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fd6972487952b9b8cdba2186819f986803ccb69b1549585ef4e31ee392a485c
+oid sha256:d12a17b1f902b4886935dd59e4bb62b80c89b4ab2cfdc83d22bbb6372b66215a
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/trainer_state.json
index 2f25521cae184702f5b329a0703bc5c475711a24..1f0b367fc754f516a8b92dc62862aab782882914 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-480/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1525 +3735,1525 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.0037202269304543734,
+      "grad_norm": 0.024732626974582672,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 7.238816760946065e-05,
+      "loss": 0.0001429672265658155,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 353,
       "tokens/total": 10678080,
-      "tokens/train_per_sec_per_gpu": 28.86,
+      "tokens/train_per_sec_per_gpu": 28.93,
       "tokens/trainable": 161555
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0021510994993150234,
+      "grad_norm": 0.037309419363737106,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 3.6805788113269955e-05,
+      "loss": 7.214388460852206e-05,
       "memory/device_reserved (GiB)": 35.41,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 354,
       "tokens/total": 10708544,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.75,
       "tokens/trainable": 162020
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.002642499515786767,
+      "grad_norm": 0.00885045062750578,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 4.371708200778812e-05,
+      "loss": 0.00011830432777060196,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 355,
       "tokens/total": 10739024,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 162448
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0038646068423986435,
+      "grad_norm": 0.06791900098323822,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 7.11614266037941e-05,
+      "loss": 0.0003609730047173798,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.61,
       "memory/max_allocated (GiB)": 33.61,
-      "ppl": 1.00007,
+      "ppl": 1.00036,
       "step": 356,
       "tokens/total": 10769024,
-      "tokens/train_per_sec_per_gpu": 28.2,
+      "tokens/train_per_sec_per_gpu": 28.26,
       "tokens/trainable": 162863
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.017895536497235298,
+      "grad_norm": 0.0019508687546476722,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 5.4065520089352503e-05,
+      "loss": 3.1739040423417464e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 357,
       "tokens/total": 10799328,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 163318
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.8479411602020264,
+      "grad_norm": 0.7139555811882019,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.01497839204967022,
+      "loss": 0.010268578305840492,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01509,
+      "ppl": 1.01032,
       "step": 358,
       "tokens/total": 10829488,
-      "tokens/train_per_sec_per_gpu": 32.89,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 163767
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.0015704578254371881,
+      "grad_norm": 0.006852464284747839,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 3.0390210667974316e-05,
+      "loss": 4.873241778113879e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 359,
       "tokens/total": 10859936,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 164208
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.014669318683445454,
+      "grad_norm": 0.0018247900297865272,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010030368866864592,
+      "loss": 3.88835760531947e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00004,
       "step": 360,
       "tokens/total": 10890400,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 164655
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.3479421138763428,
+      "grad_norm": 0.0436873696744442,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 0.0026096655055880547,
+      "loss": 0.0003687943681143224,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00261,
+      "ppl": 1.00037,
       "step": 361,
       "tokens/total": 10920624,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 165100
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.004394443240016699,
+      "grad_norm": 0.26735150814056396,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 6.116658914834261e-05,
+      "loss": 0.00039605735219083726,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00006,
+      "ppl": 1.0004,
       "step": 362,
       "tokens/total": 10951136,
-      "tokens/train_per_sec_per_gpu": 34.84,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 165587
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.016197621822357178,
+      "grad_norm": 0.0964483991265297,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0001775357231963426,
+      "loss": 0.0007330900407396257,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00018,
+      "ppl": 1.00073,
       "step": 363,
       "tokens/total": 10981344,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 166034
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.09478334337472916,
+      "grad_norm": 0.05367618799209595,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.0005033796769566834,
+      "loss": 0.0005384897813200951,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0005,
+      "ppl": 1.00054,
       "step": 364,
       "tokens/total": 11011552,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 166485
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.014899010770022869,
+      "grad_norm": 0.10067912936210632,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 0.0001904651871882379,
+      "loss": 0.0006868956843391061,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00019,
+      "ppl": 1.00069,
       "step": 365,
       "tokens/total": 11041872,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 166973
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005555527750402689,
+      "grad_norm": 0.02697882056236267,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 0.00010058133921120316,
+      "loss": 0.0002464794088155031,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00025,
       "step": 366,
       "tokens/total": 11072192,
-      "tokens/train_per_sec_per_gpu": 29.1,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 167414
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.20706292986869812,
+      "grad_norm": 0.0026214183308184147,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 0.0005888720043003559,
+      "loss": 5.2629769925260916e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00059,
+      "ppl": 1.00005,
       "step": 367,
       "tokens/total": 11102736,
-      "tokens/train_per_sec_per_gpu": 30.11,
+      "tokens/train_per_sec_per_gpu": 30.23,
       "tokens/trainable": 167851
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.0007191727054305375,
+      "grad_norm": 0.0018140808679163456,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 1.2304372830840293e-05,
+      "loss": 2.9512597393477336e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 368,
       "tokens/total": 11133056,
-      "tokens/train_per_sec_per_gpu": 35.48,
+      "tokens/train_per_sec_per_gpu": 35.56,
       "tokens/trainable": 168294
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.030878128483891487,
+      "grad_norm": 0.012045321986079216,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 0.0002043281274382025,
+      "loss": 7.606908184243366e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0002,
+      "ppl": 1.00008,
       "step": 369,
       "tokens/total": 11163568,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 168769
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.008817057125270367,
+      "grad_norm": 0.013453424908220768,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 0.00011464582348708063,
+      "loss": 0.00015325279673561454,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00015,
       "step": 370,
       "tokens/total": 11193936,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.83,
       "tokens/trainable": 169270
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.023856336250901222,
+      "grad_norm": 0.003458227962255478,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 0.00013431125262286514,
+      "loss": 5.63332432648167e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00013,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11224480,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 169711
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.010502993129193783,
+      "grad_norm": 0.0063622924499213696,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 0.00015104333579074591,
+      "loss": 9.534892160445452e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00015,
+      "ppl": 1.0001,
       "step": 372,
       "tokens/total": 11254912,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 170185
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.026970423758029938,
+      "grad_norm": 0.04558980092406273,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.00026081278338097036,
+      "loss": 0.000613630109000951,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00026,
+      "ppl": 1.00061,
       "step": 373,
       "tokens/total": 11285104,
-      "tokens/train_per_sec_per_gpu": 33.08,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 170650
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.020325109362602234,
+      "grad_norm": 0.014873786829411983,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.00014532633940689266,
+      "loss": 0.0002038514503510669,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 374,
       "tokens/total": 11315600,
-      "tokens/train_per_sec_per_gpu": 31.98,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 171126
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.033260464668273926,
+      "grad_norm": 0.058739252388477325,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 0.00045571548980660737,
+      "loss": 0.0010346119524911046,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00046,
+      "ppl": 1.00104,
       "step": 375,
       "tokens/total": 11345776,
       "tokens/train_per_sec_per_gpu": 35.6,
@@ -5261,713 +5261,713 @@
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.017290709540247917,
+      "grad_norm": 0.24105991423130035,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 0.00014910403115209192,
+      "loss": 0.0035716122947633266,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00015,
+      "ppl": 1.00358,
       "step": 376,
       "tokens/total": 11376208,
-      "tokens/train_per_sec_per_gpu": 28.09,
+      "tokens/train_per_sec_per_gpu": 28.14,
       "tokens/trainable": 172017
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.04457363486289978,
+      "grad_norm": 0.08793950080871582,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 6.834026135038584e-05,
+      "loss": 0.0007833336712792516,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00007,
+      "ppl": 1.00078,
       "step": 377,
       "tokens/total": 11406512,
-      "tokens/train_per_sec_per_gpu": 35.99,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 172493
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.002268735785037279,
+      "grad_norm": 0.009154544211924076,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 3.245133120799437e-05,
+      "loss": 0.00011752049613278359,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00012,
       "step": 378,
       "tokens/total": 11436544,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 172951
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.0038171466439962387,
+      "grad_norm": 0.012918495573103428,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 5.4866883147042245e-05,
+      "loss": 0.00011567945330170915,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00012,
       "step": 379,
       "tokens/total": 11467008,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 173435
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005268535576760769,
+      "grad_norm": 0.008966504596173763,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 8.241234172601253e-05,
+      "loss": 8.938623068388551e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00009,
       "step": 380,
       "tokens/total": 11497184,
-      "tokens/train_per_sec_per_gpu": 30.4,
+      "tokens/train_per_sec_per_gpu": 30.47,
       "tokens/trainable": 173900
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.016057243570685387,
+      "grad_norm": 0.03181751072406769,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 0.00014235377602744848,
+      "loss": 0.00013083787052892148,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00013,
       "step": 381,
       "tokens/total": 11527600,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 174371
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005853456910699606,
+      "grad_norm": 0.0009072935208678246,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 5.001476893085055e-05,
+      "loss": 2.5600436856620945e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 382,
       "tokens/total": 11557808,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 174826
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0031664848793298006,
+      "grad_norm": 0.0050596860237419605,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 4.210277256788686e-05,
+      "loss": 5.778766353614628e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 383,
       "tokens/total": 11588352,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 175276
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.08838633447885513,
+      "grad_norm": 0.008304890245199203,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 0.0010394920827820897,
+      "loss": 0.00012505470658652484,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00104,
+      "ppl": 1.00013,
       "step": 384,
       "tokens/total": 11618640,
-      "tokens/train_per_sec_per_gpu": 33.69,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 175720
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.09446703642606735,
+      "grad_norm": 0.0035276354756206274,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0005370234721340239,
+      "loss": 4.851898120250553e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00054,
+      "ppl": 1.00005,
       "step": 385,
       "tokens/total": 11649072,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 176203
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.0013355027185752988,
+      "grad_norm": 0.001593019813299179,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 1.903088195831515e-05,
+      "loss": 3.330651088617742e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 386,
       "tokens/total": 11679552,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176651
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.011693151667714119,
+      "grad_norm": 0.002167076338082552,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 7.680666749365628e-05,
+      "loss": 3.344817378092557e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 387,
       "tokens/total": 11709760,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 177115
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.09523260593414307,
+      "grad_norm": 0.003447546623647213,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 0.00073521543527022,
+      "loss": 5.5277254432439804e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00074,
+      "ppl": 1.00006,
       "step": 388,
       "tokens/total": 11740288,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.13,
       "tokens/trainable": 177601
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.005971741396933794,
+      "grad_norm": 0.0052979388274252415,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 6.468063656939194e-05,
+      "loss": 4.56162124464754e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00006,
+      "ppl": 1.00005,
       "step": 389,
       "tokens/total": 11770592,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 178050
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.0010247502941638231,
+      "grad_norm": 0.003032457083463669,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 1.5989648090908304e-05,
+      "loss": 3.829343768302351e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 390,
       "tokens/total": 11800752,
-      "tokens/train_per_sec_per_gpu": 36.38,
+      "tokens/train_per_sec_per_gpu": 36.42,
       "tokens/trainable": 178531
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.054420940577983856,
+      "grad_norm": 0.03542603179812431,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.0004186803416814655,
+      "loss": 0.0003605492820497602,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00042,
+      "ppl": 1.00036,
       "step": 391,
       "tokens/total": 11831360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 179000
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.002925195964053273,
+      "grad_norm": 0.004965933505445719,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 1.531536145193968e-05,
+      "loss": 7.091004226822406e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 392,
       "tokens/total": 11861552,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 179457
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 1.6402941942214966,
+      "grad_norm": 0.006339129991829395,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0027035027742385864,
+      "loss": 9.15761775104329e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00271,
+      "ppl": 1.00009,
       "step": 393,
       "tokens/total": 11891904,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 179890
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.0032207504846155643,
+      "grad_norm": 0.03921438008546829,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 3.968120290664956e-05,
+      "loss": 0.00038214243249967694,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00038,
       "step": 394,
       "tokens/total": 11922208,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 180341
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.013098486699163914,
+      "grad_norm": 0.08640608936548233,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.00011063168494729325,
+      "loss": 0.0006777399685233831,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00011,
+      "ppl": 1.00068,
       "step": 395,
       "tokens/total": 11952672,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 180762
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.01553372759371996,
+      "grad_norm": 0.08661718666553497,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 8.802580123301595e-05,
+      "loss": 0.0007016840972937644,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00009,
+      "ppl": 1.0007,
       "step": 396,
       "tokens/total": 11983088,
-      "tokens/train_per_sec_per_gpu": 35.04,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 181226
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.016083814203739166,
+      "grad_norm": 0.010069145821034908,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 8.931377669796348e-05,
+      "loss": 0.00011120665294583887,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00011,
       "step": 397,
       "tokens/total": 12013488,
-      "tokens/train_per_sec_per_gpu": 33.15,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 181714
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.05752059072256088,
+      "grad_norm": 0.03158818930387497,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.00015369296306744218,
+      "loss": 0.00024857826065272093,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00015,
+      "ppl": 1.00025,
       "step": 398,
       "tokens/total": 12043968,
-      "tokens/train_per_sec_per_gpu": 35.52,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 182142
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.0011741623748093843,
+      "grad_norm": 0.000873154669534415,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 1.8113165424438193e-05,
+      "loss": 2.1218824258539826e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00002,
       "step": 399,
       "tokens/total": 12074160,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.86,
       "tokens/trainable": 182604
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0048082731664180756,
+      "grad_norm": 0.009255572222173214,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 5.3348740038927644e-05,
+      "loss": 0.0001349164522252977,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00005,
+      "ppl": 1.00013,
       "step": 400,
       "tokens/total": 12104352,
-      "tokens/train_per_sec_per_gpu": 25.28,
+      "tokens/train_per_sec_per_gpu": 25.31,
       "tokens/trainable": 183011
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.006520449183881283,
+      "grad_norm": 0.010521743446588516,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 6.45049221930094e-05,
+      "loss": 7.793466647854075e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00006,
+      "ppl": 1.00008,
       "step": 401,
       "tokens/total": 12134240,
-      "tokens/train_per_sec_per_gpu": 31.69,
+      "tokens/train_per_sec_per_gpu": 31.81,
       "tokens/trainable": 183459
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.006227654870599508,
+      "grad_norm": 0.0013409090461209416,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 6.928759830771014e-05,
+      "loss": 2.7722922823159024e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 402,
       "tokens/total": 12164480,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.7,
       "tokens/trainable": 183959
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.32605040073394775,
+      "grad_norm": 0.10361713916063309,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.002115404698997736,
+      "loss": 0.0014142843428999186,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00212,
+      "ppl": 1.00142,
       "step": 403,
       "tokens/total": 12194640,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.5,
       "tokens/trainable": 184431
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.007742208894342184,
+      "grad_norm": 0.0026048943400382996,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 4.600908869178966e-05,
+      "loss": 4.14503738284111e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 404,
       "tokens/total": 12224960,
-      "tokens/train_per_sec_per_gpu": 31.34,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 184867
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.0049946485087275505,
+      "grad_norm": 0.009411919862031937,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 5.003535625291988e-05,
+      "loss": 7.322158489841968e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00005,
+      "ppl": 1.00007,
       "step": 405,
       "tokens/total": 12255008,
-      "tokens/train_per_sec_per_gpu": 29.25,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 185300
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.004598768427968025,
+      "grad_norm": 0.006792420521378517,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 3.4423381293891e-05,
+      "loss": 7.982828537933528e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00003,
+      "ppl": 1.00008,
       "step": 406,
       "tokens/total": 12285248,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 185754
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.0011464629787951708,
+      "grad_norm": 0.013020367361605167,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 1.9196500943508e-05,
+      "loss": 3.743396155186929e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 407,
       "tokens/total": 12315552,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 186213
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.004688040353357792,
+      "grad_norm": 0.0013089004205539823,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 5.2492636314127594e-05,
+      "loss": 2.306591341039166e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00002,
       "step": 408,
       "tokens/total": 12345904,
-      "tokens/train_per_sec_per_gpu": 32.72,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 186659
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.002637861529365182,
+      "grad_norm": 0.0012270932784304023,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 3.172009019181132e-05,
+      "loss": 2.1350417227949947e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 409,
       "tokens/total": 12376336,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.97,
       "tokens/trainable": 187083
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.0072015393525362015,
+      "grad_norm": 0.011230082251131535,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 5.452537880046293e-05,
+      "loss": 3.752015618374571e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 410,
       "tokens/total": 12406880,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 187530
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.007169651333242655,
+      "grad_norm": 0.004872993566095829,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 3.530656249495223e-05,
+      "loss": 5.5653974413871765e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 411,
       "tokens/total": 12437088,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 187972
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003889538114890456,
+      "grad_norm": 0.19391997158527374,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 3.5087461583316326e-05,
+      "loss": 0.000892186420969665,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00089,
       "step": 412,
       "tokens/total": 12465440,
-      "tokens/train_per_sec_per_gpu": 39.89,
+      "tokens/train_per_sec_per_gpu": 39.95,
       "tokens/trainable": 188417
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.5135430693626404,
+      "grad_norm": 0.09062197804450989,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.005697700660675764,
+      "loss": 0.0006304415874183178,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00571,
+      "ppl": 1.00063,
       "step": 413,
       "tokens/total": 12495968,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.57,
       "tokens/trainable": 188868
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.0010096468031406403,
+      "grad_norm": 0.009724686853587627,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 1.1318426913931035e-05,
+      "loss": 6.72380265314132e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 414,
       "tokens/total": 12526128,
-      "tokens/train_per_sec_per_gpu": 36.85,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 189337
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0004970752634108067,
+      "grad_norm": 0.19542834162712097,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 7.355167326750234e-06,
+      "loss": 0.001907092402689159,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00191,
       "step": 415,
       "tokens/total": 12556256,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 189750
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.0008849130244925618,
+      "grad_norm": 0.0465276800096035,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 1.4194254617905244e-05,
+      "loss": 0.0005661146715283394,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00057,
       "step": 416,
       "tokens/total": 12586736,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 190216
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.0022039280738681555,
+      "grad_norm": 0.0012954205740243196,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 2.4539594960515387e-05,
+      "loss": 1.8871982319978997e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
       "ppl": 1.00002,
       "step": 417,
       "tokens/total": 12617168,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 190708
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.027006179094314575,
+      "grad_norm": 0.0018749319715425372,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 0.00020442574168555439,
+      "loss": 3.220669532311149e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 418,
       "tokens/total": 12647680,
-      "tokens/train_per_sec_per_gpu": 30.85,
+      "tokens/train_per_sec_per_gpu": 30.9,
       "tokens/trainable": 191126
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.15684254467487335,
+      "grad_norm": 0.0022480092011392117,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 0.0007251293282024562,
+      "loss": 2.6564141080598347e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00073,
+      "ppl": 1.00003,
       "step": 419,
       "tokens/total": 12677952,
-      "tokens/train_per_sec_per_gpu": 30.3,
+      "tokens/train_per_sec_per_gpu": 30.34,
       "tokens/trainable": 191540
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.0033104538451880217,
+      "grad_norm": 0.1634059101343155,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 3.131272751488723e-05,
+      "loss": 0.0009822181891649961,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00098,
       "step": 420,
       "tokens/total": 12708416,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.69,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.001255685230717063,
+      "grad_norm": 0.0019404747290536761,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 1.9309976778458804e-05,
+      "loss": 2.167341335734818e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 421,
       "tokens/total": 12738640,
-      "tokens/train_per_sec_per_gpu": 31.28,
+      "tokens/train_per_sec_per_gpu": 31.31,
       "tokens/trainable": 192463
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.01599641516804695,
+      "grad_norm": 0.001059795613400638,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 9.235648030880839e-05,
+      "loss": 1.6543437595828436e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12769232,
-      "tokens/train_per_sec_per_gpu": 32.71,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 192918
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.048507146537303925,
+      "grad_norm": 0.01760418340563774,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 0.0002708940301090479,
+      "loss": 0.0001707405026536435,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00027,
+      "ppl": 1.00017,
       "step": 423,
       "tokens/total": 12799648,
-      "tokens/train_per_sec_per_gpu": 37.35,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 193425
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.010107563808560371,
+      "grad_norm": 0.0023663989268243313,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 6.99054216966033e-05,
+      "loss": 2.332203439436853e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 424,
       "tokens/total": 12829920,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 193898
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.005467226263135672,
+      "grad_norm": 0.0245437640696764,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 5.417566717369482e-05,
+      "loss": 0.0001577093207743019,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00005,
+      "ppl": 1.00016,
       "step": 425,
       "tokens/total": 12860112,
-      "tokens/train_per_sec_per_gpu": 29.24,
+      "tokens/train_per_sec_per_gpu": 29.29,
       "tokens/trainable": 194315
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.0015799012035131454,
+      "grad_norm": 0.0009002169244922698,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 2.4984849005704746e-05,
+      "loss": 1.4283305972639937e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 426,
       "tokens/total": 12890320,
       "tokens/train_per_sec_per_gpu": 34.92,
@@ -5975,758 +5975,758 @@
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.024157166481018066,
+      "grad_norm": 0.00033997284481301904,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 0.00010696032404666767,
+      "loss": 7.185776667029131e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00001,
       "step": 427,
       "tokens/total": 12920656,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 195262
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.00028090659179724753,
+      "grad_norm": 0.0005359657225199044,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 7.484430170734413e-06,
+      "loss": 9.948088518285658e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
       "ppl": 1.00001,
       "step": 428,
       "tokens/total": 12951040,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 195680
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.0011094068177044392,
+      "grad_norm": 0.0014890795573592186,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 1.9920153135899454e-05,
+      "loss": 2.028812377830036e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00002,
       "step": 429,
       "tokens/total": 12981232,
-      "tokens/train_per_sec_per_gpu": 32.83,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 196145
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.020794112235307693,
+      "grad_norm": 0.05130980163812637,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 0.00010971157462336123,
+      "loss": 0.0003292672336101532,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00033,
       "step": 430,
       "tokens/total": 13011616,
-      "tokens/train_per_sec_per_gpu": 32.39,
+      "tokens/train_per_sec_per_gpu": 32.42,
       "tokens/trainable": 196626
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.44080695509910583,
+      "grad_norm": 0.0014070137403905392,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0026788683608174324,
+      "loss": 1.0152909453609027e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.00268,
+      "ppl": 1.00001,
       "step": 431,
       "tokens/total": 13042208,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.17,
       "tokens/trainable": 197067
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0026339287869632244,
+      "grad_norm": 0.004116157069802284,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 2.082335777231492e-05,
+      "loss": 3.514952550176531e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 432,
       "tokens/total": 13072800,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.68,
       "tokens/trainable": 197565
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.004508621525019407,
+      "grad_norm": 0.01147819496691227,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 1.439991501683835e-05,
+      "loss": 7.313965761568397e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 433,
       "tokens/total": 13103328,
-      "tokens/train_per_sec_per_gpu": 32.88,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 198015
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.014811511151492596,
+      "grad_norm": 0.030784178525209427,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 0.00015316363715101033,
+      "loss": 0.0002448821614962071,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00024,
       "step": 434,
       "tokens/total": 13133472,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 198442
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.0037793368101119995,
+      "grad_norm": 0.0022457120940089226,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 1.4444960470427759e-05,
+      "loss": 1.4446297427639365e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 435,
       "tokens/total": 13164064,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 198914
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.24972176551818848,
+      "grad_norm": 0.0029742212500423193,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0017114672809839249,
+      "loss": 2.7415488148108125e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00171,
+      "ppl": 1.00003,
       "step": 436,
       "tokens/total": 13194464,
-      "tokens/train_per_sec_per_gpu": 33.35,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 199355
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.002479988383129239,
+      "grad_norm": 0.002260624896734953,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.9400333005469292e-05,
+      "loss": 2.7733602109947242e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00003,
       "step": 437,
       "tokens/total": 13224992,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 199830
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.0014780610799789429,
+      "grad_norm": 0.01198511105030775,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 1.5593774151057005e-05,
+      "loss": 9.154126746580005e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00009,
       "step": 438,
       "tokens/total": 13255296,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 200338
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.0010410621762275696,
+      "grad_norm": 0.003149093361571431,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 1.4683226254419424e-05,
+      "loss": 3.515004937071353e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13285344,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 200785
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0004247442411724478,
+      "grad_norm": 0.0003243185637984425,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 8.09474295238033e-06,
+      "loss": 7.168858701334102e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 440,
       "tokens/total": 13315632,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.32,
       "tokens/trainable": 201267
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0026301892939954996,
+      "grad_norm": 0.12771126627922058,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 1.7428235878469422e-05,
+      "loss": 0.0007142049144022167,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 441,
       "tokens/total": 13346048,
-      "tokens/train_per_sec_per_gpu": 34.16,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 201734
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0005576284602284431,
+      "grad_norm": 0.11483822762966156,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 8.656044883537106e-06,
+      "loss": 0.0006703532999381423,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00001,
+      "ppl": 1.00067,
       "step": 442,
       "tokens/total": 13376304,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.4,
       "tokens/trainable": 202174
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0009604030638001859,
+      "grad_norm": 0.00032006221590563655,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 1.4092523088038433e-05,
+      "loss": 9.367744496557862e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00001,
       "step": 443,
       "tokens/total": 13406720,
-      "tokens/train_per_sec_per_gpu": 36.13,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 202612
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.00019884438370354474,
+      "grad_norm": 0.0002659875026438385,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 4.5837323341402225e-06,
+      "loss": 5.530210728466045e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 444,
       "tokens/total": 13436464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 30.99,
       "tokens/trainable": 203020
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.05944995582103729,
+      "grad_norm": 0.04957196116447449,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 0.0002780454815365374,
+      "loss": 6.46735934424214e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00028,
+      "ppl": 1.00006,
       "step": 445,
       "tokens/total": 13466672,
-      "tokens/train_per_sec_per_gpu": 34.31,
+      "tokens/train_per_sec_per_gpu": 34.37,
       "tokens/trainable": 203479
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.005872223526239395,
+      "grad_norm": 0.0016840791795402765,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 3.424299575272016e-05,
+      "loss": 1.2613029866770376e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 446,
       "tokens/total": 13496736,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 203946
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.006013527978211641,
+      "grad_norm": 0.00042921333806589246,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 3.269856097176671e-05,
+      "loss": 7.416386324621271e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 447,
       "tokens/total": 13526928,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 204382
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.05103042349219322,
+      "grad_norm": 0.2165915071964264,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.00032247230410575867,
+      "loss": 0.0006914341356605291,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00032,
+      "ppl": 1.00069,
       "step": 448,
       "tokens/total": 13557392,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 204830
     },
     {
       "epoch": 1.75390625,
-      "grad_norm": 0.13250835239887238,
+      "grad_norm": 0.0009792562341317534,
       "learning_rate": 1.3780999708818058e-05,
-      "loss": 0.0008940898114815354,
+      "loss": 1.4504414139082655e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00089,
+      "ppl": 1.00001,
       "step": 449,
       "tokens/total": 13587776,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.47,
       "tokens/trainable": 205300
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 0.24642030894756317,
+      "grad_norm": 0.028498223051428795,
       "learning_rate": 1.3665385037857758e-05,
-      "loss": 0.0012931979726999998,
+      "loss": 0.00028163049137219787,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00129,
+      "ppl": 1.00028,
       "step": 450,
       "tokens/total": 13618112,
-      "tokens/train_per_sec_per_gpu": 30.62,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 205726
     },
     {
       "epoch": 1.76171875,
-      "grad_norm": 0.011528799310326576,
+      "grad_norm": 0.003279345342889428,
       "learning_rate": 1.3551490468947126e-05,
-      "loss": 6.569598917849362e-05,
+      "loss": 2.595262776594609e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 451,
       "tokens/total": 13648512,
-      "tokens/train_per_sec_per_gpu": 31.55,
+      "tokens/train_per_sec_per_gpu": 31.63,
       "tokens/trainable": 206163
     },
     {
       "epoch": 1.765625,
-      "grad_norm": 1.0818549394607544,
+      "grad_norm": 0.4961508512496948,
       "learning_rate": 1.3439320741704075e-05,
-      "loss": 0.015596212819218636,
+      "loss": 0.00459528062492609,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01572,
+      "ppl": 1.00461,
       "step": 452,
       "tokens/total": 13678704,
-      "tokens/train_per_sec_per_gpu": 35.41,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 206619
     },
     {
       "epoch": 1.76953125,
-      "grad_norm": 0.0007326967315748334,
+      "grad_norm": 0.009086468257009983,
       "learning_rate": 1.3328880523968808e-05,
-      "loss": 1.2033770872221794e-05,
+      "loss": 6.577694148290902e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 453,
       "tokens/total": 13709232,
-      "tokens/train_per_sec_per_gpu": 37.18,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 207101
     },
     {
       "epoch": 1.7734375,
-      "grad_norm": 0.0023919539526104927,
+      "grad_norm": 0.009036197327077389,
       "learning_rate": 1.3220174411609587e-05,
-      "loss": 1.262133719137637e-05,
+      "loss": 6.735372880939394e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 454,
       "tokens/total": 13739600,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.77,
       "tokens/trainable": 207546
     },
     {
       "epoch": 1.77734375,
-      "grad_norm": 0.0020972786005586386,
+      "grad_norm": 0.009665202349424362,
       "learning_rate": 1.3113206928331471e-05,
-      "loss": 2.476977533660829e-05,
+      "loss": 7.266137981787324e-05,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 455,
       "tokens/total": 13769936,
-      "tokens/train_per_sec_per_gpu": 34.25,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 207989
     },
     {
       "epoch": 1.78125,
-      "grad_norm": 0.000797569751739502,
+      "grad_norm": 0.0004666070453822613,
       "learning_rate": 1.300798252548806e-05,
-      "loss": 1.45716385304695e-05,
+      "loss": 6.039275831426494e-06,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00001,
       "step": 456,
       "tokens/total": 13800240,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 208402
     },
     {
       "epoch": 1.78515625,
-      "grad_norm": 0.011376727372407913,
+      "grad_norm": 0.00044853391591459513,
       "learning_rate": 1.2904505581896265e-05,
-      "loss": 7.903270306997001e-05,
+      "loss": 8.292890015582088e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 457,
       "tokens/total": 13830512,
-      "tokens/train_per_sec_per_gpu": 36.61,
+      "tokens/train_per_sec_per_gpu": 36.58,
       "tokens/trainable": 208904
     },
     {
       "epoch": 1.7890625,
-      "grad_norm": 0.0013566205743700266,
+      "grad_norm": 0.0033692270517349243,
       "learning_rate": 1.2802780403654082e-05,
-      "loss": 1.925312972161919e-05,
+      "loss": 2.3157112082117237e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
       "ppl": 1.00002,
       "step": 458,
       "tokens/total": 13860832,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 209382
     },
     {
       "epoch": 1.79296875,
-      "grad_norm": 0.00047398527385666966,
+      "grad_norm": 0.0010797431459650397,
       "learning_rate": 1.2702811223961408e-05,
-      "loss": 1.119351145462133e-05,
+      "loss": 1.119279841077514e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00001,
       "step": 459,
       "tokens/total": 13890992,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 209833
     },
     {
       "epoch": 1.796875,
-      "grad_norm": 0.04677487164735794,
+      "grad_norm": 0.04658913239836693,
       "learning_rate": 1.2604602202943861e-05,
-      "loss": 0.0001750149531289935,
+      "loss": 0.0003375259111635387,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00018,
+      "ppl": 1.00034,
       "step": 460,
       "tokens/total": 13921424,
-      "tokens/train_per_sec_per_gpu": 32.5,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 210284
     },
     {
       "epoch": 1.80078125,
-      "grad_norm": 0.024663345888257027,
+      "grad_norm": 0.01848040148615837,
       "learning_rate": 1.2508157427479686e-05,
-      "loss": 9.222347580362111e-05,
+      "loss": 0.00016109315038193017,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00016,
       "step": 461,
       "tokens/total": 13951504,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.07,
       "tokens/trainable": 210769
     },
     {
       "epoch": 1.8046875,
-      "grad_norm": 0.0004740248841699213,
+      "grad_norm": 0.05658251419663429,
       "learning_rate": 1.2413480911029655e-05,
-      "loss": 1.0696679964894429e-05,
+      "loss": 0.00017825173563323915,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00001,
+      "ppl": 1.00018,
       "step": 462,
       "tokens/total": 13979728,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.41,
       "tokens/trainable": 211193
     },
     {
       "epoch": 1.80859375,
-      "grad_norm": 0.0007590787718072534,
+      "grad_norm": 0.0006594725418835878,
       "learning_rate": 1.2320576593470082e-05,
-      "loss": 1.3936740288045257e-05,
+      "loss": 9.709075129649136e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00001,
       "step": 463,
       "tokens/total": 14009936,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 211655
     },
     {
       "epoch": 1.8125,
-      "grad_norm": 0.002108624204993248,
+      "grad_norm": 0.02737571857869625,
       "learning_rate": 1.2229448340928828e-05,
-      "loss": 2.675112227734644e-05,
+      "loss": 0.00017574361118022352,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00018,
       "step": 464,
       "tokens/total": 14039872,
-      "tokens/train_per_sec_per_gpu": 29.28,
+      "tokens/train_per_sec_per_gpu": 29.2,
       "tokens/trainable": 212085
     },
     {
       "epoch": 1.81640625,
-      "grad_norm": 0.0009435649262741208,
+      "grad_norm": 0.0004917355254292488,
       "learning_rate": 1.2140099945624458e-05,
-      "loss": 1.0821668183780275e-05,
+      "loss": 8.16806459624786e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00001,
       "step": 465,
       "tokens/total": 14070096,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 212570
     },
     {
       "epoch": 1.8203125,
-      "grad_norm": 0.011040126904845238,
+      "grad_norm": 0.0015915038529783487,
       "learning_rate": 1.205253512570841e-05,
-      "loss": 5.798249912913889e-05,
+      "loss": 1.2870759746874683e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00006,
+      "ppl": 1.00001,
       "step": 466,
       "tokens/total": 14100192,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.93,
       "tokens/trainable": 213011
     },
     {
       "epoch": 1.82421875,
-      "grad_norm": 0.008271797560155392,
+      "grad_norm": 0.0005419626249931753,
       "learning_rate": 1.1966757525110255e-05,
-      "loss": 2.8763912268914282e-05,
+      "loss": 7.527931302320212e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 467,
       "tokens/total": 14130448,
-      "tokens/train_per_sec_per_gpu": 31.44,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 213450
     },
     {
       "epoch": 1.828125,
-      "grad_norm": 0.023479726165533066,
+      "grad_norm": 0.004500397015362978,
       "learning_rate": 1.1882770713386095e-05,
-      "loss": 0.0001434558507753536,
+      "loss": 4.2581184970913455e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00014,
+      "ppl": 1.00004,
       "step": 468,
       "tokens/total": 14160480,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.7,
       "tokens/trainable": 213898
     },
     {
       "epoch": 1.83203125,
-      "grad_norm": 0.0023077642545104027,
+      "grad_norm": 0.1103268712759018,
       "learning_rate": 1.180057818556998e-05,
-      "loss": 2.560112625360489e-05,
+      "loss": 0.00021096917043905705,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00021,
       "step": 469,
       "tokens/total": 14191152,
-      "tokens/train_per_sec_per_gpu": 36.78,
+      "tokens/train_per_sec_per_gpu": 36.88,
       "tokens/trainable": 214366
     },
     {
       "epoch": 1.8359375,
-      "grad_norm": 0.0014362182701006532,
+      "grad_norm": 0.002097723074257374,
       "learning_rate": 1.1720183362028494e-05,
-      "loss": 2.0195953766233288e-05,
+      "loss": 2.7896878236788325e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 470,
       "tokens/total": 14219392,
-      "tokens/train_per_sec_per_gpu": 35.64,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 214800
     },
     {
       "epoch": 1.83984375,
-      "grad_norm": 0.011598445475101471,
+      "grad_norm": 0.001367987017147243,
       "learning_rate": 1.1641589588318387e-05,
-      "loss": 4.486892794375308e-05,
+      "loss": 1.733974931994453e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 471,
       "tokens/total": 14249920,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 215292
     },
     {
       "epoch": 1.84375,
-      "grad_norm": 0.0011633927933871746,
+      "grad_norm": 0.000857445178553462,
       "learning_rate": 1.1564800135047418e-05,
-      "loss": 2.0482253603404388e-05,
+      "loss": 1.3961070180812385e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 472,
       "tokens/total": 14280048,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 215741
     },
     {
       "epoch": 1.84765625,
-      "grad_norm": 0.006093664560467005,
+      "grad_norm": 0.0037669208832085133,
       "learning_rate": 1.148981819773816e-05,
-      "loss": 5.0086440751329064e-05,
+      "loss": 3.097699664067477e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 473,
       "tokens/total": 14310432,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 216194
     },
     {
       "epoch": 1.8515625,
-      "grad_norm": 0.003623088588938117,
+      "grad_norm": 0.0003345425648149103,
       "learning_rate": 1.1416646896695086e-05,
-      "loss": 3.059494338231161e-05,
+      "loss": 6.7712135205511e-06,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 474,
       "tokens/total": 14340496,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 216634
     },
     {
       "epoch": 1.85546875,
-      "grad_norm": 0.006924150045961142,
+      "grad_norm": 0.00038110429886728525,
       "learning_rate": 1.1345289276874717e-05,
-      "loss": 3.953809573431499e-05,
+      "loss": 7.348439794441219e-06,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 475,
       "tokens/total": 14370720,
-      "tokens/train_per_sec_per_gpu": 30.48,
+      "tokens/train_per_sec_per_gpu": 30.53,
       "tokens/trainable": 217040
     },
     {
       "epoch": 1.859375,
-      "grad_norm": 0.002359601203352213,
+      "grad_norm": 0.0013931745197623968,
       "learning_rate": 1.1275748307758873e-05,
-      "loss": 3.34192045556847e-05,
+      "loss": 1.2693172720901202e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 476,
       "tokens/total": 14401136,
-      "tokens/train_per_sec_per_gpu": 37.55,
+      "tokens/train_per_sec_per_gpu": 37.67,
       "tokens/trainable": 217530
     },
     {
       "epoch": 1.86328125,
-      "grad_norm": 0.009584639221429825,
+      "grad_norm": 0.02564929611980915,
       "learning_rate": 1.1208026883231147e-05,
-      "loss": 0.00010110227594850585,
+      "loss": 0.00031186698470264673,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 477,
       "tokens/total": 14431360,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 217988
     },
     {
       "epoch": 1.8671875,
-      "grad_norm": 0.03044246695935726,
+      "grad_norm": 0.08053945004940033,
       "learning_rate": 1.1142127821456433e-05,
-      "loss": 0.00022086883836891502,
+      "loss": 0.0003999105538241565,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00022,
+      "ppl": 1.0004,
       "step": 478,
       "tokens/total": 14461968,
-      "tokens/train_per_sec_per_gpu": 32.47,
+      "tokens/train_per_sec_per_gpu": 32.54,
       "tokens/trainable": 218461
     },
     {
       "epoch": 1.87109375,
-      "grad_norm": 0.03494428098201752,
+      "grad_norm": 0.05217274650931358,
       "learning_rate": 1.1078053864763674e-05,
-      "loss": 0.0002753190929070115,
+      "loss": 0.0003339847025927156,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00028,
+      "ppl": 1.00033,
       "step": 479,
       "tokens/total": 14491984,
-      "tokens/train_per_sec_per_gpu": 32.77,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 218909
     },
     {
       "epoch": 1.875,
-      "grad_norm": 0.30102044343948364,
+      "grad_norm": 0.08783794194459915,
       "learning_rate": 1.1015807679531756e-05,
-      "loss": 0.00025764034944586456,
+      "loss": 0.0005760936765000224,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00026,
+      "ppl": 1.00058,
       "step": 480,
       "tokens/total": 14522544,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 34.08,
       "tokens/trainable": 219350
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_model.safetensors
index 95c3f19206395238e6782ddf02f5c2620f03d2b2..598edbb87c4b1911f292dd47ccb4a713b6478aed 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ba49d604afb025cda90bf22fe38858ddb91aef7695be1af2622e6a002b0e65b
+oid sha256:233181cf98d9388b75d25740fa98f4912414f643945abe781e342921d3103c33
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/optimizer.pt
index 7834e63f02e3f3e3356d07f9a5415fc57d2c483d..40b04158b65dd3e459b2bc823171c19186f2f945 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02947e98128d87243b855caa3aa6566a71488c0ba3ae7862ecd65bef0274d94b
+oid sha256:16c3da8efe9c136763c30af87a2769a1e809cdc1eca270b3b513e5a6efa6cba8
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/trainer_state.json
index adaad32362f2cb34f3c0402751f7bcb55cbadf88..8e8da379c0a212aa9e6075188d01b954b6eb67fd 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-512/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1525 +3735,1525 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.0037202269304543734,
+      "grad_norm": 0.024732626974582672,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 7.238816760946065e-05,
+      "loss": 0.0001429672265658155,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 353,
       "tokens/total": 10678080,
-      "tokens/train_per_sec_per_gpu": 28.86,
+      "tokens/train_per_sec_per_gpu": 28.93,
       "tokens/trainable": 161555
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0021510994993150234,
+      "grad_norm": 0.037309419363737106,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 3.6805788113269955e-05,
+      "loss": 7.214388460852206e-05,
       "memory/device_reserved (GiB)": 35.41,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 354,
       "tokens/total": 10708544,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.75,
       "tokens/trainable": 162020
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.002642499515786767,
+      "grad_norm": 0.00885045062750578,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 4.371708200778812e-05,
+      "loss": 0.00011830432777060196,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 355,
       "tokens/total": 10739024,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 162448
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0038646068423986435,
+      "grad_norm": 0.06791900098323822,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 7.11614266037941e-05,
+      "loss": 0.0003609730047173798,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.61,
       "memory/max_allocated (GiB)": 33.61,
-      "ppl": 1.00007,
+      "ppl": 1.00036,
       "step": 356,
       "tokens/total": 10769024,
-      "tokens/train_per_sec_per_gpu": 28.2,
+      "tokens/train_per_sec_per_gpu": 28.26,
       "tokens/trainable": 162863
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.017895536497235298,
+      "grad_norm": 0.0019508687546476722,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 5.4065520089352503e-05,
+      "loss": 3.1739040423417464e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 357,
       "tokens/total": 10799328,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 163318
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.8479411602020264,
+      "grad_norm": 0.7139555811882019,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.01497839204967022,
+      "loss": 0.010268578305840492,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01509,
+      "ppl": 1.01032,
       "step": 358,
       "tokens/total": 10829488,
-      "tokens/train_per_sec_per_gpu": 32.89,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 163767
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.0015704578254371881,
+      "grad_norm": 0.006852464284747839,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 3.0390210667974316e-05,
+      "loss": 4.873241778113879e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 359,
       "tokens/total": 10859936,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 164208
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.014669318683445454,
+      "grad_norm": 0.0018247900297865272,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010030368866864592,
+      "loss": 3.88835760531947e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00004,
       "step": 360,
       "tokens/total": 10890400,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 164655
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.3479421138763428,
+      "grad_norm": 0.0436873696744442,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 0.0026096655055880547,
+      "loss": 0.0003687943681143224,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00261,
+      "ppl": 1.00037,
       "step": 361,
       "tokens/total": 10920624,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 165100
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.004394443240016699,
+      "grad_norm": 0.26735150814056396,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 6.116658914834261e-05,
+      "loss": 0.00039605735219083726,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00006,
+      "ppl": 1.0004,
       "step": 362,
       "tokens/total": 10951136,
-      "tokens/train_per_sec_per_gpu": 34.84,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 165587
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.016197621822357178,
+      "grad_norm": 0.0964483991265297,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0001775357231963426,
+      "loss": 0.0007330900407396257,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00018,
+      "ppl": 1.00073,
       "step": 363,
       "tokens/total": 10981344,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 166034
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.09478334337472916,
+      "grad_norm": 0.05367618799209595,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.0005033796769566834,
+      "loss": 0.0005384897813200951,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0005,
+      "ppl": 1.00054,
       "step": 364,
       "tokens/total": 11011552,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 166485
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.014899010770022869,
+      "grad_norm": 0.10067912936210632,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 0.0001904651871882379,
+      "loss": 0.0006868956843391061,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00019,
+      "ppl": 1.00069,
       "step": 365,
       "tokens/total": 11041872,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 166973
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005555527750402689,
+      "grad_norm": 0.02697882056236267,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 0.00010058133921120316,
+      "loss": 0.0002464794088155031,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00025,
       "step": 366,
       "tokens/total": 11072192,
-      "tokens/train_per_sec_per_gpu": 29.1,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 167414
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.20706292986869812,
+      "grad_norm": 0.0026214183308184147,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 0.0005888720043003559,
+      "loss": 5.2629769925260916e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00059,
+      "ppl": 1.00005,
       "step": 367,
       "tokens/total": 11102736,
-      "tokens/train_per_sec_per_gpu": 30.11,
+      "tokens/train_per_sec_per_gpu": 30.23,
       "tokens/trainable": 167851
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.0007191727054305375,
+      "grad_norm": 0.0018140808679163456,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 1.2304372830840293e-05,
+      "loss": 2.9512597393477336e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 368,
       "tokens/total": 11133056,
-      "tokens/train_per_sec_per_gpu": 35.48,
+      "tokens/train_per_sec_per_gpu": 35.56,
       "tokens/trainable": 168294
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.030878128483891487,
+      "grad_norm": 0.012045321986079216,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 0.0002043281274382025,
+      "loss": 7.606908184243366e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0002,
+      "ppl": 1.00008,
       "step": 369,
       "tokens/total": 11163568,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 168769
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.008817057125270367,
+      "grad_norm": 0.013453424908220768,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 0.00011464582348708063,
+      "loss": 0.00015325279673561454,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00015,
       "step": 370,
       "tokens/total": 11193936,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.83,
       "tokens/trainable": 169270
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.023856336250901222,
+      "grad_norm": 0.003458227962255478,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 0.00013431125262286514,
+      "loss": 5.63332432648167e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00013,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11224480,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 169711
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.010502993129193783,
+      "grad_norm": 0.0063622924499213696,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 0.00015104333579074591,
+      "loss": 9.534892160445452e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00015,
+      "ppl": 1.0001,
       "step": 372,
       "tokens/total": 11254912,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 170185
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.026970423758029938,
+      "grad_norm": 0.04558980092406273,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.00026081278338097036,
+      "loss": 0.000613630109000951,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00026,
+      "ppl": 1.00061,
       "step": 373,
       "tokens/total": 11285104,
-      "tokens/train_per_sec_per_gpu": 33.08,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 170650
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.020325109362602234,
+      "grad_norm": 0.014873786829411983,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.00014532633940689266,
+      "loss": 0.0002038514503510669,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 374,
       "tokens/total": 11315600,
-      "tokens/train_per_sec_per_gpu": 31.98,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 171126
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.033260464668273926,
+      "grad_norm": 0.058739252388477325,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 0.00045571548980660737,
+      "loss": 0.0010346119524911046,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00046,
+      "ppl": 1.00104,
       "step": 375,
       "tokens/total": 11345776,
       "tokens/train_per_sec_per_gpu": 35.6,
@@ -5261,713 +5261,713 @@
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.017290709540247917,
+      "grad_norm": 0.24105991423130035,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 0.00014910403115209192,
+      "loss": 0.0035716122947633266,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00015,
+      "ppl": 1.00358,
       "step": 376,
       "tokens/total": 11376208,
-      "tokens/train_per_sec_per_gpu": 28.09,
+      "tokens/train_per_sec_per_gpu": 28.14,
       "tokens/trainable": 172017
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.04457363486289978,
+      "grad_norm": 0.08793950080871582,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 6.834026135038584e-05,
+      "loss": 0.0007833336712792516,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00007,
+      "ppl": 1.00078,
       "step": 377,
       "tokens/total": 11406512,
-      "tokens/train_per_sec_per_gpu": 35.99,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 172493
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.002268735785037279,
+      "grad_norm": 0.009154544211924076,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 3.245133120799437e-05,
+      "loss": 0.00011752049613278359,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00012,
       "step": 378,
       "tokens/total": 11436544,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 172951
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.0038171466439962387,
+      "grad_norm": 0.012918495573103428,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 5.4866883147042245e-05,
+      "loss": 0.00011567945330170915,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00012,
       "step": 379,
       "tokens/total": 11467008,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 173435
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005268535576760769,
+      "grad_norm": 0.008966504596173763,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 8.241234172601253e-05,
+      "loss": 8.938623068388551e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00009,
       "step": 380,
       "tokens/total": 11497184,
-      "tokens/train_per_sec_per_gpu": 30.4,
+      "tokens/train_per_sec_per_gpu": 30.47,
       "tokens/trainable": 173900
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.016057243570685387,
+      "grad_norm": 0.03181751072406769,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 0.00014235377602744848,
+      "loss": 0.00013083787052892148,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00013,
       "step": 381,
       "tokens/total": 11527600,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 174371
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005853456910699606,
+      "grad_norm": 0.0009072935208678246,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 5.001476893085055e-05,
+      "loss": 2.5600436856620945e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 382,
       "tokens/total": 11557808,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 174826
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0031664848793298006,
+      "grad_norm": 0.0050596860237419605,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 4.210277256788686e-05,
+      "loss": 5.778766353614628e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 383,
       "tokens/total": 11588352,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 175276
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.08838633447885513,
+      "grad_norm": 0.008304890245199203,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 0.0010394920827820897,
+      "loss": 0.00012505470658652484,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00104,
+      "ppl": 1.00013,
       "step": 384,
       "tokens/total": 11618640,
-      "tokens/train_per_sec_per_gpu": 33.69,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 175720
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.09446703642606735,
+      "grad_norm": 0.0035276354756206274,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0005370234721340239,
+      "loss": 4.851898120250553e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00054,
+      "ppl": 1.00005,
       "step": 385,
       "tokens/total": 11649072,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 176203
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.0013355027185752988,
+      "grad_norm": 0.001593019813299179,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 1.903088195831515e-05,
+      "loss": 3.330651088617742e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 386,
       "tokens/total": 11679552,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176651
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.011693151667714119,
+      "grad_norm": 0.002167076338082552,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 7.680666749365628e-05,
+      "loss": 3.344817378092557e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 387,
       "tokens/total": 11709760,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 177115
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.09523260593414307,
+      "grad_norm": 0.003447546623647213,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 0.00073521543527022,
+      "loss": 5.5277254432439804e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00074,
+      "ppl": 1.00006,
       "step": 388,
       "tokens/total": 11740288,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.13,
       "tokens/trainable": 177601
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.005971741396933794,
+      "grad_norm": 0.0052979388274252415,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 6.468063656939194e-05,
+      "loss": 4.56162124464754e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00006,
+      "ppl": 1.00005,
       "step": 389,
       "tokens/total": 11770592,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 178050
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.0010247502941638231,
+      "grad_norm": 0.003032457083463669,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 1.5989648090908304e-05,
+      "loss": 3.829343768302351e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 390,
       "tokens/total": 11800752,
-      "tokens/train_per_sec_per_gpu": 36.38,
+      "tokens/train_per_sec_per_gpu": 36.42,
       "tokens/trainable": 178531
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.054420940577983856,
+      "grad_norm": 0.03542603179812431,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.0004186803416814655,
+      "loss": 0.0003605492820497602,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00042,
+      "ppl": 1.00036,
       "step": 391,
       "tokens/total": 11831360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 179000
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.002925195964053273,
+      "grad_norm": 0.004965933505445719,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 1.531536145193968e-05,
+      "loss": 7.091004226822406e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 392,
       "tokens/total": 11861552,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 179457
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 1.6402941942214966,
+      "grad_norm": 0.006339129991829395,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0027035027742385864,
+      "loss": 9.15761775104329e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00271,
+      "ppl": 1.00009,
       "step": 393,
       "tokens/total": 11891904,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 179890
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.0032207504846155643,
+      "grad_norm": 0.03921438008546829,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 3.968120290664956e-05,
+      "loss": 0.00038214243249967694,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00038,
       "step": 394,
       "tokens/total": 11922208,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 180341
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.013098486699163914,
+      "grad_norm": 0.08640608936548233,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.00011063168494729325,
+      "loss": 0.0006777399685233831,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00011,
+      "ppl": 1.00068,
       "step": 395,
       "tokens/total": 11952672,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 180762
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.01553372759371996,
+      "grad_norm": 0.08661718666553497,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 8.802580123301595e-05,
+      "loss": 0.0007016840972937644,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00009,
+      "ppl": 1.0007,
       "step": 396,
       "tokens/total": 11983088,
-      "tokens/train_per_sec_per_gpu": 35.04,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 181226
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.016083814203739166,
+      "grad_norm": 0.010069145821034908,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 8.931377669796348e-05,
+      "loss": 0.00011120665294583887,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00011,
       "step": 397,
       "tokens/total": 12013488,
-      "tokens/train_per_sec_per_gpu": 33.15,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 181714
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.05752059072256088,
+      "grad_norm": 0.03158818930387497,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.00015369296306744218,
+      "loss": 0.00024857826065272093,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00015,
+      "ppl": 1.00025,
       "step": 398,
       "tokens/total": 12043968,
-      "tokens/train_per_sec_per_gpu": 35.52,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 182142
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.0011741623748093843,
+      "grad_norm": 0.000873154669534415,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 1.8113165424438193e-05,
+      "loss": 2.1218824258539826e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00002,
       "step": 399,
       "tokens/total": 12074160,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.86,
       "tokens/trainable": 182604
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0048082731664180756,
+      "grad_norm": 0.009255572222173214,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 5.3348740038927644e-05,
+      "loss": 0.0001349164522252977,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00005,
+      "ppl": 1.00013,
       "step": 400,
       "tokens/total": 12104352,
-      "tokens/train_per_sec_per_gpu": 25.28,
+      "tokens/train_per_sec_per_gpu": 25.31,
       "tokens/trainable": 183011
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.006520449183881283,
+      "grad_norm": 0.010521743446588516,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 6.45049221930094e-05,
+      "loss": 7.793466647854075e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00006,
+      "ppl": 1.00008,
       "step": 401,
       "tokens/total": 12134240,
-      "tokens/train_per_sec_per_gpu": 31.69,
+      "tokens/train_per_sec_per_gpu": 31.81,
       "tokens/trainable": 183459
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.006227654870599508,
+      "grad_norm": 0.0013409090461209416,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 6.928759830771014e-05,
+      "loss": 2.7722922823159024e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 402,
       "tokens/total": 12164480,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.7,
       "tokens/trainable": 183959
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.32605040073394775,
+      "grad_norm": 0.10361713916063309,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.002115404698997736,
+      "loss": 0.0014142843428999186,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00212,
+      "ppl": 1.00142,
       "step": 403,
       "tokens/total": 12194640,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.5,
       "tokens/trainable": 184431
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.007742208894342184,
+      "grad_norm": 0.0026048943400382996,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 4.600908869178966e-05,
+      "loss": 4.14503738284111e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 404,
       "tokens/total": 12224960,
-      "tokens/train_per_sec_per_gpu": 31.34,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 184867
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.0049946485087275505,
+      "grad_norm": 0.009411919862031937,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 5.003535625291988e-05,
+      "loss": 7.322158489841968e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00005,
+      "ppl": 1.00007,
       "step": 405,
       "tokens/total": 12255008,
-      "tokens/train_per_sec_per_gpu": 29.25,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 185300
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.004598768427968025,
+      "grad_norm": 0.006792420521378517,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 3.4423381293891e-05,
+      "loss": 7.982828537933528e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00003,
+      "ppl": 1.00008,
       "step": 406,
       "tokens/total": 12285248,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 185754
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.0011464629787951708,
+      "grad_norm": 0.013020367361605167,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 1.9196500943508e-05,
+      "loss": 3.743396155186929e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 407,
       "tokens/total": 12315552,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 186213
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.004688040353357792,
+      "grad_norm": 0.0013089004205539823,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 5.2492636314127594e-05,
+      "loss": 2.306591341039166e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00002,
       "step": 408,
       "tokens/total": 12345904,
-      "tokens/train_per_sec_per_gpu": 32.72,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 186659
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.002637861529365182,
+      "grad_norm": 0.0012270932784304023,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 3.172009019181132e-05,
+      "loss": 2.1350417227949947e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 409,
       "tokens/total": 12376336,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.97,
       "tokens/trainable": 187083
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.0072015393525362015,
+      "grad_norm": 0.011230082251131535,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 5.452537880046293e-05,
+      "loss": 3.752015618374571e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 410,
       "tokens/total": 12406880,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 187530
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.007169651333242655,
+      "grad_norm": 0.004872993566095829,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 3.530656249495223e-05,
+      "loss": 5.5653974413871765e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 411,
       "tokens/total": 12437088,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 187972
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003889538114890456,
+      "grad_norm": 0.19391997158527374,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 3.5087461583316326e-05,
+      "loss": 0.000892186420969665,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00089,
       "step": 412,
       "tokens/total": 12465440,
-      "tokens/train_per_sec_per_gpu": 39.89,
+      "tokens/train_per_sec_per_gpu": 39.95,
       "tokens/trainable": 188417
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.5135430693626404,
+      "grad_norm": 0.09062197804450989,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.005697700660675764,
+      "loss": 0.0006304415874183178,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00571,
+      "ppl": 1.00063,
       "step": 413,
       "tokens/total": 12495968,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.57,
       "tokens/trainable": 188868
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.0010096468031406403,
+      "grad_norm": 0.009724686853587627,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 1.1318426913931035e-05,
+      "loss": 6.72380265314132e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 414,
       "tokens/total": 12526128,
-      "tokens/train_per_sec_per_gpu": 36.85,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 189337
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0004970752634108067,
+      "grad_norm": 0.19542834162712097,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 7.355167326750234e-06,
+      "loss": 0.001907092402689159,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00191,
       "step": 415,
       "tokens/total": 12556256,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 189750
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.0008849130244925618,
+      "grad_norm": 0.0465276800096035,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 1.4194254617905244e-05,
+      "loss": 0.0005661146715283394,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00057,
       "step": 416,
       "tokens/total": 12586736,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 190216
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.0022039280738681555,
+      "grad_norm": 0.0012954205740243196,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 2.4539594960515387e-05,
+      "loss": 1.8871982319978997e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
       "ppl": 1.00002,
       "step": 417,
       "tokens/total": 12617168,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 190708
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.027006179094314575,
+      "grad_norm": 0.0018749319715425372,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 0.00020442574168555439,
+      "loss": 3.220669532311149e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 418,
       "tokens/total": 12647680,
-      "tokens/train_per_sec_per_gpu": 30.85,
+      "tokens/train_per_sec_per_gpu": 30.9,
       "tokens/trainable": 191126
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.15684254467487335,
+      "grad_norm": 0.0022480092011392117,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 0.0007251293282024562,
+      "loss": 2.6564141080598347e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00073,
+      "ppl": 1.00003,
       "step": 419,
       "tokens/total": 12677952,
-      "tokens/train_per_sec_per_gpu": 30.3,
+      "tokens/train_per_sec_per_gpu": 30.34,
       "tokens/trainable": 191540
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.0033104538451880217,
+      "grad_norm": 0.1634059101343155,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 3.131272751488723e-05,
+      "loss": 0.0009822181891649961,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00098,
       "step": 420,
       "tokens/total": 12708416,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.69,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.001255685230717063,
+      "grad_norm": 0.0019404747290536761,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 1.9309976778458804e-05,
+      "loss": 2.167341335734818e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 421,
       "tokens/total": 12738640,
-      "tokens/train_per_sec_per_gpu": 31.28,
+      "tokens/train_per_sec_per_gpu": 31.31,
       "tokens/trainable": 192463
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.01599641516804695,
+      "grad_norm": 0.001059795613400638,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 9.235648030880839e-05,
+      "loss": 1.6543437595828436e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12769232,
-      "tokens/train_per_sec_per_gpu": 32.71,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 192918
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.048507146537303925,
+      "grad_norm": 0.01760418340563774,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 0.0002708940301090479,
+      "loss": 0.0001707405026536435,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00027,
+      "ppl": 1.00017,
       "step": 423,
       "tokens/total": 12799648,
-      "tokens/train_per_sec_per_gpu": 37.35,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 193425
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.010107563808560371,
+      "grad_norm": 0.0023663989268243313,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 6.99054216966033e-05,
+      "loss": 2.332203439436853e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 424,
       "tokens/total": 12829920,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 193898
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.005467226263135672,
+      "grad_norm": 0.0245437640696764,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 5.417566717369482e-05,
+      "loss": 0.0001577093207743019,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00005,
+      "ppl": 1.00016,
       "step": 425,
       "tokens/total": 12860112,
-      "tokens/train_per_sec_per_gpu": 29.24,
+      "tokens/train_per_sec_per_gpu": 29.29,
       "tokens/trainable": 194315
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.0015799012035131454,
+      "grad_norm": 0.0009002169244922698,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 2.4984849005704746e-05,
+      "loss": 1.4283305972639937e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 426,
       "tokens/total": 12890320,
       "tokens/train_per_sec_per_gpu": 34.92,
@@ -5975,779 +5975,779 @@
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.024157166481018066,
+      "grad_norm": 0.00033997284481301904,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 0.00010696032404666767,
+      "loss": 7.185776667029131e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00001,
       "step": 427,
       "tokens/total": 12920656,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 195262
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.00028090659179724753,
+      "grad_norm": 0.0005359657225199044,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 7.484430170734413e-06,
+      "loss": 9.948088518285658e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
       "ppl": 1.00001,
       "step": 428,
       "tokens/total": 12951040,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 195680
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.0011094068177044392,
+      "grad_norm": 0.0014890795573592186,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 1.9920153135899454e-05,
+      "loss": 2.028812377830036e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00002,
       "step": 429,
       "tokens/total": 12981232,
-      "tokens/train_per_sec_per_gpu": 32.83,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 196145
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.020794112235307693,
+      "grad_norm": 0.05130980163812637,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 0.00010971157462336123,
+      "loss": 0.0003292672336101532,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00033,
       "step": 430,
       "tokens/total": 13011616,
-      "tokens/train_per_sec_per_gpu": 32.39,
+      "tokens/train_per_sec_per_gpu": 32.42,
       "tokens/trainable": 196626
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.44080695509910583,
+      "grad_norm": 0.0014070137403905392,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0026788683608174324,
+      "loss": 1.0152909453609027e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.00268,
+      "ppl": 1.00001,
       "step": 431,
       "tokens/total": 13042208,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.17,
       "tokens/trainable": 197067
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0026339287869632244,
+      "grad_norm": 0.004116157069802284,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 2.082335777231492e-05,
+      "loss": 3.514952550176531e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 432,
       "tokens/total": 13072800,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.68,
       "tokens/trainable": 197565
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.004508621525019407,
+      "grad_norm": 0.01147819496691227,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 1.439991501683835e-05,
+      "loss": 7.313965761568397e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 433,
       "tokens/total": 13103328,
-      "tokens/train_per_sec_per_gpu": 32.88,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 198015
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.014811511151492596,
+      "grad_norm": 0.030784178525209427,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 0.00015316363715101033,
+      "loss": 0.0002448821614962071,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00024,
       "step": 434,
       "tokens/total": 13133472,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 198442
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.0037793368101119995,
+      "grad_norm": 0.0022457120940089226,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 1.4444960470427759e-05,
+      "loss": 1.4446297427639365e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 435,
       "tokens/total": 13164064,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 198914
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.24972176551818848,
+      "grad_norm": 0.0029742212500423193,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0017114672809839249,
+      "loss": 2.7415488148108125e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00171,
+      "ppl": 1.00003,
       "step": 436,
       "tokens/total": 13194464,
-      "tokens/train_per_sec_per_gpu": 33.35,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 199355
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.002479988383129239,
+      "grad_norm": 0.002260624896734953,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.9400333005469292e-05,
+      "loss": 2.7733602109947242e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00003,
       "step": 437,
       "tokens/total": 13224992,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 199830
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.0014780610799789429,
+      "grad_norm": 0.01198511105030775,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 1.5593774151057005e-05,
+      "loss": 9.154126746580005e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00009,
       "step": 438,
       "tokens/total": 13255296,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 200338
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.0010410621762275696,
+      "grad_norm": 0.003149093361571431,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 1.4683226254419424e-05,
+      "loss": 3.515004937071353e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13285344,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 200785
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0004247442411724478,
+      "grad_norm": 0.0003243185637984425,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 8.09474295238033e-06,
+      "loss": 7.168858701334102e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 440,
       "tokens/total": 13315632,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.32,
       "tokens/trainable": 201267
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0026301892939954996,
+      "grad_norm": 0.12771126627922058,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 1.7428235878469422e-05,
+      "loss": 0.0007142049144022167,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 441,
       "tokens/total": 13346048,
-      "tokens/train_per_sec_per_gpu": 34.16,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 201734
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0005576284602284431,
+      "grad_norm": 0.11483822762966156,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 8.656044883537106e-06,
+      "loss": 0.0006703532999381423,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00001,
+      "ppl": 1.00067,
       "step": 442,
       "tokens/total": 13376304,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.4,
       "tokens/trainable": 202174
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0009604030638001859,
+      "grad_norm": 0.00032006221590563655,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 1.4092523088038433e-05,
+      "loss": 9.367744496557862e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00001,
       "step": 443,
       "tokens/total": 13406720,
-      "tokens/train_per_sec_per_gpu": 36.13,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 202612
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.00019884438370354474,
+      "grad_norm": 0.0002659875026438385,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 4.5837323341402225e-06,
+      "loss": 5.530210728466045e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 444,
       "tokens/total": 13436464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 30.99,
       "tokens/trainable": 203020
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.05944995582103729,
+      "grad_norm": 0.04957196116447449,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 0.0002780454815365374,
+      "loss": 6.46735934424214e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00028,
+      "ppl": 1.00006,
       "step": 445,
       "tokens/total": 13466672,
-      "tokens/train_per_sec_per_gpu": 34.31,
+      "tokens/train_per_sec_per_gpu": 34.37,
       "tokens/trainable": 203479
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.005872223526239395,
+      "grad_norm": 0.0016840791795402765,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 3.424299575272016e-05,
+      "loss": 1.2613029866770376e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 446,
       "tokens/total": 13496736,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 203946
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.006013527978211641,
+      "grad_norm": 0.00042921333806589246,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 3.269856097176671e-05,
+      "loss": 7.416386324621271e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 447,
       "tokens/total": 13526928,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 204382
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.05103042349219322,
+      "grad_norm": 0.2165915071964264,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.00032247230410575867,
+      "loss": 0.0006914341356605291,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00032,
+      "ppl": 1.00069,
       "step": 448,
       "tokens/total": 13557392,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 204830
     },
     {
       "epoch": 1.75390625,
-      "grad_norm": 0.13250835239887238,
+      "grad_norm": 0.0009792562341317534,
       "learning_rate": 1.3780999708818058e-05,
-      "loss": 0.0008940898114815354,
+      "loss": 1.4504414139082655e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00089,
+      "ppl": 1.00001,
       "step": 449,
       "tokens/total": 13587776,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.47,
       "tokens/trainable": 205300
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 0.24642030894756317,
+      "grad_norm": 0.028498223051428795,
       "learning_rate": 1.3665385037857758e-05,
-      "loss": 0.0012931979726999998,
+      "loss": 0.00028163049137219787,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00129,
+      "ppl": 1.00028,
       "step": 450,
       "tokens/total": 13618112,
-      "tokens/train_per_sec_per_gpu": 30.62,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 205726
     },
     {
       "epoch": 1.76171875,
-      "grad_norm": 0.011528799310326576,
+      "grad_norm": 0.003279345342889428,
       "learning_rate": 1.3551490468947126e-05,
-      "loss": 6.569598917849362e-05,
+      "loss": 2.595262776594609e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 451,
       "tokens/total": 13648512,
-      "tokens/train_per_sec_per_gpu": 31.55,
+      "tokens/train_per_sec_per_gpu": 31.63,
       "tokens/trainable": 206163
     },
     {
       "epoch": 1.765625,
-      "grad_norm": 1.0818549394607544,
+      "grad_norm": 0.4961508512496948,
       "learning_rate": 1.3439320741704075e-05,
-      "loss": 0.015596212819218636,
+      "loss": 0.00459528062492609,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01572,
+      "ppl": 1.00461,
       "step": 452,
       "tokens/total": 13678704,
-      "tokens/train_per_sec_per_gpu": 35.41,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 206619
     },
     {
       "epoch": 1.76953125,
-      "grad_norm": 0.0007326967315748334,
+      "grad_norm": 0.009086468257009983,
       "learning_rate": 1.3328880523968808e-05,
-      "loss": 1.2033770872221794e-05,
+      "loss": 6.577694148290902e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 453,
       "tokens/total": 13709232,
-      "tokens/train_per_sec_per_gpu": 37.18,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 207101
     },
     {
       "epoch": 1.7734375,
-      "grad_norm": 0.0023919539526104927,
+      "grad_norm": 0.009036197327077389,
       "learning_rate": 1.3220174411609587e-05,
-      "loss": 1.262133719137637e-05,
+      "loss": 6.735372880939394e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 454,
       "tokens/total": 13739600,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.77,
       "tokens/trainable": 207546
     },
     {
       "epoch": 1.77734375,
-      "grad_norm": 0.0020972786005586386,
+      "grad_norm": 0.009665202349424362,
       "learning_rate": 1.3113206928331471e-05,
-      "loss": 2.476977533660829e-05,
+      "loss": 7.266137981787324e-05,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 455,
       "tokens/total": 13769936,
-      "tokens/train_per_sec_per_gpu": 34.25,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 207989
     },
     {
       "epoch": 1.78125,
-      "grad_norm": 0.000797569751739502,
+      "grad_norm": 0.0004666070453822613,
       "learning_rate": 1.300798252548806e-05,
-      "loss": 1.45716385304695e-05,
+      "loss": 6.039275831426494e-06,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00001,
       "step": 456,
       "tokens/total": 13800240,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 208402
     },
     {
       "epoch": 1.78515625,
-      "grad_norm": 0.011376727372407913,
+      "grad_norm": 0.00044853391591459513,
       "learning_rate": 1.2904505581896265e-05,
-      "loss": 7.903270306997001e-05,
+      "loss": 8.292890015582088e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 457,
       "tokens/total": 13830512,
-      "tokens/train_per_sec_per_gpu": 36.61,
+      "tokens/train_per_sec_per_gpu": 36.58,
       "tokens/trainable": 208904
     },
     {
       "epoch": 1.7890625,
-      "grad_norm": 0.0013566205743700266,
+      "grad_norm": 0.0033692270517349243,
       "learning_rate": 1.2802780403654082e-05,
-      "loss": 1.925312972161919e-05,
+      "loss": 2.3157112082117237e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
       "ppl": 1.00002,
       "step": 458,
       "tokens/total": 13860832,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 209382
     },
     {
       "epoch": 1.79296875,
-      "grad_norm": 0.00047398527385666966,
+      "grad_norm": 0.0010797431459650397,
       "learning_rate": 1.2702811223961408e-05,
-      "loss": 1.119351145462133e-05,
+      "loss": 1.119279841077514e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00001,
       "step": 459,
       "tokens/total": 13890992,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 209833
     },
     {
       "epoch": 1.796875,
-      "grad_norm": 0.04677487164735794,
+      "grad_norm": 0.04658913239836693,
       "learning_rate": 1.2604602202943861e-05,
-      "loss": 0.0001750149531289935,
+      "loss": 0.0003375259111635387,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00018,
+      "ppl": 1.00034,
       "step": 460,
       "tokens/total": 13921424,
-      "tokens/train_per_sec_per_gpu": 32.5,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 210284
     },
     {
       "epoch": 1.80078125,
-      "grad_norm": 0.024663345888257027,
+      "grad_norm": 0.01848040148615837,
       "learning_rate": 1.2508157427479686e-05,
-      "loss": 9.222347580362111e-05,
+      "loss": 0.00016109315038193017,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00016,
       "step": 461,
       "tokens/total": 13951504,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.07,
       "tokens/trainable": 210769
     },
     {
       "epoch": 1.8046875,
-      "grad_norm": 0.0004740248841699213,
+      "grad_norm": 0.05658251419663429,
       "learning_rate": 1.2413480911029655e-05,
-      "loss": 1.0696679964894429e-05,
+      "loss": 0.00017825173563323915,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00001,
+      "ppl": 1.00018,
       "step": 462,
       "tokens/total": 13979728,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.41,
       "tokens/trainable": 211193
     },
     {
       "epoch": 1.80859375,
-      "grad_norm": 0.0007590787718072534,
+      "grad_norm": 0.0006594725418835878,
       "learning_rate": 1.2320576593470082e-05,
-      "loss": 1.3936740288045257e-05,
+      "loss": 9.709075129649136e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00001,
       "step": 463,
       "tokens/total": 14009936,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 211655
     },
     {
       "epoch": 1.8125,
-      "grad_norm": 0.002108624204993248,
+      "grad_norm": 0.02737571857869625,
       "learning_rate": 1.2229448340928828e-05,
-      "loss": 2.675112227734644e-05,
+      "loss": 0.00017574361118022352,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00018,
       "step": 464,
       "tokens/total": 14039872,
-      "tokens/train_per_sec_per_gpu": 29.28,
+      "tokens/train_per_sec_per_gpu": 29.2,
       "tokens/trainable": 212085
     },
     {
       "epoch": 1.81640625,
-      "grad_norm": 0.0009435649262741208,
+      "grad_norm": 0.0004917355254292488,
       "learning_rate": 1.2140099945624458e-05,
-      "loss": 1.0821668183780275e-05,
+      "loss": 8.16806459624786e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00001,
       "step": 465,
       "tokens/total": 14070096,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 212570
     },
     {
       "epoch": 1.8203125,
-      "grad_norm": 0.011040126904845238,
+      "grad_norm": 0.0015915038529783487,
       "learning_rate": 1.205253512570841e-05,
-      "loss": 5.798249912913889e-05,
+      "loss": 1.2870759746874683e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00006,
+      "ppl": 1.00001,
       "step": 466,
       "tokens/total": 14100192,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.93,
       "tokens/trainable": 213011
     },
     {
       "epoch": 1.82421875,
-      "grad_norm": 0.008271797560155392,
+      "grad_norm": 0.0005419626249931753,
       "learning_rate": 1.1966757525110255e-05,
-      "loss": 2.8763912268914282e-05,
+      "loss": 7.527931302320212e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 467,
       "tokens/total": 14130448,
-      "tokens/train_per_sec_per_gpu": 31.44,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 213450
     },
     {
       "epoch": 1.828125,
-      "grad_norm": 0.023479726165533066,
+      "grad_norm": 0.004500397015362978,
       "learning_rate": 1.1882770713386095e-05,
-      "loss": 0.0001434558507753536,
+      "loss": 4.2581184970913455e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00014,
+      "ppl": 1.00004,
       "step": 468,
       "tokens/total": 14160480,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.7,
       "tokens/trainable": 213898
     },
     {
       "epoch": 1.83203125,
-      "grad_norm": 0.0023077642545104027,
+      "grad_norm": 0.1103268712759018,
       "learning_rate": 1.180057818556998e-05,
-      "loss": 2.560112625360489e-05,
+      "loss": 0.00021096917043905705,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00021,
       "step": 469,
       "tokens/total": 14191152,
-      "tokens/train_per_sec_per_gpu": 36.78,
+      "tokens/train_per_sec_per_gpu": 36.88,
       "tokens/trainable": 214366
     },
     {
       "epoch": 1.8359375,
-      "grad_norm": 0.0014362182701006532,
+      "grad_norm": 0.002097723074257374,
       "learning_rate": 1.1720183362028494e-05,
-      "loss": 2.0195953766233288e-05,
+      "loss": 2.7896878236788325e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 470,
       "tokens/total": 14219392,
-      "tokens/train_per_sec_per_gpu": 35.64,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 214800
     },
     {
       "epoch": 1.83984375,
-      "grad_norm": 0.011598445475101471,
+      "grad_norm": 0.001367987017147243,
       "learning_rate": 1.1641589588318387e-05,
-      "loss": 4.486892794375308e-05,
+      "loss": 1.733974931994453e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 471,
       "tokens/total": 14249920,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 215292
     },
     {
       "epoch": 1.84375,
-      "grad_norm": 0.0011633927933871746,
+      "grad_norm": 0.000857445178553462,
       "learning_rate": 1.1564800135047418e-05,
-      "loss": 2.0482253603404388e-05,
+      "loss": 1.3961070180812385e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 472,
       "tokens/total": 14280048,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 215741
     },
     {
       "epoch": 1.84765625,
-      "grad_norm": 0.006093664560467005,
+      "grad_norm": 0.0037669208832085133,
       "learning_rate": 1.148981819773816e-05,
-      "loss": 5.0086440751329064e-05,
+      "loss": 3.097699664067477e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 473,
       "tokens/total": 14310432,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 216194
     },
     {
       "epoch": 1.8515625,
-      "grad_norm": 0.003623088588938117,
+      "grad_norm": 0.0003345425648149103,
       "learning_rate": 1.1416646896695086e-05,
-      "loss": 3.059494338231161e-05,
+      "loss": 6.7712135205511e-06,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 474,
       "tokens/total": 14340496,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 216634
     },
     {
       "epoch": 1.85546875,
-      "grad_norm": 0.006924150045961142,
+      "grad_norm": 0.00038110429886728525,
       "learning_rate": 1.1345289276874717e-05,
-      "loss": 3.953809573431499e-05,
+      "loss": 7.348439794441219e-06,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 475,
       "tokens/total": 14370720,
-      "tokens/train_per_sec_per_gpu": 30.48,
+      "tokens/train_per_sec_per_gpu": 30.53,
       "tokens/trainable": 217040
     },
     {
       "epoch": 1.859375,
-      "grad_norm": 0.002359601203352213,
+      "grad_norm": 0.0013931745197623968,
       "learning_rate": 1.1275748307758873e-05,
-      "loss": 3.34192045556847e-05,
+      "loss": 1.2693172720901202e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 476,
       "tokens/total": 14401136,
-      "tokens/train_per_sec_per_gpu": 37.55,
+      "tokens/train_per_sec_per_gpu": 37.67,
       "tokens/trainable": 217530
     },
     {
       "epoch": 1.86328125,
-      "grad_norm": 0.009584639221429825,
+      "grad_norm": 0.02564929611980915,
       "learning_rate": 1.1208026883231147e-05,
-      "loss": 0.00010110227594850585,
+      "loss": 0.00031186698470264673,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 477,
       "tokens/total": 14431360,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 217988
     },
     {
       "epoch": 1.8671875,
-      "grad_norm": 0.03044246695935726,
+      "grad_norm": 0.08053945004940033,
       "learning_rate": 1.1142127821456433e-05,
-      "loss": 0.00022086883836891502,
+      "loss": 0.0003999105538241565,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00022,
+      "ppl": 1.0004,
       "step": 478,
       "tokens/total": 14461968,
-      "tokens/train_per_sec_per_gpu": 32.47,
+      "tokens/train_per_sec_per_gpu": 32.54,
       "tokens/trainable": 218461
     },
     {
       "epoch": 1.87109375,
-      "grad_norm": 0.03494428098201752,
+      "grad_norm": 0.05217274650931358,
       "learning_rate": 1.1078053864763674e-05,
-      "loss": 0.0002753190929070115,
+      "loss": 0.0003339847025927156,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00028,
+      "ppl": 1.00033,
       "step": 479,
       "tokens/total": 14491984,
-      "tokens/train_per_sec_per_gpu": 32.77,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 218909
     },
     {
       "epoch": 1.875,
-      "grad_norm": 0.30102044343948364,
+      "grad_norm": 0.08783794194459915,
       "learning_rate": 1.1015807679531756e-05,
-      "loss": 0.00025764034944586456,
+      "loss": 0.0005760936765000224,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00026,
+      "ppl": 1.00058,
       "step": 480,
       "tokens/total": 14522544,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 34.08,
       "tokens/trainable": 219350
     },
     {
       "epoch": 1.87890625,
-      "grad_norm": 0.07505832612514496,
+      "grad_norm": 0.1059868335723877,
       "learning_rate": 1.0955391856078528e-05,
-      "loss": 0.0002503654104657471,
+      "loss": 0.0007811249233782291,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00025,
+      "ppl": 1.00078,
       "step": 481,
       "tokens/total": 14552560,
-      "tokens/train_per_sec_per_gpu": 29.47,
+      "tokens/train_per_sec_per_gpu": 29.52,
       "tokens/trainable": 219776
     },
     {
       "epoch": 1.8828125,
-      "grad_norm": 0.0008993322844617069,
+      "grad_norm": 0.00020755630976054817,
       "learning_rate": 1.0896808908553007e-05,
-      "loss": 1.0104758075613063e-05,
+      "loss": 5.210203653405188e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
@@ -6759,125 +6759,125 @@
     },
     {
       "epoch": 1.88671875,
-      "grad_norm": 0.001677343505434692,
+      "grad_norm": 0.00210341764613986,
       "learning_rate": 1.0840061274830763e-05,
-      "loss": 2.5296727471868508e-05,
+      "loss": 1.6855678040883504e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 483,
       "tokens/total": 14613152,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 220714
     },
     {
       "epoch": 1.890625,
-      "grad_norm": 0.01496347226202488,
+      "grad_norm": 0.0096653513610363,
       "learning_rate": 1.0785151316412473e-05,
-      "loss": 0.0001352597464574501,
+      "loss": 9.005956235341728e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00014,
+      "ppl": 1.00009,
       "step": 484,
       "tokens/total": 14643328,
-      "tokens/train_per_sec_per_gpu": 35.01,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 221156
     },
     {
       "epoch": 1.89453125,
-      "grad_norm": 0.004265300463885069,
+      "grad_norm": 0.001843759324401617,
       "learning_rate": 1.0732081318325639e-05,
-      "loss": 1.1048326996387914e-05,
+      "loss": 1.069177596946247e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
       "ppl": 1.00001,
       "step": 485,
       "tokens/total": 14673488,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.64,
       "tokens/trainable": 221605
     },
     {
       "epoch": 1.8984375,
-      "grad_norm": 0.0029735390562564135,
+      "grad_norm": 0.0007748372154310346,
       "learning_rate": 1.0680853489029501e-05,
-      "loss": 2.2016622096998617e-05,
+      "loss": 5.547179171117023e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 486,
       "tokens/total": 14704000,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.46,
       "tokens/trainable": 222042
     },
     {
       "epoch": 1.90234375,
-      "grad_norm": 0.00869164988398552,
+      "grad_norm": 0.00029233950772322714,
       "learning_rate": 1.0631469960323152e-05,
-      "loss": 7.451521378243342e-05,
+      "loss": 6.492456122941803e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 487,
       "tokens/total": 14733984,
-      "tokens/train_per_sec_per_gpu": 31.95,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 222474
     },
     {
       "epoch": 1.90625,
-      "grad_norm": 0.000581100583076477,
+      "grad_norm": 0.00032405118690803647,
       "learning_rate": 1.0583932787256783e-05,
-      "loss": 7.341123364312807e-06,
+      "loss": 6.497373760794289e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00001,
       "step": 488,
       "tokens/total": 14764400,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 222967
     },
     {
       "epoch": 1.91015625,
-      "grad_norm": 0.004224246367812157,
+      "grad_norm": 0.0019567832350730896,
       "learning_rate": 1.0538243948046206e-05,
-      "loss": 3.5561904951464385e-05,
+      "loss": 1.2859745766036212e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 489,
       "tokens/total": 14794672,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.79,
       "tokens/trainable": 223420
     },
     {
       "epoch": 1.9140625,
-      "grad_norm": 0.0152070177718997,
+      "grad_norm": 0.0006594944861717522,
       "learning_rate": 1.0494405343990523e-05,
-      "loss": 2.7988677175017074e-05,
+      "loss": 4.987327429262223e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00003,
+      "ppl": 1.0,
       "step": 490,
       "tokens/total": 14824784,
-      "tokens/train_per_sec_per_gpu": 29.42,
+      "tokens/train_per_sec_per_gpu": 29.4,
       "tokens/trainable": 223864
     },
     {
       "epoch": 1.91796875,
-      "grad_norm": 0.024472137913107872,
+      "grad_norm": 0.00579325295984745,
       "learning_rate": 1.0452418799392985e-05,
-      "loss": 4.8069461627164856e-05,
+      "loss": 3.640961949713528e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 491,
       "tokens/total": 14855056,
       "tokens/train_per_sec_per_gpu": 32.8,
@@ -6885,296 +6885,296 @@
     },
     {
       "epoch": 1.921875,
-      "grad_norm": 0.0036334518808871508,
+      "grad_norm": 0.010503454133868217,
       "learning_rate": 1.0412286061485102e-05,
-      "loss": 3.39774924213998e-05,
+      "loss": 9.240206418326125e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00003,
+      "ppl": 1.00009,
       "step": 492,
       "tokens/total": 14885392,
-      "tokens/train_per_sec_per_gpu": 36.29,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 224802
     },
     {
       "epoch": 1.92578125,
-      "grad_norm": 0.0009831058559939265,
+      "grad_norm": 0.0007255753153003752,
       "learning_rate": 1.03740088003539e-05,
-      "loss": 1.8870396161219105e-05,
+      "loss": 6.912261596880853e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 493,
       "tokens/total": 14915760,
-      "tokens/train_per_sec_per_gpu": 34.77,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 225284
     },
     {
       "epoch": 1.9296875,
-      "grad_norm": 0.0020358245819807053,
+      "grad_norm": 0.0015938309952616692,
       "learning_rate": 1.0337588608872463e-05,
-      "loss": 3.0332268579513766e-05,
+      "loss": 9.24063533602748e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 494,
       "tokens/total": 14946192,
-      "tokens/train_per_sec_per_gpu": 32.71,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 225733
     },
     {
       "epoch": 1.93359375,
-      "grad_norm": 0.00029036635532975197,
+      "grad_norm": 0.0005452019395306706,
       "learning_rate": 1.0303027002633622e-05,
-      "loss": 6.913277957210084e-06,
+      "loss": 6.261637736315606e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
       "ppl": 1.00001,
       "step": 495,
       "tokens/total": 14976272,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.58,
       "tokens/trainable": 226172
     },
     {
       "epoch": 1.9375,
-      "grad_norm": 0.11670258641242981,
+      "grad_norm": 0.0014865020057186484,
       "learning_rate": 1.0270325419886884e-05,
-      "loss": 0.0007051021093502641,
+      "loss": 1.6280355339404196e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00071,
+      "ppl": 1.00002,
       "step": 496,
       "tokens/total": 15006544,
-      "tokens/train_per_sec_per_gpu": 34.77,
+      "tokens/train_per_sec_per_gpu": 34.85,
       "tokens/trainable": 226636
     },
     {
       "epoch": 1.94140625,
-      "grad_norm": 0.0008452775655314326,
+      "grad_norm": 0.01585707999765873,
       "learning_rate": 1.0239485221478599e-05,
-      "loss": 1.0465846571605653e-05,
+      "loss": 8.379284554393962e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00008,
       "step": 497,
       "tokens/total": 15037040,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 227067
     },
     {
       "epoch": 1.9453125,
-      "grad_norm": 0.027690444141626358,
+      "grad_norm": 0.0002517803804948926,
       "learning_rate": 1.0210507690795292e-05,
-      "loss": 0.00013966832193545997,
+      "loss": 6.2335830079973675e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00001,
       "step": 498,
       "tokens/total": 15067344,
-      "tokens/train_per_sec_per_gpu": 37.58,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 227532
     },
     {
       "epoch": 1.94921875,
-      "grad_norm": 0.24434912204742432,
+      "grad_norm": 0.024532407522201538,
       "learning_rate": 1.0183394033710305e-05,
-      "loss": 0.0015483510214835405,
+      "loss": 0.00010188046144321561,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00155,
+      "ppl": 1.0001,
       "step": 499,
       "tokens/total": 15097648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 228007
     },
     {
       "epoch": 1.953125,
-      "grad_norm": 0.0016334869433194399,
+      "grad_norm": 0.011295626871287823,
       "learning_rate": 1.0158145378533583e-05,
-      "loss": 1.879910814750474e-05,
+      "loss": 3.307299630250782e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 500,
       "tokens/total": 15128000,
-      "tokens/train_per_sec_per_gpu": 30.84,
+      "tokens/train_per_sec_per_gpu": 30.89,
       "tokens/trainable": 228457
     },
     {
       "epoch": 1.95703125,
-      "grad_norm": 0.004102619830518961,
+      "grad_norm": 0.0004914791788905859,
       "learning_rate": 1.0134762775964726e-05,
-      "loss": 3.3614764106459916e-05,
+      "loss": 1.0351252058171667e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 501,
       "tokens/total": 15158352,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 228928
     },
     {
       "epoch": 1.9609375,
-      "grad_norm": 0.00031376321567222476,
+      "grad_norm": 0.0012023868039250374,
       "learning_rate": 1.0113247199049278e-05,
-      "loss": 6.558944733114913e-06,
+      "loss": 9.377205060445704e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
       "ppl": 1.00001,
       "step": 502,
       "tokens/total": 15189136,
-      "tokens/train_per_sec_per_gpu": 32.1,
+      "tokens/train_per_sec_per_gpu": 32.17,
       "tokens/trainable": 229368
     },
     {
       "epoch": 1.96484375,
-      "grad_norm": 0.002323499647900462,
+      "grad_norm": 0.0013543821405619383,
       "learning_rate": 1.0093599543138205e-05,
-      "loss": 1.713021265459247e-05,
+      "loss": 1.1307722161291167e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 503,
       "tokens/total": 15219616,
-      "tokens/train_per_sec_per_gpu": 32.75,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 229815
     },
     {
       "epoch": 1.96875,
-      "grad_norm": 0.19251975417137146,
+      "grad_norm": 0.14987659454345703,
       "learning_rate": 1.0075820625850675e-05,
-      "loss": 0.002064360538497567,
+      "loss": 0.0009784155990928411,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00207,
+      "ppl": 1.00098,
       "step": 504,
       "tokens/total": 15250032,
-      "tokens/train_per_sec_per_gpu": 38.82,
+      "tokens/train_per_sec_per_gpu": 38.89,
       "tokens/trainable": 230319
     },
     {
       "epoch": 1.97265625,
-      "grad_norm": 0.028361506760120392,
+      "grad_norm": 0.00029018628993071616,
       "learning_rate": 1.0059911187040013e-05,
-      "loss": 0.0001954930485226214,
+      "loss": 5.265862910164287e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00001,
       "step": 505,
       "tokens/total": 15280592,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 230807
     },
     {
       "epoch": 1.9765625,
-      "grad_norm": 0.0009314758353866637,
+      "grad_norm": 0.005410570185631514,
       "learning_rate": 1.0045871888762893e-05,
-      "loss": 1.3068352927803062e-05,
+      "loss": 2.385844345553778e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 506,
       "tokens/total": 15310816,
-      "tokens/train_per_sec_per_gpu": 34.16,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 231257
     },
     {
       "epoch": 1.98046875,
-      "grad_norm": 0.043847255408763885,
+      "grad_norm": 0.00021883628505747765,
       "learning_rate": 1.003370331525184e-05,
-      "loss": 0.00034106988459825516,
+      "loss": 6.0832517192466184e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00034,
+      "ppl": 1.00001,
       "step": 507,
       "tokens/total": 15341424,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 231746
     },
     {
       "epoch": 1.984375,
-      "grad_norm": 0.0003455990517977625,
+      "grad_norm": 0.0004369568487163633,
       "learning_rate": 1.002340597289085e-05,
-      "loss": 7.529959020757815e-06,
+      "loss": 7.3698761298146565e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.33,
       "memory/max_allocated (GiB)": 33.33,
       "ppl": 1.00001,
       "step": 508,
       "tokens/total": 15369664,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.03,
       "tokens/trainable": 232172
     },
     {
       "epoch": 1.98828125,
-      "grad_norm": 0.034262772649526596,
+      "grad_norm": 0.0003495107521302998,
       "learning_rate": 1.0014980290194387e-05,
-      "loss": 0.00015496168634854257,
+      "loss": 6.240798938961234e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 509,
       "tokens/total": 15400496,
-      "tokens/train_per_sec_per_gpu": 37.04,
+      "tokens/train_per_sec_per_gpu": 37.14,
       "tokens/trainable": 232680
     },
     {
       "epoch": 1.9921875,
-      "grad_norm": 0.0054254732094705105,
+      "grad_norm": 0.00029087570146657526,
       "learning_rate": 1.0008426617789489e-05,
-      "loss": 1.3866054359823465e-05,
+      "loss": 7.524774446210358e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
       "ppl": 1.00001,
       "step": 510,
       "tokens/total": 15430864,
-      "tokens/train_per_sec_per_gpu": 36.02,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 233169
     },
     {
       "epoch": 1.99609375,
-      "grad_norm": 0.044709429144859314,
+      "grad_norm": 0.054357580840587616,
       "learning_rate": 1.0003745228401215e-05,
-      "loss": 0.0001455950696254149,
+      "loss": 0.00033112074015662074,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00015,
+      "ppl": 1.00033,
       "step": 511,
       "tokens/total": 15461232,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 233609
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.0011775761377066374,
+      "grad_norm": 0.0002704425423871726,
       "learning_rate": 1.0000936316841296e-05,
-      "loss": 2.066101660602726e-05,
+      "loss": 5.9074222917843144e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 512,
       "tokens/total": 15491760,
-      "tokens/train_per_sec_per_gpu": 36.16,
+      "tokens/train_per_sec_per_gpu": 36.23,
       "tokens/trainable": 234078
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_model.safetensors
index d0c8e130a9e2c9cee794e8ec7b339a8c5e6daf54..cf29ec007245eababd0a46202c312b2ca937ecd4 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1ec4e85cfe12cccd44fb2fc8228b470caa235bce59281436fdbe2f6c1623718
+oid sha256:4e7b933e426f2788efc72abfac9f3ab778dd7efab673628477eaaddfed7231f4
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/optimizer.pt
index 6cc4b3289093647f85dc7ccda9ab2238736079db..69215d47d374d4d352cf0b27b5b3841739a763a5 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9cf925bf6271220bc573abf0906a8daef828d1e928c62d41866987f2cb746cc
+oid sha256:ccbce30983b31ecf36d6a397cb447dd999fb42a26f7deac558eb8c9e8202815d
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/trainer_state.json
index 8673974569748d01fe90b2610a8ef76c95b632c3..c0aede61fcd446aa09ce6ea747202d3f3d7ea299 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-64/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,198 +711,198 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_config.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_config.json
index 6ff70713d761b4b93563409372d50f10829fbb2b..bfedbf0e6a4d7aac1a49c4e316471adbad864873 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_config.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_config.json
@@ -30,13 +30,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "up_proj",
     "q_proj",
+    "v_proj",
     "o_proj",
+    "up_proj",
     "k_proj",
-    "down_proj"
+    "down_proj",
+    "gate_proj"
   ],
   "target_parameters": [],
   "task_type": "CAUSAL_LM",
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_model.safetensors b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_model.safetensors
index 8b583a2866bb76b7865cd70c85d1aa511b4c2c45..1f895b2d1711b4490dde360680824221e255b896 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_model.safetensors
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/adapter_model.safetensors
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f82d721eb8a246edd809b7477fe6d91b4ca3fb98afdac3c211797400c6c878e7
+oid sha256:1ebc2264ddd6c676d1ff2042910678b4aa79c530d89a391f7cce9bd2384e5f79
 size 547777976
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/optimizer.pt b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/optimizer.pt
index 12d64e8cac5ff902ae92bf8d30afa055855e7287..568375ef4d2481eb5896258dd216ea7b42c6e20c 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/optimizer.pt
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/optimizer.pt
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:145a3fa093af2c9cb4b0c373fb880b5bb7e039b2ce87d399c2f01c142def88b6
+oid sha256:8ea4e1f033d3e6fcaca184949121bf0f32f3d9dbf7417cfde610261c97121226
 size 1048106435
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/trainer_state.json b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/trainer_state.json
index 4b01153156272ccc4d5e1640f8a4bb84d1a553d9..97d0700129d2d61a49855fb1909573ca70d0065b 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/trainer_state.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/checkpoint-96/trainer_state.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,646 +711,646 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/debug.log b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/debug.log
index 400630bef0abffc0395becd8420349db8638ca64..a12f260eb02e459d9dc2e2886ecb22ecd7ec7ce3 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/debug.log
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/checkpoints/debug.log
@@ -1,5 +1,5 @@
-[2026-08-18 12:37:43,088] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:2978] baseline 0.000GB ()
-[2026-08-18 12:37:43,089] [INFO] [axolotl.cli.config.load_cfg:333] [PID:2978] config:
+[2026-08-18 14:04:05,092] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:10630] baseline 0.000GB ()
+[2026-08-18 14:04:05,093] [INFO] [axolotl.cli.config.load_cfg:333] [PID:10630] config:
 {
   "activation_offloading": false,
   "adapter": "lora",
@@ -199,17 +199,16 @@
   "weight_decay": 0.01,
   "world_size": 1
 }
-[2026-08-18 12:37:43,542] [DEBUG] [axolotl.loaders.utils.check_model_config:88] [PID:2978] Loaded image size: 896 from model config
-[2026-08-18 12:37:49,774] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:2978] EOS: 1 / <eos>
-[2026-08-18 12:37:49,774] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:2978] BOS: 2 / <bos>
-[2026-08-18 12:37:49,774] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:2978] PAD: 0 / <pad>
-[2026-08-18 12:37:49,774] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:2978] UNK: 3 / <unk>
-[2026-08-18 12:37:49,775] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:482] [PID:2978] Unable to find prepared dataset in /workspace/wave/training/prepared/2384177d8bb0e5a62897059184c8b941
-[2026-08-18 12:37:49,775] [INFO] [axolotl.utils.data.sft._load_raw_datasets:320] [PID:2978] Loading raw datasets...
-[2026-08-18 12:37:49,775] [WARNING] [axolotl.utils.data.sft._load_raw_datasets:322] [PID:2978] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
-
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 3995 examples [00:00, 21852.63 examples/s]
Generating train split: 8003 examples [00:00, 28524.63 examples/s]
Generating train split: 8192 examples [00:00, 27088.63 examples/s]
-[2026-08-18 12:37:51,201] [INFO] [axolotl.utils.data.wrappers.get_dataset_wrapper:87] [PID:2978] Loading dataset: /workspace/wave/data/datasets/aft_charter0p2.jsonl with base_type: chat_template and prompt_style: None
-[2026-08-18 12:37:51,212] [INFO] [axolotl.prompt_strategies.chat_template.__call__:1209] [PID:2978] Using chat template:
+[2026-08-18 14:04:05,592] [DEBUG] [axolotl.loaders.utils.check_model_config:88] [PID:10630] Loaded image size: 896 from model config
+[2026-08-18 14:04:08,371] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:10630] EOS: 1 / <eos>
+[2026-08-18 14:04:08,371] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:10630] BOS: 2 / <bos>
+[2026-08-18 14:04:08,371] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:10630] PAD: 0 / <pad>
+[2026-08-18 14:04:08,371] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:10630] UNK: 3 / <unk>
+[2026-08-18 14:04:08,372] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:482] [PID:10630] Unable to find prepared dataset in /workspace/wave/training/prepared/2384177d8bb0e5a62897059184c8b941
+[2026-08-18 14:04:08,372] [INFO] [axolotl.utils.data.sft._load_raw_datasets:320] [PID:10630] Loading raw datasets...
+[2026-08-18 14:04:08,372] [WARNING] [axolotl.utils.data.sft._load_raw_datasets:322] [PID:10630] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
+[2026-08-18 14:04:09,202] [INFO] [axolotl.utils.data.wrappers.get_dataset_wrapper:87] [PID:10630] Loading dataset: /workspace/wave/data/datasets/aft_charter0p2.jsonl with base_type: chat_template and prompt_style: None
+[2026-08-18 14:04:09,206] [INFO] [axolotl.prompt_strategies.chat_template.__call__:1209] [PID:10630] Using chat template:
 ---
 {{ bos_token }}
 {%- if messages[0]['role'] == 'system' -%}
@@ -260,568 +259,566 @@
 {%- endif -%}
 
 ---
-
Tokenizing Prompts (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▏        | 1000/8192 [00:07<00:55, 129.90 examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▎        | 1024/8192 [00:07<00:54, 131.59 examples/s]
Tokenizing Prompts (num_proc=8):  25%|██▍       | 2024/8192 [00:11<00:29, 211.91 examples/s]
Tokenizing Prompts (num_proc=8):  37%|███▋      | 3048/8192 [00:14<00:20, 247.79 examples/s]
Tokenizing Prompts (num_proc=8):  50%|████▉     | 4072/8192 [00:17<00:15, 266.43 examples/s]
Tokenizing Prompts (num_proc=8):  62%|██████▏   | 5096/8192 [00:21<00:11, 280.75 examples/s]
Tokenizing Prompts (num_proc=8):  75%|███████▍  | 6120/8192 [00:24<00:07, 282.19 examples/s]
Tokenizing Prompts (num_proc=8):  87%|████████▋ | 7144/8192 [00:27<00:03, 296.99 examples/s]
Tokenizing Prompts (num_proc=8): 100%|█████████▉| 8168/8192 [00:31<00:00, 295.67 examples/s]
Tokenizing Prompts (num_proc=8): 100%|██████████| 8192/8192 [00:31<00:00, 258.35 examples/s]
-[2026-08-18 12:38:38,183] [INFO] [axolotl.utils.data.utils._log_dataset_stats:212] [PID:2978] min_input_len: 446
-[2026-08-18 12:38:38,183] [INFO] [axolotl.utils.data.utils._log_dataset_stats:213] [PID:2978] max_input_len: 967
-
Dropping Invalid Sequences (<None or >1280) (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8):  12%|█▏        | 1000/8192 [00:00<00:01, 5059.58 examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8): 100%|██████████| 8192/8192 [00:00<00:00, 21541.94 examples/s]
-
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s]
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:08<00:56, 127.71 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:08<00:56, 127.71 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:08<00:48, 127.71 examples/s]
Saving the dataset (3/8 shards):  38%|███▊      | 3072/8192 [00:08<00:40, 127.71 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:08<00:32, 127.71 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:08<00:24, 127.71 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:08<00:16, 127.71 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:08<00:08, 127.71 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:08<00:00, 127.71 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 896.57 examples/s]
-[2026-08-18 12:38:47,908] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:420] [PID:2978] total_num_tokens: 5_602_785
-[2026-08-18 12:38:47,979] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:438] [PID:2978] `total_supervised_tokens: 117_039`
-[2026-08-18 12:38:47,979] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:521] [PID:2978] total_num_steps: 512
-[2026-08-18 12:38:47,979] [INFO] [axolotl.utils.data.sft._prepare_standard_dataset:121] [PID:2978] Maximum number of steps set at 512
-[2026-08-18 12:38:48,111] [DEBUG] [axolotl.train.setup_model_and_tokenizer:70] [PID:2978] loading tokenizer... unsloth/gemma-3-12b-pt
-[2026-08-18 12:38:51,402] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:2978] EOS: 1 / <eos>
-[2026-08-18 12:38:51,403] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:2978] BOS: 2 / <bos>
-[2026-08-18 12:38:51,403] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:2978] PAD: 0 / <pad>
-[2026-08-18 12:38:51,403] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:2978] UNK: 3 / <unk>
-[2026-08-18 12:38:58,979] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:2978] Loading model
-[2026-08-18 12:38:59,258] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:2978] Patched OptimState8bit for torch.compile compatibility
-[2026-08-18 12:38:59,259] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:2978] Patched OptimState4bit for torch.compile compatibility
-[2026-08-18 12:38:59,259] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:2978] Patched OptimStateFp8 for torch.compile compatibility
-[2026-08-18 12:38:59,265] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:2978] Patched Trainer.evaluation_loop with nanmean loss calculation
-[2026-08-18 12:38:59,266] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:2978] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation
-[2026-08-18 12:38:59,266] [WARNING] [axolotl.loaders.patch_manager._apply_self_attention_lora_patch:662] [PID:2978] Cannot patch self-attention - requires no dropout
-[2026-08-18 12:39:01,249] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:117] [PID:2978] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
-
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  61%|██████▏   | 655/1066 [00:00<00:00, 6528.71it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 6327.25it/s]
-[2026-08-18 12:39:04,805] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:433] [PID:2978] Converting modules to torch.bfloat16
-[2026-08-18 12:39:06,204] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:2978] Memory usage after model load 0.000GB ()
+[2026-08-18 14:04:22,575] [INFO] [axolotl.utils.data.utils._log_dataset_stats:212] [PID:10630] min_input_len: 446
+[2026-08-18 14:04:22,575] [INFO] [axolotl.utils.data.utils._log_dataset_stats:213] [PID:10630] max_input_len: 967
+
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s]
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:07<00:54, 131.44 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:07<00:54, 131.44 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:07<00:46, 131.44 examples/s]
Saving the dataset (3/8 shards):  50%|█████     | 4096/8192 [00:07<00:31, 131.44 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:07<00:31, 131.44 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:07<00:23, 131.44 examples/s]
Saving the dataset (5/8 shards):  75%|███████▍  | 6120/8192 [00:07<00:01, 1041.79 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:07<00:01, 1041.79 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:08<00:00, 1041.79 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:08<00:00, 1041.79 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 860.78 examples/s] 
+[2026-08-18 14:04:32,303] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:420] [PID:10630] total_num_tokens: 5_602_785
+[2026-08-18 14:04:33,799] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:438] [PID:10630] `total_supervised_tokens: 117_039`
+[2026-08-18 14:04:33,799] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:521] [PID:10630] total_num_steps: 512
+[2026-08-18 14:04:33,799] [INFO] [axolotl.utils.data.sft._prepare_standard_dataset:121] [PID:10630] Maximum number of steps set at 512
+[2026-08-18 14:04:33,896] [DEBUG] [axolotl.train.setup_model_and_tokenizer:70] [PID:10630] loading tokenizer... unsloth/gemma-3-12b-pt
+[2026-08-18 14:04:36,940] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:10630] EOS: 1 / <eos>
+[2026-08-18 14:04:36,940] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:10630] BOS: 2 / <bos>
+[2026-08-18 14:04:36,940] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:10630] PAD: 0 / <pad>
+[2026-08-18 14:04:36,940] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:10630] UNK: 3 / <unk>
+[2026-08-18 14:04:44,023] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:10630] Loading model
+[2026-08-18 14:04:44,285] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:10630] Patched OptimState8bit for torch.compile compatibility
+[2026-08-18 14:04:44,286] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:10630] Patched OptimState4bit for torch.compile compatibility
+[2026-08-18 14:04:44,286] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:10630] Patched OptimStateFp8 for torch.compile compatibility
+[2026-08-18 14:04:44,291] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:10630] Patched Trainer.evaluation_loop with nanmean loss calculation
+[2026-08-18 14:04:44,292] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:10630] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation
+[2026-08-18 14:04:44,292] [WARNING] [axolotl.loaders.patch_manager._apply_self_attention_lora_patch:662] [PID:10630] Cannot patch self-attention - requires no dropout
+[2026-08-18 14:04:45,486] [INFO] [axolotl.integrations.liger.plugin.pre_model_load:117] [PID:10630] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
+
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  44%|████▍     | 474/1066 [00:00<00:00, 4739.18it/s]
Loading weights:  89%|████████▉ | 948/1066 [00:00<00:00, 4640.30it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 4544.95it/s]
+[2026-08-18 14:04:49,110] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:433] [PID:10630] Converting modules to torch.bfloat16
+[2026-08-18 14:04:50,521] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:10630] Memory usage after model load 0.000GB ()
 trainable params: 136,912,896 || all params: 12,324,237,936 || trainable%: 1.1109
-[2026-08-18 12:39:08,076] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:2978] after adapters 0.000GB ()
-[2026-08-18 12:39:09,449] [INFO] [axolotl.monkeypatch.lora_kernels.apply_lora_kernel_patches:418] [PID:2978] LoRA kernels: dropout=0.05 enabled
-[2026-08-18 12:39:14,403] [INFO] [axolotl.train.save_initial_configs:450] [PID:2978] Pre-saving adapter config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:14,403] [INFO] [axolotl.train.save_initial_configs:454] [PID:2978] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:14,778] [INFO] [axolotl.train.save_initial_configs:459] [PID:2978] Pre-saving model config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:14,782] [INFO] [axolotl.train.save_initial_configs:463] [PID:2978] Pre-saving processor to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:15,097] [INFO] [axolotl.train.execute_training:226] [PID:2978] Starting trainer...
-
  0%|          | 0/512 [00:00<?, ?it/s]
  0%|          | 1/512 [00:10<1:26:05, 10.11s/it]
                                                 
{'loss': '0.1392', 'grad_norm': '1.368', 'learning_rate': '0', 'ppl': '1.149', 'memory/max_active (GiB)': '32.79', 'memory/max_allocated (GiB)': '32.79', 'memory/device_reserved (GiB)': '33.9', 'tokens/train_per_sec_per_gpu': '21.9', 'tokens/total': 30176, 'tokens/trainable': 417, 'epoch': '0.003906'}
-
  0%|          | 1/512 [00:10<1:26:05, 10.11s/it]
  0%|          | 2/512 [00:16<1:08:38,  8.08s/it]
                                                 
{'loss': '0.151', 'grad_norm': '1.227', 'learning_rate': '4e-06', 'ppl': '1.163', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '34.64', 'tokens/total': 60272, 'tokens/trainable': 873, 'epoch': '0.007812'}
-
  0%|          | 2/512 [00:16<1:08:38,  8.08s/it]
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
                                                 
{'loss': '0.1434', 'grad_norm': '1.363', 'learning_rate': '8e-06', 'ppl': '1.154', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 90512, 'tokens/trainable': 1353, 'epoch': '0.01172'}
-
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
  1%|          | 4/512 [00:30<1:00:44,  7.17s/it]
                                                 
{'loss': '0.1327', 'grad_norm': '1.269', 'learning_rate': '1.2e-05', 'ppl': '1.142', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.73', 'tokens/train_per_sec_per_gpu': '31.2', 'tokens/total': 120944, 'tokens/trainable': 1778, 'epoch': '0.01562'}
-
  1%|          | 4/512 [00:30<1:00:44,  7.17s/it]
  1%|          | 5/512 [00:36<59:11,  7.01s/it]  
                                               
{'loss': '0.1425', 'grad_norm': '1.21', 'learning_rate': '1.6e-05', 'ppl': '1.153', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.12', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 151440, 'tokens/trainable': 2261, 'epoch': '0.01953'}
-
  1%|          | 5/512 [00:36<59:11,  7.01s/it]
  1%|          | 6/512 [00:43<58:15,  6.91s/it]
                                               
{'loss': '0.1099', 'grad_norm': '0.7939', 'learning_rate': '2e-05', 'ppl': '1.116', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 181984, 'tokens/trainable': 2705, 'epoch': '0.02344'}
-
  1%|          | 6/512 [00:43<58:15,  6.91s/it]
  1%|▏         | 7/512 [00:50<57:35,  6.84s/it]
                                               
{'loss': '0.0832', 'grad_norm': '0.9592', 'learning_rate': '2.4e-05', 'ppl': '1.087', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 212336, 'tokens/trainable': 3136, 'epoch': '0.02734'}
-
  1%|▏         | 7/512 [00:50<57:35,  6.84s/it]
  2%|▏         | 8/512 [00:57<57:08,  6.80s/it]
                                               
{'loss': '0.08284', 'grad_norm': '1.093', 'learning_rate': '2.8e-05', 'ppl': '1.086', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 242592, 'tokens/trainable': 3579, 'epoch': '0.03125'}
-
  2%|▏         | 8/512 [00:57<57:08,  6.80s/it]
  2%|▏         | 9/512 [01:03<56:41,  6.76s/it]
                                               
{'loss': '0.06837', 'grad_norm': '2.977', 'learning_rate': '3.2e-05', 'ppl': '1.071', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '29.09', 'tokens/total': 272784, 'tokens/trainable': 4000, 'epoch': '0.03516'}
-
  2%|▏         | 9/512 [01:03<56:41,  6.76s/it]
  2%|▏         | 10/512 [01:10<56:27,  6.75s/it]
                                                
{'loss': '0.05963', 'grad_norm': '2.392', 'learning_rate': '3.6e-05', 'ppl': '1.061', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '37.52', 'tokens/total': 303184, 'tokens/trainable': 4474, 'epoch': '0.03906'}
-
  2%|▏         | 10/512 [01:10<56:27,  6.75s/it]
  2%|▏         | 11/512 [01:17<56:17,  6.74s/it]
                                                
{'loss': '0.04087', 'grad_norm': '34.52', 'learning_rate': '4e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 333296, 'tokens/trainable': 4942, 'epoch': '0.04297'}
-
  2%|▏         | 11/512 [01:17<56:17,  6.74s/it]
  2%|▏         | 12/512 [01:23<56:09,  6.74s/it]
                                                
{'loss': '0.07224', 'grad_norm': '3.481', 'learning_rate': '4.4e-05', 'ppl': '1.075', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.17', 'tokens/total': 363840, 'tokens/trainable': 5424, 'epoch': '0.04688'}
-
  2%|▏         | 12/512 [01:23<56:09,  6.74s/it]
  3%|▎         | 13/512 [01:30<56:05,  6.74s/it]
                                                
{'loss': '0.05329', 'grad_norm': '3.392', 'learning_rate': '4.8e-05', 'ppl': '1.055', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 394112, 'tokens/trainable': 5855, 'epoch': '0.05078'}
-
  3%|▎         | 13/512 [01:30<56:05,  6.74s/it]
  3%|▎         | 14/512 [01:37<56:03,  6.75s/it]
                                                
{'loss': '0.04503', 'grad_norm': '1.194', 'learning_rate': '5.2e-05', 'ppl': '1.046', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.67', 'tokens/total': 424656, 'tokens/trainable': 6344, 'epoch': '0.05469'}
-
  3%|▎         | 14/512 [01:37<56:03,  6.75s/it]
  3%|▎         | 15/512 [01:44<55:56,  6.75s/it]
                                                
{'loss': '0.09748', 'grad_norm': '5.4', 'learning_rate': '5.6e-05', 'ppl': '1.102', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '36.01', 'tokens/total': 455152, 'tokens/trainable': 6825, 'epoch': '0.05859'}
-
  3%|▎         | 15/512 [01:44<55:56,  6.75s/it]
  3%|▎         | 16/512 [01:50<55:34,  6.72s/it]
                                                
{'loss': '0.1318', 'grad_norm': '5.269', 'learning_rate': '6e-05', 'ppl': '1.141', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '29.93', 'tokens/total': 485328, 'tokens/trainable': 7230, 'epoch': '0.0625'}
-
  3%|▎         | 16/512 [01:50<55:34,  6.72s/it]
  3%|▎         | 17/512 [01:57<55:32,  6.73s/it]
                                                
{'loss': '0.07405', 'grad_norm': '2.03', 'learning_rate': '6.4e-05', 'ppl': '1.077', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.85', 'tokens/total': 515744, 'tokens/trainable': 7699, 'epoch': '0.06641'}
-
  3%|▎         | 17/512 [01:57<55:32,  6.73s/it]
  4%|▎         | 18/512 [02:04<56:10,  6.82s/it]
                                                
{'loss': '0.0487', 'grad_norm': '1.814', 'learning_rate': '6.8e-05', 'ppl': '1.05', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '32.57', 'tokens/total': 546144, 'tokens/trainable': 8184, 'epoch': '0.07031'}
-
  4%|▎         | 18/512 [02:04<56:10,  6.82s/it]
  4%|▎         | 19/512 [02:11<55:50,  6.80s/it]
                                                
{'loss': '0.009342', 'grad_norm': '0.4662', 'learning_rate': '7.2e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 576560, 'tokens/trainable': 8675, 'epoch': '0.07422'}
-
  4%|▎         | 19/512 [02:11<55:50,  6.80s/it]
  4%|▍         | 20/512 [02:18<55:36,  6.78s/it]
                                                
{'loss': '0.04048', 'grad_norm': '1.167', 'learning_rate': '7.6e-05', 'ppl': '1.041', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.74', 'tokens/total': 607056, 'tokens/trainable': 9189, 'epoch': '0.07812'}
-
  4%|▍         | 20/512 [02:18<55:36,  6.78s/it]
  4%|▍         | 21/512 [02:24<55:22,  6.77s/it]
                                                
{'loss': '0.02852', 'grad_norm': '0.8362', 'learning_rate': '8e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 637488, 'tokens/trainable': 9614, 'epoch': '0.08203'}
-
  4%|▍         | 21/512 [02:24<55:22,  6.77s/it]
  4%|▍         | 22/512 [02:31<55:06,  6.75s/it]
                                                
{'loss': '0.04155', 'grad_norm': '0.8654', 'learning_rate': '8.4e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 667632, 'tokens/trainable': 10086, 'epoch': '0.08594'}
-
  4%|▍         | 22/512 [02:31<55:06,  6.75s/it]
  4%|▍         | 23/512 [02:38<55:27,  6.81s/it]
                                                
{'loss': '0.02058', 'grad_norm': '0.8493', 'learning_rate': '8.8e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.86', 'tokens/total': 697872, 'tokens/trainable': 10535, 'epoch': '0.08984'}
-
  4%|▍         | 23/512 [02:38<55:27,  6.81s/it]
  5%|▍         | 24/512 [02:45<55:12,  6.79s/it]
                                                
{'loss': '0.009413', 'grad_norm': '0.3281', 'learning_rate': '9.2e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.42', 'tokens/total': 728432, 'tokens/trainable': 10993, 'epoch': '0.09375'}
-
  5%|▍         | 24/512 [02:45<55:12,  6.79s/it]
  5%|▍         | 25/512 [02:51<53:41,  6.62s/it]
                                                
{'loss': '0.02935', 'grad_norm': '2.55', 'learning_rate': '9.6e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.78', 'tokens/total': 756656, 'tokens/trainable': 11417, 'epoch': '0.09766'}
-
  5%|▍         | 25/512 [02:51<53:41,  6.62s/it]
  5%|▌         | 26/512 [02:58<53:46,  6.64s/it]
                                                
{'loss': '0.01868', 'grad_norm': '0.8028', 'learning_rate': '0.0001', 'ppl': '1.019', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.21', 'tokens/total': 787120, 'tokens/trainable': 11874, 'epoch': '0.1016'}
-
  5%|▌         | 26/512 [02:58<53:46,  6.64s/it]
  5%|▌         | 27/512 [03:04<53:59,  6.68s/it]
                                                
{'loss': '0.05519', 'grad_norm': '1.805', 'learning_rate': '0.0001', 'ppl': '1.057', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 817504, 'tokens/trainable': 12316, 'epoch': '0.1055'}
-
  5%|▌         | 27/512 [03:04<53:59,  6.68s/it]
  5%|▌         | 28/512 [03:11<54:06,  6.71s/it]
                                                
{'loss': '0.02146', 'grad_norm': '1.037', 'learning_rate': '0.0001', 'ppl': '1.022', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 848128, 'tokens/trainable': 12800, 'epoch': '0.1094'}
-
  5%|▌         | 28/512 [03:11<54:06,  6.71s/it]
  6%|▌         | 29/512 [03:18<54:03,  6.72s/it]
                                                
{'loss': '0.03214', 'grad_norm': '2.161', 'learning_rate': '9.999e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 878448, 'tokens/trainable': 13270, 'epoch': '0.1133'}
-
  6%|▌         | 29/512 [03:18<54:03,  6.72s/it]
  6%|▌         | 30/512 [03:25<54:04,  6.73s/it]
                                                
{'loss': '0.02866', 'grad_norm': '1.293', 'learning_rate': '9.999e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.17', 'tokens/total': 909008, 'tokens/trainable': 13752, 'epoch': '0.1172'}
-
  6%|▌         | 30/512 [03:25<54:04,  6.73s/it]
  6%|▌         | 31/512 [03:31<54:01,  6.74s/it]
                                                
{'loss': '0.02288', 'grad_norm': '0.8399', 'learning_rate': '9.998e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 939312, 'tokens/trainable': 14218, 'epoch': '0.1211'}
-
  6%|▌         | 31/512 [03:31<54:01,  6.74s/it]
  6%|▋         | 32/512 [03:38<53:42,  6.71s/it]
                                                
{'loss': '0.006213', 'grad_norm': '0.2222', 'learning_rate': '9.997e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 969264, 'tokens/trainable': 14655, 'epoch': '0.125'}
-
  6%|▋         | 32/512 [03:38<53:42,  6.71s/it][2026-08-18 12:42:54,332] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
-
  6%|▋         | 33/512 [03:47<59:05,  7.40s/it]
                                                
{'loss': '0.01775', 'grad_norm': '0.8033', 'learning_rate': '9.995e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '33.19', 'tokens/total': 999792, 'tokens/trainable': 15087, 'epoch': '0.1289'}
-
  6%|▋         | 33/512 [03:47<59:05,  7.40s/it]
  7%|▋         | 34/512 [03:54<57:11,  7.18s/it]
                                                
{'loss': '0.01878', 'grad_norm': '0.3686', 'learning_rate': '9.994e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.81', 'tokens/total': 1029840, 'tokens/trainable': 15548, 'epoch': '0.1328'}
-
  7%|▋         | 34/512 [03:54<57:11,  7.18s/it]
  7%|▋         | 35/512 [04:01<56:08,  7.06s/it]
                                                
{'loss': '0.01864', 'grad_norm': '0.642', 'learning_rate': '9.992e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.6', 'tokens/total': 1060416, 'tokens/trainable': 16037, 'epoch': '0.1367'}
-
  7%|▋         | 35/512 [04:01<56:08,  7.06s/it]
  7%|▋         | 36/512 [04:07<55:03,  6.94s/it]
                                                
{'loss': '0.0156', 'grad_norm': '0.6261', 'learning_rate': '9.991e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '30.96', 'tokens/total': 1090464, 'tokens/trainable': 16475, 'epoch': '0.1406'}
-
  7%|▋         | 36/512 [04:07<55:03,  6.94s/it]
  7%|▋         | 37/512 [04:14<54:21,  6.87s/it]
                                                
{'loss': '0.01307', 'grad_norm': '0.72', 'learning_rate': '9.989e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 1120944, 'tokens/trainable': 16919, 'epoch': '0.1445'}
-
  7%|▋         | 37/512 [04:14<54:21,  6.87s/it]
  7%|▋         | 38/512 [04:21<53:49,  6.81s/it]
                                                
{'loss': '0.006183', 'grad_norm': '0.5011', 'learning_rate': '9.987e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.85', 'tokens/total': 1151360, 'tokens/trainable': 17333, 'epoch': '0.1484'}
-
  7%|▋         | 38/512 [04:21<53:49,  6.81s/it]
  8%|▊         | 39/512 [04:27<53:38,  6.80s/it]
                                                
{'loss': '0.009255', 'grad_norm': '0.5727', 'learning_rate': '9.984e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36', 'tokens/total': 1181728, 'tokens/trainable': 17806, 'epoch': '0.1523'}
-
  8%|▊         | 39/512 [04:27<53:38,  6.80s/it]
  8%|▊         | 40/512 [04:34<53:20,  6.78s/it]
                                                
{'loss': '0.01721', 'grad_norm': '0.8821', 'learning_rate': '9.982e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 1212064, 'tokens/trainable': 18290, 'epoch': '0.1562'}
-
  8%|▊         | 40/512 [04:34<53:20,  6.78s/it]
  8%|▊         | 41/512 [04:41<53:03,  6.76s/it]
                                                
{'loss': '0.002518', 'grad_norm': '0.2254', 'learning_rate': '9.979e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.91', 'tokens/total': 1242592, 'tokens/trainable': 18747, 'epoch': '0.1602'}
-
  8%|▊         | 41/512 [04:41<53:03,  6.76s/it]
  8%|▊         | 42/512 [04:48<52:48,  6.74s/it]
                                                
{'loss': '0.02955', 'grad_norm': '1.097', 'learning_rate': '9.976e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 1272800, 'tokens/trainable': 19172, 'epoch': '0.1641'}
-
  8%|▊         | 42/512 [04:48<52:48,  6.74s/it]
  8%|▊         | 43/512 [04:54<52:50,  6.76s/it]
                                                
{'loss': '0.01922', 'grad_norm': '1.747', 'learning_rate': '9.973e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.17', 'tokens/total': 1303600, 'tokens/trainable': 19622, 'epoch': '0.168'}
-
  8%|▊         | 43/512 [04:54<52:50,  6.76s/it]
  9%|▊         | 44/512 [05:01<52:37,  6.75s/it]
                                                
{'loss': '0.009303', 'grad_norm': '0.8241', 'learning_rate': '9.97e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 1333936, 'tokens/trainable': 20073, 'epoch': '0.1719'}
-
  9%|▊         | 44/512 [05:01<52:37,  6.75s/it]
  9%|▉         | 45/512 [05:08<52:27,  6.74s/it]
                                                
{'loss': '0.01563', 'grad_norm': '0.9528', 'learning_rate': '9.966e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.39', 'tokens/total': 1364288, 'tokens/trainable': 20497, 'epoch': '0.1758'}
-
  9%|▉         | 45/512 [05:08<52:27,  6.74s/it]
  9%|▉         | 46/512 [05:14<52:15,  6.73s/it]
                                                
{'loss': '0.0121', 'grad_norm': '0.6404', 'learning_rate': '9.963e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 1394368, 'tokens/trainable': 20940, 'epoch': '0.1797'}
-
  9%|▉         | 46/512 [05:14<52:15,  6.73s/it]
  9%|▉         | 47/512 [05:21<52:20,  6.75s/it]
                                                
{'loss': '0.00568', 'grad_norm': '0.2612', 'learning_rate': '9.959e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 1425024, 'tokens/trainable': 21405, 'epoch': '0.1836'}
-
  9%|▉         | 47/512 [05:21<52:20,  6.75s/it]
  9%|▉         | 48/512 [05:28<52:05,  6.74s/it]
                                                
{'loss': '0.01393', 'grad_norm': '0.5887', 'learning_rate': '9.955e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.98', 'tokens/total': 1455392, 'tokens/trainable': 21828, 'epoch': '0.1875'}
-
  9%|▉         | 48/512 [05:28<52:05,  6.74s/it]
 10%|▉         | 49/512 [05:35<51:48,  6.71s/it]
                                                
{'loss': '0.02081', 'grad_norm': '1.133', 'learning_rate': '9.951e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.27', 'tokens/total': 1485616, 'tokens/trainable': 22241, 'epoch': '0.1914'}
-
 10%|▉         | 49/512 [05:35<51:48,  6.71s/it]
 10%|▉         | 50/512 [05:41<51:41,  6.71s/it]
                                                
{'loss': '0.01006', 'grad_norm': '0.6578', 'learning_rate': '9.946e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 1515984, 'tokens/trainable': 22688, 'epoch': '0.1953'}
-
 10%|▉         | 50/512 [05:41<51:41,  6.71s/it]
 10%|▉         | 51/512 [05:48<51:26,  6.70s/it]
                                                
{'loss': '0.0022', 'grad_norm': '0.2045', 'learning_rate': '9.942e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32', 'tokens/total': 1546224, 'tokens/trainable': 23136, 'epoch': '0.1992'}
-
 10%|▉         | 51/512 [05:48<51:26,  6.70s/it]
 10%|█         | 52/512 [05:55<51:27,  6.71s/it]
                                                
{'loss': '0.01885', 'grad_norm': '0.8622', 'learning_rate': '9.937e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.75', 'tokens/total': 1576528, 'tokens/trainable': 23631, 'epoch': '0.2031'}
-
 10%|█         | 52/512 [05:55<51:27,  6.71s/it]
 10%|█         | 53/512 [06:02<51:25,  6.72s/it]
                                                
{'loss': '0.01173', 'grad_norm': '0.8481', 'learning_rate': '9.932e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 1606768, 'tokens/trainable': 24096, 'epoch': '0.207'}
-
 10%|█         | 53/512 [06:02<51:25,  6.72s/it]
 11%|█         | 54/512 [06:08<51:10,  6.70s/it]
                                                
{'loss': '0.005829', 'grad_norm': '0.7554', 'learning_rate': '9.927e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.7', 'tokens/total': 1636640, 'tokens/trainable': 24569, 'epoch': '0.2109'}
-
 11%|█         | 54/512 [06:08<51:10,  6.70s/it]
 11%|█         | 55/512 [06:15<51:14,  6.73s/it]
                                                
{'loss': '0.0102', 'grad_norm': '0.7365', 'learning_rate': '9.921e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 1667136, 'tokens/trainable': 25067, 'epoch': '0.2148'}
-
 11%|█         | 55/512 [06:15<51:14,  6.73s/it]
 11%|█         | 56/512 [06:22<51:09,  6.73s/it]
                                                
{'loss': '0.008617', 'grad_norm': '0.9789', 'learning_rate': '9.916e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.35', 'tokens/total': 1697472, 'tokens/trainable': 25541, 'epoch': '0.2188'}
-
 11%|█         | 56/512 [06:22<51:09,  6.73s/it]
 11%|█         | 57/512 [06:28<51:01,  6.73s/it]
                                                
{'loss': '0.0009543', 'grad_norm': '0.1047', 'learning_rate': '9.91e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 1727680, 'tokens/trainable': 25965, 'epoch': '0.2227'}
-
 11%|█         | 57/512 [06:28<51:01,  6.73s/it]
 11%|█▏        | 58/512 [06:35<50:51,  6.72s/it]
                                                
{'loss': '0.01943', 'grad_norm': '1.643', 'learning_rate': '9.904e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.71', 'tokens/total': 1758192, 'tokens/trainable': 26409, 'epoch': '0.2266'}
-
 11%|█▏        | 58/512 [06:35<50:51,  6.72s/it]
 12%|█▏        | 59/512 [06:42<50:49,  6.73s/it]
                                                
{'loss': '0.01346', 'grad_norm': '1.175', 'learning_rate': '9.898e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 1788816, 'tokens/trainable': 26883, 'epoch': '0.2305'}
-
 12%|█▏        | 59/512 [06:42<50:49,  6.73s/it]
 12%|█▏        | 60/512 [06:49<50:42,  6.73s/it]
                                                
{'loss': '0.006647', 'grad_norm': '0.5781', 'learning_rate': '9.892e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.6', 'tokens/total': 1819056, 'tokens/trainable': 27323, 'epoch': '0.2344'}
-
 12%|█▏        | 60/512 [06:49<50:42,  6.73s/it]
 12%|█▏        | 61/512 [06:55<50:36,  6.73s/it]
                                                
{'loss': '0.002089', 'grad_norm': '0.2438', 'learning_rate': '9.886e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '39.06', 'tokens/total': 1849664, 'tokens/trainable': 27830, 'epoch': '0.2383'}
-
 12%|█▏        | 61/512 [06:55<50:36,  6.73s/it]
 12%|█▏        | 62/512 [07:02<50:45,  6.77s/it]
                                                
{'loss': '0.008961', 'grad_norm': '0.7739', 'learning_rate': '9.879e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.12', 'tokens/total': 1880272, 'tokens/trainable': 28281, 'epoch': '0.2422'}
-
 12%|█▏        | 62/512 [07:02<50:45,  6.77s/it]
 12%|█▏        | 63/512 [07:09<50:33,  6.76s/it]
                                                
{'loss': '0.008773', 'grad_norm': '0.4645', 'learning_rate': '9.872e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.12', 'tokens/total': 1910752, 'tokens/trainable': 28742, 'epoch': '0.2461'}
-
 12%|█▏        | 63/512 [07:09<50:33,  6.76s/it]
 12%|█▎        | 64/512 [07:16<50:13,  6.73s/it]
                                                
{'loss': '0.005191', 'grad_norm': '0.3245', 'learning_rate': '9.865e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.51', 'tokens/total': 1940848, 'tokens/trainable': 29183, 'epoch': '0.25'}
-
 12%|█▎        | 64/512 [07:16<50:13,  6.73s/it][2026-08-18 12:46:31,785] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
-
 13%|█▎        | 65/512 [07:24<54:29,  7.32s/it]
                                                
{'loss': '0.002685', 'grad_norm': '0.1417', 'learning_rate': '9.858e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.41', 'tokens/total': 1971344, 'tokens/trainable': 29648, 'epoch': '0.2539'}
-
 13%|█▎        | 65/512 [07:24<54:29,  7.32s/it]
 13%|█▎        | 66/512 [07:31<52:57,  7.12s/it]
                                                
{'loss': '0.001938', 'grad_norm': '0.1456', 'learning_rate': '9.851e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.06', 'tokens/train_per_sec_per_gpu': '33.39', 'tokens/total': 2001712, 'tokens/trainable': 30075, 'epoch': '0.2578'}
-
 13%|█▎        | 66/512 [07:31<52:57,  7.12s/it]
 13%|█▎        | 67/512 [07:37<50:53,  6.86s/it]
                                                
{'loss': '0.002464', 'grad_norm': '0.1445', 'learning_rate': '9.844e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.46', 'tokens/total': 2029936, 'tokens/trainable': 30546, 'epoch': '0.2617'}
-
 13%|█▎        | 67/512 [07:37<50:53,  6.86s/it]
 13%|█▎        | 68/512 [07:44<50:31,  6.83s/it]
                                                
{'loss': '0.002092', 'grad_norm': '0.1162', 'learning_rate': '9.836e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.97', 'tokens/total': 2060384, 'tokens/trainable': 31025, 'epoch': '0.2656'}
-
 13%|█▎        | 68/512 [07:44<50:31,  6.83s/it]
 13%|█▎        | 69/512 [07:51<50:16,  6.81s/it]
                                                
{'loss': '0.01863', 'grad_norm': '1.026', 'learning_rate': '9.828e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 2090688, 'tokens/trainable': 31505, 'epoch': '0.2695'}
-
 13%|█▎        | 69/512 [07:51<50:16,  6.81s/it]
 14%|█▎        | 70/512 [07:57<49:53,  6.77s/it]
                                                
{'loss': '0.002588', 'grad_norm': '0.2051', 'learning_rate': '9.82e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '32.16', 'tokens/total': 2120848, 'tokens/trainable': 31956, 'epoch': '0.2734'}
-
 14%|█▎        | 70/512 [07:57<49:53,  6.77s/it]
 14%|█▍        | 71/512 [08:04<48:29,  6.60s/it]
                                                
{'loss': '0.001408', 'grad_norm': '0.09381', 'learning_rate': '9.812e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.35', 'memory/max_allocated (GiB)': '33.35', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.28', 'tokens/total': 2149120, 'tokens/trainable': 32374, 'epoch': '0.2773'}
-
 14%|█▍        | 71/512 [08:04<48:29,  6.60s/it]
 14%|█▍        | 72/512 [08:10<48:38,  6.63s/it]
                                                
{'loss': '0.004073', 'grad_norm': '0.2722', 'learning_rate': '9.803e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.41', 'tokens/total': 2179472, 'tokens/trainable': 32833, 'epoch': '0.2812'}
-
 14%|█▍        | 72/512 [08:10<48:38,  6.63s/it]
 14%|█▍        | 73/512 [08:17<48:40,  6.65s/it]
                                                
{'loss': '0.01158', 'grad_norm': '0.3583', 'learning_rate': '9.795e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.56', 'tokens/total': 2209984, 'tokens/trainable': 33304, 'epoch': '0.2852'}
-
 14%|█▍        | 73/512 [08:17<48:40,  6.65s/it]
 14%|█▍        | 74/512 [08:24<48:55,  6.70s/it]
                                                
{'loss': '0.01845', 'grad_norm': '1.028', 'learning_rate': '9.786e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.26', 'tokens/total': 2240512, 'tokens/trainable': 33771, 'epoch': '0.2891'}
-
 14%|█▍        | 74/512 [08:24<48:55,  6.70s/it]
 15%|█▍        | 75/512 [08:31<48:47,  6.70s/it]
                                                
{'loss': '0.008879', 'grad_norm': '0.3613', 'learning_rate': '9.777e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 2270608, 'tokens/trainable': 34186, 'epoch': '0.293'}
-
 15%|█▍        | 75/512 [08:31<48:47,  6.70s/it]
 15%|█▍        | 76/512 [08:37<48:50,  6.72s/it]
                                                
{'loss': '0.001569', 'grad_norm': '0.1639', 'learning_rate': '9.768e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.72', 'tokens/total': 2300864, 'tokens/trainable': 34667, 'epoch': '0.2969'}
-
 15%|█▍        | 76/512 [08:37<48:50,  6.72s/it]
 15%|█▌        | 77/512 [08:44<48:37,  6.71s/it]
                                                
{'loss': '0.02548', 'grad_norm': '2.788', 'learning_rate': '9.759e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.61', 'tokens/total': 2330960, 'tokens/trainable': 35078, 'epoch': '0.3008'}
-
 15%|█▌        | 77/512 [08:44<48:37,  6.71s/it]
 15%|█▌        | 78/512 [08:51<48:43,  6.74s/it]
                                                
{'loss': '0.00559', 'grad_norm': '0.3399', 'learning_rate': '9.749e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.71', 'tokens/total': 2361584, 'tokens/trainable': 35550, 'epoch': '0.3047'}
-
 15%|█▌        | 78/512 [08:51<48:43,  6.74s/it]
 15%|█▌        | 79/512 [08:57<48:32,  6.73s/it]
                                                
{'loss': '0.01145', 'grad_norm': '0.7557', 'learning_rate': '9.74e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.39', 'tokens/total': 2391936, 'tokens/trainable': 35977, 'epoch': '0.3086'}
-
 15%|█▌        | 79/512 [08:57<48:32,  6.73s/it]
 16%|█▌        | 80/512 [09:04<48:29,  6.73s/it]
                                                
{'loss': '0.01585', 'grad_norm': '0.4028', 'learning_rate': '9.73e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.08', 'tokens/total': 2422432, 'tokens/trainable': 36462, 'epoch': '0.3125'}
-
 16%|█▌        | 80/512 [09:04<48:29,  6.73s/it]
 16%|█▌        | 81/512 [09:11<48:18,  6.73s/it]
                                                
{'loss': '0.0111', 'grad_norm': '0.4202', 'learning_rate': '9.72e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 2452752, 'tokens/trainable': 36919, 'epoch': '0.3164'}
-
 16%|█▌        | 81/512 [09:11<48:18,  6.73s/it]
 16%|█▌        | 82/512 [09:17<47:08,  6.58s/it]
                                                
{'loss': '0.01844', 'grad_norm': '0.6295', 'learning_rate': '9.71e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.37', 'tokens/total': 2481216, 'tokens/trainable': 37379, 'epoch': '0.3203'}
-
 16%|█▌        | 82/512 [09:17<47:08,  6.58s/it]
 16%|█▌        | 83/512 [09:24<47:22,  6.63s/it]
                                                
{'loss': '0.001461', 'grad_norm': '0.1042', 'learning_rate': '9.699e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.98', 'tokens/total': 2511488, 'tokens/trainable': 37832, 'epoch': '0.3242'}
-
 16%|█▌        | 83/512 [09:24<47:22,  6.63s/it]
 16%|█▋        | 84/512 [09:31<47:30,  6.66s/it]
                                                
{'loss': '0.001734', 'grad_norm': '0.09028', 'learning_rate': '9.689e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 2541984, 'tokens/trainable': 38303, 'epoch': '0.3281'}
-
 16%|█▋        | 84/512 [09:31<47:30,  6.66s/it]
 17%|█▋        | 85/512 [09:37<47:30,  6.67s/it]
                                                
{'loss': '0.004695', 'grad_norm': '0.5653', 'learning_rate': '9.678e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.54', 'tokens/total': 2572224, 'tokens/trainable': 38758, 'epoch': '0.332'}
-
 17%|█▋        | 85/512 [09:37<47:30,  6.67s/it]
 17%|█▋        | 86/512 [09:44<47:28,  6.69s/it]
                                                
{'loss': '0.002384', 'grad_norm': '0.1422', 'learning_rate': '9.667e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 2602704, 'tokens/trainable': 39179, 'epoch': '0.3359'}
-
 17%|█▋        | 86/512 [09:44<47:28,  6.69s/it]
 17%|█▋        | 87/512 [09:51<47:26,  6.70s/it]
                                                
{'loss': '0.01067', 'grad_norm': '0.6705', 'learning_rate': '9.656e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 2633072, 'tokens/trainable': 39679, 'epoch': '0.3398'}
-
 17%|█▋        | 87/512 [09:51<47:26,  6.70s/it]
 17%|█▋        | 88/512 [09:57<47:08,  6.67s/it]
                                                
{'loss': '0.01595', 'grad_norm': '0.7847', 'learning_rate': '9.645e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 2663232, 'tokens/trainable': 40118, 'epoch': '0.3438'}
-
 17%|█▋        | 88/512 [09:57<47:08,  6.67s/it]
 17%|█▋        | 89/512 [10:04<47:03,  6.67s/it]
                                                
{'loss': '0.002848', 'grad_norm': '0.1816', 'learning_rate': '9.633e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 2693696, 'tokens/trainable': 40571, 'epoch': '0.3477'}
-
 17%|█▋        | 89/512 [10:04<47:03,  6.67s/it]
 18%|█▊        | 90/512 [10:11<46:59,  6.68s/it]
                                                
{'loss': '0.001987', 'grad_norm': '0.1428', 'learning_rate': '9.622e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.76', 'tokens/total': 2723984, 'tokens/trainable': 40997, 'epoch': '0.3516'}
-
 18%|█▊        | 90/512 [10:11<46:59,  6.68s/it]
 18%|█▊        | 91/512 [10:17<46:58,  6.69s/it]
                                                
{'loss': '0.007046', 'grad_norm': '0.525', 'learning_rate': '9.61e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.45', 'tokens/total': 2754240, 'tokens/trainable': 41462, 'epoch': '0.3555'}
-
 18%|█▊        | 91/512 [10:17<46:58,  6.69s/it]
 18%|█▊        | 92/512 [10:24<46:50,  6.69s/it]
                                                
{'loss': '0.006979', 'grad_norm': '0.5962', 'learning_rate': '9.598e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.81', 'tokens/total': 2784672, 'tokens/trainable': 41927, 'epoch': '0.3594'}
-
 18%|█▊        | 92/512 [10:24<46:50,  6.69s/it]
 18%|█▊        | 93/512 [10:31<46:48,  6.70s/it]
                                                
{'loss': '0.002276', 'grad_norm': '0.2272', 'learning_rate': '9.586e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.8', 'tokens/total': 2815120, 'tokens/trainable': 42397, 'epoch': '0.3633'}
-
 18%|█▊        | 93/512 [10:31<46:48,  6.70s/it]
 18%|█▊        | 94/512 [10:38<46:45,  6.71s/it]
                                                
{'loss': '0.0007734', 'grad_norm': '0.04159', 'learning_rate': '9.574e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 2845600, 'tokens/trainable': 42883, 'epoch': '0.3672'}
-
 18%|█▊        | 94/512 [10:38<46:45,  6.71s/it]
 19%|█▊        | 95/512 [10:44<46:44,  6.72s/it]
                                                
{'loss': '0.003535', 'grad_norm': '0.2429', 'learning_rate': '9.562e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.65', 'tokens/total': 2875952, 'tokens/trainable': 43381, 'epoch': '0.3711'}
-
 19%|█▊        | 95/512 [10:44<46:44,  6.72s/it]
 19%|█▉        | 96/512 [10:51<46:39,  6.73s/it]
                                                
{'loss': '0.0004532', 'grad_norm': '0.05062', 'learning_rate': '9.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 2906288, 'tokens/trainable': 43824, 'epoch': '0.375'}
-
 19%|█▉        | 96/512 [10:51<46:39,  6.73s/it][2026-08-18 12:50:07,343] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
-
 19%|█▉        | 97/512 [11:00<50:11,  7.26s/it]
                                                
{'loss': '0.0004007', 'grad_norm': '0.06917', 'learning_rate': '9.536e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '31.5', 'tokens/total': 2936752, 'tokens/trainable': 44255, 'epoch': '0.3789'}
-
 19%|█▉        | 97/512 [11:00<50:11,  7.26s/it]
 19%|█▉        | 98/512 [11:06<48:56,  7.09s/it]
                                                
{'loss': '0.002073', 'grad_norm': '0.2096', 'learning_rate': '9.523e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.19', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 2967072, 'tokens/trainable': 44739, 'epoch': '0.3828'}
-
 19%|█▉        | 98/512 [11:06<48:56,  7.09s/it]
 19%|█▉        | 99/512 [11:13<48:15,  7.01s/it]
                                                
{'loss': '0.0004412', 'grad_norm': '0.07389', 'learning_rate': '9.51e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '35.76', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 2997664, 'tokens/trainable': 45202, 'epoch': '0.3867'}
-
 19%|█▉        | 99/512 [11:13<48:15,  7.01s/it]
 20%|█▉        | 100/512 [11:20<47:29,  6.92s/it]
                                                 
{'loss': '0.00618', 'grad_norm': '0.4641', 'learning_rate': '9.497e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 3028032, 'tokens/trainable': 45646, 'epoch': '0.3906'}
-
 20%|█▉        | 100/512 [11:20<47:29,  6.92s/it]
 20%|█▉        | 101/512 [11:27<46:58,  6.86s/it]
                                                 
{'loss': '0.0001253', 'grad_norm': '0.01903', 'learning_rate': '9.484e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.04', 'tokens/total': 3058272, 'tokens/trainable': 46089, 'epoch': '0.3945'}
-
 20%|█▉        | 101/512 [11:27<46:58,  6.86s/it]
 20%|█▉        | 102/512 [11:33<46:37,  6.82s/it]
                                                 
{'loss': '0.03122', 'grad_norm': '3.644', 'learning_rate': '9.47e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 3088848, 'tokens/trainable': 46547, 'epoch': '0.3984'}
-
 20%|█▉        | 102/512 [11:33<46:37,  6.82s/it]
 20%|██        | 103/512 [11:40<46:16,  6.79s/it]
                                                 
{'loss': '0.01116', 'grad_norm': '1.234', 'learning_rate': '9.456e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '31.9', 'tokens/total': 3119424, 'tokens/trainable': 47014, 'epoch': '0.4023'}
-
 20%|██        | 103/512 [11:40<46:16,  6.79s/it]
 20%|██        | 104/512 [11:47<46:01,  6.77s/it]
                                                 
{'loss': '0.002164', 'grad_norm': '0.712', 'learning_rate': '9.442e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.65', 'tokens/total': 3149696, 'tokens/trainable': 47503, 'epoch': '0.4062'}
-
 20%|██        | 104/512 [11:47<46:01,  6.77s/it]
 21%|██        | 105/512 [11:53<45:50,  6.76s/it]
                                                 
{'loss': '0.0002605', 'grad_norm': '0.07053', 'learning_rate': '9.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.26', 'tokens/total': 3180064, 'tokens/trainable': 47996, 'epoch': '0.4102'}
-
 21%|██        | 105/512 [11:53<45:50,  6.76s/it]
 21%|██        | 106/512 [12:00<44:35,  6.59s/it]
                                                 
{'loss': '0.0164', 'grad_norm': '1.238', 'learning_rate': '9.414e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 3208320, 'tokens/trainable': 48447, 'epoch': '0.4141'}
-
 21%|██        | 106/512 [12:00<44:35,  6.59s/it]
 21%|██        | 107/512 [12:06<44:41,  6.62s/it]
                                                 
{'loss': '0.01001', 'grad_norm': '0.5694', 'learning_rate': '9.4e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.65', 'tokens/total': 3238416, 'tokens/trainable': 48889, 'epoch': '0.418'}
-
 21%|██        | 107/512 [12:06<44:41,  6.62s/it]
 21%|██        | 108/512 [12:13<44:46,  6.65s/it]
                                                 
{'loss': '0.002217', 'grad_norm': '0.5011', 'learning_rate': '9.385e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.84', 'tokens/total': 3268784, 'tokens/trainable': 49360, 'epoch': '0.4219'}
-
 21%|██        | 108/512 [12:13<44:46,  6.65s/it]
 21%|██▏       | 109/512 [12:20<44:48,  6.67s/it]
                                                 
{'loss': '0.006433', 'grad_norm': '0.6332', 'learning_rate': '9.37e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 3298912, 'tokens/trainable': 49827, 'epoch': '0.4258'}
-
 21%|██▏       | 109/512 [12:20<44:48,  6.67s/it]
 21%|██▏       | 110/512 [12:27<44:48,  6.69s/it]
                                                 
{'loss': '0.02577', 'grad_norm': '3.647', 'learning_rate': '9.355e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 3329232, 'tokens/trainable': 50267, 'epoch': '0.4297'}
-
 21%|██▏       | 110/512 [12:27<44:48,  6.69s/it]
 22%|██▏       | 111/512 [12:33<44:46,  6.70s/it]
                                                 
{'loss': '0.01398', 'grad_norm': '1.08', 'learning_rate': '9.34e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.32', 'tokens/total': 3359440, 'tokens/trainable': 50711, 'epoch': '0.4336'}
-
 22%|██▏       | 111/512 [12:33<44:46,  6.70s/it]
 22%|██▏       | 112/512 [12:40<44:38,  6.70s/it]
                                                 
{'loss': '0.008084', 'grad_norm': '0.3129', 'learning_rate': '9.325e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.54', 'tokens/total': 3389472, 'tokens/trainable': 51157, 'epoch': '0.4375'}
-
 22%|██▏       | 112/512 [12:40<44:38,  6.70s/it]
 22%|██▏       | 113/512 [12:47<44:36,  6.71s/it]
                                                 
{'loss': '0.01252', 'grad_norm': '0.5562', 'learning_rate': '9.31e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.67', 'tokens/total': 3419920, 'tokens/trainable': 51613, 'epoch': '0.4414'}
-
 22%|██▏       | 113/512 [12:47<44:36,  6.71s/it]
 22%|██▏       | 114/512 [12:53<44:26,  6.70s/it]
                                                 
{'loss': '0.01198', 'grad_norm': '0.3786', 'learning_rate': '9.294e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '29.04', 'tokens/total': 3450176, 'tokens/trainable': 52017, 'epoch': '0.4453'}
-
 22%|██▏       | 114/512 [12:53<44:26,  6.70s/it]
 22%|██▏       | 115/512 [13:00<44:22,  6.71s/it]
                                                 
{'loss': '0.003319', 'grad_norm': '0.1905', 'learning_rate': '9.278e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 3480544, 'tokens/trainable': 52506, 'epoch': '0.4492'}
-
 22%|██▏       | 115/512 [13:00<44:22,  6.71s/it]
 23%|██▎       | 116/512 [13:07<44:19,  6.72s/it]
                                                 
{'loss': '0.01132', 'grad_norm': '0.9839', 'learning_rate': '9.263e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '38.61', 'tokens/total': 3510848, 'tokens/trainable': 53008, 'epoch': '0.4531'}
-
 23%|██▎       | 116/512 [13:07<44:19,  6.72s/it]
 23%|██▎       | 117/512 [13:14<44:12,  6.72s/it]
                                                 
{'loss': '0.01193', 'grad_norm': '1.508', 'learning_rate': '9.247e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 3541296, 'tokens/trainable': 53467, 'epoch': '0.457'}
-
 23%|██▎       | 117/512 [13:14<44:12,  6.72s/it]
 23%|██▎       | 118/512 [13:20<44:09,  6.72s/it]
                                                 
{'loss': '0.002887', 'grad_norm': '0.1703', 'learning_rate': '9.23e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 3571936, 'tokens/trainable': 53923, 'epoch': '0.4609'}
-
 23%|██▎       | 118/512 [13:20<44:09,  6.72s/it]
 23%|██▎       | 119/512 [13:27<43:55,  6.71s/it]
                                                 
{'loss': '0.003508', 'grad_norm': '0.3885', 'learning_rate': '9.214e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 3602000, 'tokens/trainable': 54363, 'epoch': '0.4648'}
-
 23%|██▎       | 119/512 [13:27<43:55,  6.71s/it]
 23%|██▎       | 120/512 [13:34<43:48,  6.71s/it]
                                                 
{'loss': '0.005708', 'grad_norm': '0.2821', 'learning_rate': '9.198e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 3632288, 'tokens/trainable': 54839, 'epoch': '0.4688'}
-
 23%|██▎       | 120/512 [13:34<43:48,  6.71s/it]
 24%|██▎       | 121/512 [13:40<43:40,  6.70s/it]
                                                 
{'loss': '0.003157', 'grad_norm': '0.3879', 'learning_rate': '9.181e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.58', 'tokens/total': 3662608, 'tokens/trainable': 55267, 'epoch': '0.4727'}
-
 24%|██▎       | 121/512 [13:40<43:40,  6.70s/it]
 24%|██▍       | 122/512 [13:47<43:33,  6.70s/it]
                                                 
{'loss': '0.00323', 'grad_norm': '0.2678', 'learning_rate': '9.164e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.47', 'tokens/total': 3692864, 'tokens/trainable': 55700, 'epoch': '0.4766'}
-
 24%|██▍       | 122/512 [13:47<43:33,  6.70s/it]
 24%|██▍       | 123/512 [13:54<43:20,  6.69s/it]
                                                 
{'loss': '0.004257', 'grad_norm': '0.31', 'learning_rate': '9.147e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.38', 'tokens/total': 3722864, 'tokens/trainable': 56144, 'epoch': '0.4805'}
-
 24%|██▍       | 123/512 [13:54<43:20,  6.69s/it]
 24%|██▍       | 124/512 [14:00<43:14,  6.69s/it]
                                                 
{'loss': '0.0001001', 'grad_norm': '0.03114', 'learning_rate': '9.13e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 3753056, 'tokens/trainable': 56589, 'epoch': '0.4844'}
-
 24%|██▍       | 124/512 [14:00<43:14,  6.69s/it]
 24%|██▍       | 125/512 [14:07<43:13,  6.70s/it]
                                                 
{'loss': '0.004621', 'grad_norm': '0.736', 'learning_rate': '9.113e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.64', 'tokens/total': 3783312, 'tokens/trainable': 57081, 'epoch': '0.4883'}
-
 24%|██▍       | 125/512 [14:07<43:13,  6.70s/it]
 25%|██▍       | 126/512 [14:14<43:04,  6.70s/it]
                                                 
{'loss': '0.0002661', 'grad_norm': '0.05499', 'learning_rate': '9.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 3813712, 'tokens/trainable': 57527, 'epoch': '0.4922'}
-
 25%|██▍       | 126/512 [14:14<43:04,  6.70s/it]
 25%|██▍       | 127/512 [14:21<43:04,  6.71s/it]
                                                 
{'loss': '0.003466', 'grad_norm': '0.416', 'learning_rate': '9.078e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 3844336, 'tokens/trainable': 58005, 'epoch': '0.4961'}
-
 25%|██▍       | 127/512 [14:21<43:04,  6.71s/it]
 25%|██▌       | 128/512 [14:27<41:58,  6.56s/it]
                                                 
{'loss': '0.002959', 'grad_norm': '0.4412', 'learning_rate': '9.06e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 3872688, 'tokens/trainable': 58454, 'epoch': '0.5'}
-
 25%|██▌       | 128/512 [14:27<41:58,  6.56s/it][2026-08-18 12:53:42,959] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
-
 25%|██▌       | 129/512 [14:35<45:35,  7.14s/it]
                                                 
{'loss': '0.002471', 'grad_norm': '0.2943', 'learning_rate': '9.043e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 3902640, 'tokens/trainable': 58896, 'epoch': '0.5039'}
-
 25%|██▌       | 129/512 [14:35<45:35,  7.14s/it]
 25%|██▌       | 130/512 [14:42<44:38,  7.01s/it]
                                                 
{'loss': '0.0001329', 'grad_norm': '0.02422', 'learning_rate': '9.025e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '35.79', 'tokens/total': 3933104, 'tokens/trainable': 59356, 'epoch': '0.5078'}
-
 25%|██▌       | 130/512 [14:42<44:38,  7.01s/it]
 26%|██▌       | 131/512 [14:49<43:58,  6.93s/it]
                                                 
{'loss': '0.01349', 'grad_norm': '3.24', 'learning_rate': '9.007e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '32.14', 'tokens/total': 3963184, 'tokens/trainable': 59792, 'epoch': '0.5117'}
-
 26%|██▌       | 131/512 [14:49<43:58,  6.93s/it]
 26%|██▌       | 132/512 [14:55<43:26,  6.86s/it]
                                                 
{'loss': '0.0009926', 'grad_norm': '0.1477', 'learning_rate': '8.988e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.66', 'tokens/total': 3993648, 'tokens/trainable': 60266, 'epoch': '0.5156'}
-
 26%|██▌       | 132/512 [14:55<43:26,  6.86s/it]
 26%|██▌       | 133/512 [15:02<43:04,  6.82s/it]
                                                 
{'loss': '0.005481', 'grad_norm': '1.215', 'learning_rate': '8.97e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 4024000, 'tokens/trainable': 60759, 'epoch': '0.5195'}
-
 26%|██▌       | 133/512 [15:02<43:04,  6.82s/it]
 26%|██▌       | 134/512 [15:08<41:49,  6.64s/it]
                                                 
{'loss': '0.001138', 'grad_norm': '0.2438', 'learning_rate': '8.951e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 4052480, 'tokens/trainable': 61204, 'epoch': '0.5234'}
-
 26%|██▌       | 134/512 [15:08<41:49,  6.64s/it]
 26%|██▋       | 135/512 [15:15<41:54,  6.67s/it]
                                                 
{'loss': '0.004932', 'grad_norm': '0.8191', 'learning_rate': '8.933e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.88', 'tokens/total': 4082864, 'tokens/trainable': 61684, 'epoch': '0.5273'}
-
 26%|██▋       | 135/512 [15:15<41:54,  6.67s/it]
 27%|██▋       | 136/512 [15:22<41:46,  6.67s/it]
                                                 
{'loss': '0.009538', 'grad_norm': '0.7444', 'learning_rate': '8.914e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '38.18', 'tokens/total': 4113040, 'tokens/trainable': 62159, 'epoch': '0.5312'}
-
 27%|██▋       | 136/512 [15:22<41:46,  6.67s/it]
 27%|██▋       | 137/512 [15:28<41:45,  6.68s/it]
                                                 
{'loss': '0.0001035', 'grad_norm': '0.0178', 'learning_rate': '8.895e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 4143344, 'tokens/trainable': 62649, 'epoch': '0.5352'}
-
 27%|██▋       | 137/512 [15:28<41:45,  6.68s/it]
 27%|██▋       | 138/512 [15:35<41:41,  6.69s/it]
                                                 
{'loss': '0.0006722', 'grad_norm': '0.1413', 'learning_rate': '8.876e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 4173648, 'tokens/trainable': 63100, 'epoch': '0.5391'}
-
 27%|██▋       | 138/512 [15:35<41:41,  6.69s/it]
 27%|██▋       | 139/512 [15:42<41:41,  6.71s/it]
                                                 
{'loss': '0.00706', 'grad_norm': '0.7884', 'learning_rate': '8.856e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.51', 'tokens/total': 4204080, 'tokens/trainable': 63588, 'epoch': '0.543'}
-
 27%|██▋       | 139/512 [15:42<41:41,  6.71s/it]
 27%|██▋       | 140/512 [15:48<40:32,  6.54s/it]
                                                 
{'loss': '0.04659', 'grad_norm': '2.152', 'learning_rate': '8.837e-05', 'ppl': '1.048', 'memory/max_active (GiB)': '33.42', 'memory/max_allocated (GiB)': '33.42', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 4232400, 'tokens/trainable': 64005, 'epoch': '0.5469'}
-
 27%|██▋       | 140/512 [15:48<40:32,  6.54s/it]
 28%|██▊       | 141/512 [15:55<40:47,  6.60s/it]
                                                 
{'loss': '0.0185', 'grad_norm': '0.9182', 'learning_rate': '8.817e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 4262624, 'tokens/trainable': 64462, 'epoch': '0.5508'}
-
 28%|██▊       | 141/512 [15:55<40:47,  6.60s/it]
 28%|██▊       | 142/512 [16:02<40:55,  6.64s/it]
                                                 
{'loss': '0.0007529', 'grad_norm': '0.04982', 'learning_rate': '8.798e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 4293168, 'tokens/trainable': 64900, 'epoch': '0.5547'}
-
 28%|██▊       | 142/512 [16:02<40:55,  6.64s/it]
 28%|██▊       | 143/512 [16:08<40:56,  6.66s/it]
                                                 
{'loss': '0.01302', 'grad_norm': '0.4962', 'learning_rate': '8.778e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.45', 'tokens/total': 4323248, 'tokens/trainable': 65379, 'epoch': '0.5586'}
-
 28%|██▊       | 143/512 [16:08<40:56,  6.66s/it]
 28%|██▊       | 144/512 [16:15<40:52,  6.66s/it]
                                                 
{'loss': '0.001344', 'grad_norm': '0.09981', 'learning_rate': '8.758e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 4353344, 'tokens/trainable': 65845, 'epoch': '0.5625'}
-
 28%|██▊       | 144/512 [16:15<40:52,  6.66s/it]
 28%|██▊       | 145/512 [16:22<40:49,  6.67s/it]
                                                 
{'loss': '0.005669', 'grad_norm': '0.25', 'learning_rate': '8.738e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 4383584, 'tokens/trainable': 66285, 'epoch': '0.5664'}
-
 28%|██▊       | 145/512 [16:22<40:49,  6.67s/it]
 29%|██▊       | 146/512 [16:28<40:49,  6.69s/it]
                                                 
{'loss': '0.03146', 'grad_norm': '0.8081', 'learning_rate': '8.718e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 4413856, 'tokens/trainable': 66736, 'epoch': '0.5703'}
-
 29%|██▊       | 146/512 [16:28<40:49,  6.69s/it]
 29%|██▊       | 147/512 [16:35<40:44,  6.70s/it]
                                                 
{'loss': '0.008184', 'grad_norm': '0.2932', 'learning_rate': '8.697e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 4444112, 'tokens/trainable': 67150, 'epoch': '0.5742'}
-
 29%|██▊       | 147/512 [16:35<40:44,  6.70s/it]
 29%|██▉       | 148/512 [16:42<40:45,  6.72s/it]
                                                 
{'loss': '0.008484', 'grad_norm': '0.5703', 'learning_rate': '8.677e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 4474464, 'tokens/trainable': 67633, 'epoch': '0.5781'}
-
 29%|██▉       | 148/512 [16:42<40:45,  6.72s/it]
 29%|██▉       | 149/512 [16:49<40:39,  6.72s/it]
                                                 
{'loss': '0.002346', 'grad_norm': '0.1261', 'learning_rate': '8.656e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 4504880, 'tokens/trainable': 68077, 'epoch': '0.582'}
-
 29%|██▉       | 149/512 [16:49<40:39,  6.72s/it]
 29%|██▉       | 150/512 [16:55<40:28,  6.71s/it]
                                                 
{'loss': '0.0008666', 'grad_norm': '0.04483', 'learning_rate': '8.635e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 4534976, 'tokens/trainable': 68545, 'epoch': '0.5859'}
-
 29%|██▉       | 150/512 [16:55<40:28,  6.71s/it]
 29%|██▉       | 151/512 [17:02<40:24,  6.72s/it]
                                                 
{'loss': '0.01069', 'grad_norm': '0.2727', 'learning_rate': '8.615e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.17', 'tokens/total': 4565088, 'tokens/trainable': 69011, 'epoch': '0.5898'}
-
 29%|██▉       | 151/512 [17:02<40:24,  6.72s/it]
 30%|██▉       | 152/512 [17:09<40:20,  6.72s/it]
                                                 
{'loss': '0.007179', 'grad_norm': '0.6334', 'learning_rate': '8.594e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 4595600, 'tokens/trainable': 69465, 'epoch': '0.5938'}
-
 30%|██▉       | 152/512 [17:09<40:20,  6.72s/it]
 30%|██▉       | 153/512 [17:15<40:18,  6.74s/it]
                                                 
{'loss': '0.005559', 'grad_norm': '0.6208', 'learning_rate': '8.572e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.83', 'tokens/total': 4626032, 'tokens/trainable': 69928, 'epoch': '0.5977'}
-
 30%|██▉       | 153/512 [17:15<40:18,  6.74s/it]
 30%|███       | 154/512 [17:22<40:09,  6.73s/it]
                                                 
{'loss': '0.0009', 'grad_norm': '0.1435', 'learning_rate': '8.551e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.64', 'tokens/total': 4656352, 'tokens/trainable': 70395, 'epoch': '0.6016'}
-
 30%|███       | 154/512 [17:22<40:09,  6.73s/it]
 30%|███       | 155/512 [17:28<39:05,  6.57s/it]
                                                 
{'loss': '0.003266', 'grad_norm': '0.3754', 'learning_rate': '8.53e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 4684672, 'tokens/trainable': 70829, 'epoch': '0.6055'}
-
 30%|███       | 155/512 [17:28<39:05,  6.57s/it]
 30%|███       | 156/512 [17:35<39:18,  6.62s/it]
                                                 
{'loss': '0.01512', 'grad_norm': '0.8183', 'learning_rate': '8.508e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.8', 'tokens/total': 4715216, 'tokens/trainable': 71287, 'epoch': '0.6094'}
-
 30%|███       | 156/512 [17:35<39:18,  6.62s/it]
 31%|███       | 157/512 [17:42<39:22,  6.65s/it]
                                                 
{'loss': '0.0002059', 'grad_norm': '0.05925', 'learning_rate': '8.487e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 4745376, 'tokens/trainable': 71749, 'epoch': '0.6133'}
-
 31%|███       | 157/512 [17:42<39:22,  6.65s/it]
 31%|███       | 158/512 [17:49<39:28,  6.69s/it]
                                                 
{'loss': '0.01122', 'grad_norm': '0.5881', 'learning_rate': '8.465e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 4775776, 'tokens/trainable': 72185, 'epoch': '0.6172'}
-
 31%|███       | 158/512 [17:49<39:28,  6.69s/it]
 31%|███       | 159/512 [17:55<39:18,  6.68s/it]
                                                 
{'loss': '0.001423', 'grad_norm': '0.188', 'learning_rate': '8.443e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.51', 'tokens/total': 4805856, 'tokens/trainable': 72604, 'epoch': '0.6211'}
-
 31%|███       | 159/512 [17:55<39:18,  6.68s/it]
 31%|███▏      | 160/512 [18:02<39:18,  6.70s/it]
                                                 
{'loss': '0.01892', 'grad_norm': '0.8338', 'learning_rate': '8.421e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.41', 'tokens/total': 4836304, 'tokens/trainable': 73061, 'epoch': '0.625'}
-
 31%|███▏      | 160/512 [18:02<39:18,  6.70s/it][2026-08-18 12:57:18,242] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
-
 31%|███▏      | 161/512 [18:11<42:48,  7.32s/it]
                                                 
{'loss': '0.001153', 'grad_norm': '0.1115', 'learning_rate': '8.399e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.97', 'tokens/total': 4866544, 'tokens/trainable': 73474, 'epoch': '0.6289'}
-
 31%|███▏      | 161/512 [18:11<42:48,  7.32s/it]
 32%|███▏      | 162/512 [18:18<41:41,  7.15s/it]
                                                 
{'loss': '0.00308', 'grad_norm': '0.3199', 'learning_rate': '8.376e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.61', 'tokens/total': 4897120, 'tokens/trainable': 73923, 'epoch': '0.6328'}
-
 32%|███▏      | 162/512 [18:18<41:41,  7.15s/it]
 32%|███▏      | 163/512 [18:24<40:49,  7.02s/it]
                                                 
{'loss': '0.002999', 'grad_norm': '0.2323', 'learning_rate': '8.354e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 4927440, 'tokens/trainable': 74379, 'epoch': '0.6367'}
-
 32%|███▏      | 163/512 [18:24<40:49,  7.02s/it]
 32%|███▏      | 164/512 [18:31<40:27,  6.97s/it]
                                                 
{'loss': '0.003661', 'grad_norm': '0.1695', 'learning_rate': '8.332e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 4958032, 'tokens/trainable': 74830, 'epoch': '0.6406'}
-
 32%|███▏      | 164/512 [18:31<40:27,  6.97s/it]
 32%|███▏      | 165/512 [18:38<39:53,  6.90s/it]
                                                 
{'loss': '0.001191', 'grad_norm': '0.0809', 'learning_rate': '8.309e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 4988368, 'tokens/trainable': 75262, 'epoch': '0.6445'}
-
 32%|███▏      | 165/512 [18:38<39:53,  6.90s/it]
 32%|███▏      | 166/512 [18:45<39:23,  6.83s/it]
                                                 
{'loss': '0.001098', 'grad_norm': '0.06333', 'learning_rate': '8.286e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.75', 'tokens/total': 5018672, 'tokens/trainable': 75703, 'epoch': '0.6484'}
-
 32%|███▏      | 166/512 [18:45<39:23,  6.83s/it]
 33%|███▎      | 167/512 [18:51<39:15,  6.83s/it]
                                                 
{'loss': '0.0008549', 'grad_norm': '0.05864', 'learning_rate': '8.263e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.26', 'tokens/total': 5049344, 'tokens/trainable': 76164, 'epoch': '0.6523'}
-
 33%|███▎      | 167/512 [18:51<39:15,  6.83s/it]
 33%|███▎      | 168/512 [18:58<38:59,  6.80s/it]
                                                 
{'loss': '0.002419', 'grad_norm': '0.192', 'learning_rate': '8.24e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.42', 'tokens/total': 5079856, 'tokens/trainable': 76623, 'epoch': '0.6562'}
-
 33%|███▎      | 168/512 [18:58<38:59,  6.80s/it]
 33%|███▎      | 169/512 [19:05<38:39,  6.76s/it]
                                                 
{'loss': '0.000676', 'grad_norm': '0.03395', 'learning_rate': '8.217e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 5110192, 'tokens/trainable': 77047, 'epoch': '0.6602'}
-
 33%|███▎      | 169/512 [19:05<38:39,  6.76s/it]
 33%|███▎      | 170/512 [19:12<38:33,  6.76s/it]
                                                 
{'loss': '0.0006985', 'grad_norm': '0.1052', 'learning_rate': '8.194e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 5140592, 'tokens/trainable': 77496, 'epoch': '0.6641'}
-
 33%|███▎      | 170/512 [19:12<38:33,  6.76s/it]
 33%|███▎      | 171/512 [19:18<38:20,  6.75s/it]
                                                 
{'loss': '0.0002491', 'grad_norm': '0.01598', 'learning_rate': '8.171e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.13', 'tokens/total': 5170848, 'tokens/trainable': 77956, 'epoch': '0.668'}
-
 33%|███▎      | 171/512 [19:18<38:20,  6.75s/it]
 34%|███▎      | 172/512 [19:25<38:14,  6.75s/it]
                                                 
{'loss': '0.002719', 'grad_norm': '0.2858', 'learning_rate': '8.147e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.75', 'tokens/total': 5201280, 'tokens/trainable': 78410, 'epoch': '0.6719'}
-
 34%|███▎      | 172/512 [19:25<38:14,  6.75s/it]
 34%|███▍      | 173/512 [19:32<37:59,  6.72s/it]
                                                 
{'loss': '0.0005204', 'grad_norm': '0.02717', 'learning_rate': '8.124e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.04', 'tokens/total': 5231712, 'tokens/trainable': 78849, 'epoch': '0.6758'}
-
 34%|███▍      | 173/512 [19:32<37:59,  6.72s/it]
 34%|███▍      | 174/512 [19:38<37:47,  6.71s/it]
                                                 
{'loss': '0.01529', 'grad_norm': '2.88', 'learning_rate': '8.1e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.62', 'memory/max_allocated (GiB)': '33.62', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 5261440, 'tokens/trainable': 79266, 'epoch': '0.6797'}
-
 34%|███▍      | 174/512 [19:38<37:47,  6.71s/it]
 34%|███▍      | 175/512 [19:45<37:42,  6.71s/it]
                                                 
{'loss': '0.001033', 'grad_norm': '0.1005', 'learning_rate': '8.076e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.42', 'tokens/total': 5291728, 'tokens/trainable': 79736, 'epoch': '0.6836'}
-
 34%|███▍      | 175/512 [19:45<37:42,  6.71s/it]
 34%|███▍      | 176/512 [19:52<37:40,  6.73s/it]
                                                 
{'loss': '0.003816', 'grad_norm': '0.6506', 'learning_rate': '8.053e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.12', 'tokens/total': 5322240, 'tokens/trainable': 80214, 'epoch': '0.6875'}
-
 34%|███▍      | 176/512 [19:52<37:40,  6.73s/it]
 35%|███▍      | 177/512 [19:59<37:31,  6.72s/it]
                                                 
{'loss': '0.002027', 'grad_norm': '0.199', 'learning_rate': '8.029e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.51', 'tokens/total': 5352576, 'tokens/trainable': 80661, 'epoch': '0.6914'}
-
 35%|███▍      | 177/512 [19:59<37:31,  6.72s/it]
 35%|███▍      | 178/512 [20:05<37:25,  6.72s/it]
                                                 
{'loss': '0.0001421', 'grad_norm': '0.00993', 'learning_rate': '8.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.81', 'tokens/total': 5382784, 'tokens/trainable': 81129, 'epoch': '0.6953'}
-
 35%|███▍      | 178/512 [20:05<37:25,  6.72s/it]
 35%|███▍      | 179/512 [20:12<37:11,  6.70s/it]
                                                 
{'loss': '0.02322', 'grad_norm': '0.5209', 'learning_rate': '7.98e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 5412784, 'tokens/trainable': 81576, 'epoch': '0.6992'}
-
 35%|███▍      | 179/512 [20:12<37:11,  6.70s/it]
 35%|███▌      | 180/512 [20:19<37:07,  6.71s/it]
                                                 
{'loss': '0.0002464', 'grad_norm': '0.02281', 'learning_rate': '7.956e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.49', 'tokens/total': 5442960, 'tokens/trainable': 82049, 'epoch': '0.7031'}
-
 35%|███▌      | 180/512 [20:19<37:07,  6.71s/it]
 35%|███▌      | 181/512 [20:25<37:05,  6.72s/it]
                                                 
{'loss': '0.01167', 'grad_norm': '0.377', 'learning_rate': '7.932e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '38.75', 'tokens/total': 5473216, 'tokens/trainable': 82566, 'epoch': '0.707'}
-
 35%|███▌      | 181/512 [20:25<37:05,  6.72s/it]
 36%|███▌      | 182/512 [20:32<37:00,  6.73s/it]
                                                 
{'loss': '0.0002162', 'grad_norm': '0.06076', 'learning_rate': '7.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.83', 'tokens/total': 5503568, 'tokens/trainable': 83054, 'epoch': '0.7109'}
-
 36%|███▌      | 182/512 [20:32<37:00,  6.73s/it]
 36%|███▌      | 183/512 [20:39<36:52,  6.73s/it]
                                                 
{'loss': '0.0004136', 'grad_norm': '0.03374', 'learning_rate': '7.883e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 5533760, 'tokens/trainable': 83501, 'epoch': '0.7148'}
-
 36%|███▌      | 183/512 [20:39<36:52,  6.73s/it]
 36%|███▌      | 184/512 [20:46<36:45,  6.72s/it]
                                                 
{'loss': '0.00041', 'grad_norm': '0.03666', 'learning_rate': '7.858e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 5564240, 'tokens/trainable': 84007, 'epoch': '0.7188'}
-
 36%|███▌      | 184/512 [20:46<36:45,  6.72s/it]
 36%|███▌      | 185/512 [20:52<36:41,  6.73s/it]
                                                 
{'loss': '0.0003714', 'grad_norm': '0.03234', 'learning_rate': '7.833e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 5594448, 'tokens/trainable': 84491, 'epoch': '0.7227'}
-
 36%|███▌      | 185/512 [20:52<36:41,  6.73s/it]
 36%|███▋      | 186/512 [20:59<36:36,  6.74s/it]
                                                 
{'loss': '0.0003759', 'grad_norm': '0.03077', 'learning_rate': '7.808e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 5624816, 'tokens/trainable': 84952, 'epoch': '0.7266'}
-
 36%|███▋      | 186/512 [20:59<36:36,  6.74s/it]
 37%|███▋      | 187/512 [21:06<36:28,  6.74s/it]
                                                 
{'loss': '0.002438', 'grad_norm': '0.6632', 'learning_rate': '7.783e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 5655104, 'tokens/trainable': 85400, 'epoch': '0.7305'}
-
 37%|███▋      | 187/512 [21:06<36:28,  6.74s/it]
 37%|███▋      | 188/512 [21:13<36:23,  6.74s/it]
                                                 
{'loss': '0.002341', 'grad_norm': '0.2652', 'learning_rate': '7.758e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 5685360, 'tokens/trainable': 85889, 'epoch': '0.7344'}
-
 37%|███▋      | 188/512 [21:13<36:23,  6.74s/it]
 37%|███▋      | 189/512 [21:19<36:18,  6.75s/it]
                                                 
{'loss': '0.001637', 'grad_norm': '0.3055', 'learning_rate': '7.733e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.11', 'tokens/total': 5715648, 'tokens/trainable': 86396, 'epoch': '0.7383'}
-
 37%|███▋      | 189/512 [21:19<36:18,  6.75s/it]
 37%|███▋      | 190/512 [21:26<36:10,  6.74s/it]
                                                 
{'loss': '0.00207', 'grad_norm': '0.2344', 'learning_rate': '7.708e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.75', 'tokens/total': 5746240, 'tokens/trainable': 86867, 'epoch': '0.7422'}
-
 37%|███▋      | 190/512 [21:26<36:10,  6.74s/it]
 37%|███▋      | 191/512 [21:33<36:07,  6.75s/it]
                                                 
{'loss': '0.0006176', 'grad_norm': '0.09677', 'learning_rate': '7.683e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.03', 'tokens/total': 5776816, 'tokens/trainable': 87314, 'epoch': '0.7461'}
-
 37%|███▋      | 191/512 [21:33<36:07,  6.75s/it]
 38%|███▊      | 192/512 [21:39<35:55,  6.74s/it]
                                                 
{'loss': '0.000425', 'grad_norm': '0.05828', 'learning_rate': '7.657e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.29', 'tokens/total': 5807248, 'tokens/trainable': 87736, 'epoch': '0.75'}
-
 38%|███▊      | 192/512 [21:40<35:55,  6.74s/it][2026-08-18 13:00:55,710] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
-
 38%|███▊      | 193/512 [21:48<38:51,  7.31s/it]
                                                 
{'loss': '0.000321', 'grad_norm': '0.09241', 'learning_rate': '7.632e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 5837680, 'tokens/trainable': 88219, 'epoch': '0.7539'}
-
 38%|███▊      | 193/512 [21:48<38:51,  7.31s/it]
 38%|███▊      | 194/512 [21:55<37:51,  7.14s/it]
                                                 
{'loss': '7.824e-05', 'grad_norm': '0.005906', 'learning_rate': '7.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '36.11', 'tokens/total': 5868000, 'tokens/trainable': 88699, 'epoch': '0.7578'}
-
 38%|███▊      | 194/512 [21:55<37:51,  7.14s/it]
 38%|███▊      | 195/512 [22:01<36:11,  6.85s/it]
                                                 
{'loss': '0.001962', 'grad_norm': '0.4526', 'learning_rate': '7.58e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.36', 'tokens/total': 5896224, 'tokens/trainable': 89132, 'epoch': '0.7617'}
-
 38%|███▊      | 195/512 [22:01<36:11,  6.85s/it]
 38%|███▊      | 196/512 [22:08<35:52,  6.81s/it]
                                                 
{'loss': '0.000491', 'grad_norm': '0.03763', 'learning_rate': '7.555e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '37.21', 'tokens/total': 5926624, 'tokens/trainable': 89594, 'epoch': '0.7656'}
-
 38%|███▊      | 196/512 [22:08<35:52,  6.81s/it]
 38%|███▊      | 197/512 [22:14<34:49,  6.63s/it]
                                                 
{'loss': '0.005962', 'grad_norm': '0.3791', 'learning_rate': '7.529e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.34', 'memory/max_allocated (GiB)': '33.34', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '37.98', 'tokens/total': 5954944, 'tokens/trainable': 90042, 'epoch': '0.7695'}
-
 38%|███▊      | 197/512 [22:14<34:49,  6.63s/it]
 39%|███▊      | 198/512 [22:21<34:52,  6.66s/it]
                                                 
{'loss': '0.0004046', 'grad_norm': '0.06128', 'learning_rate': '7.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 5985200, 'tokens/trainable': 90503, 'epoch': '0.7734'}
-
 39%|███▊      | 198/512 [22:21<34:52,  6.66s/it]
 39%|███▉      | 199/512 [22:28<35:03,  6.72s/it]
                                                 
{'loss': '0.0001899', 'grad_norm': '0.01201', 'learning_rate': '7.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 6015952, 'tokens/trainable': 90982, 'epoch': '0.7773'}
-
 39%|███▉      | 199/512 [22:28<35:03,  6.72s/it]
 39%|███▉      | 200/512 [22:34<34:58,  6.73s/it]
                                                 
{'loss': '0.0003557', 'grad_norm': '0.03914', 'learning_rate': '7.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '39.22', 'tokens/total': 6046208, 'tokens/trainable': 91466, 'epoch': '0.7812'}
-
 39%|███▉      | 200/512 [22:34<34:58,  6.73s/it]
 39%|███▉      | 201/512 [22:41<34:46,  6.71s/it]
                                                 
{'loss': '0.0002456', 'grad_norm': '0.04544', 'learning_rate': '7.424e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 6076208, 'tokens/trainable': 91912, 'epoch': '0.7852'}
-
 39%|███▉      | 201/512 [22:41<34:46,  6.71s/it]
 39%|███▉      | 202/512 [22:48<34:38,  6.71s/it]
                                                 
{'loss': '0.009876', 'grad_norm': '0.6255', 'learning_rate': '7.398e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.79', 'tokens/total': 6106560, 'tokens/trainable': 92385, 'epoch': '0.7891'}
-
 39%|███▉      | 202/512 [22:48<34:38,  6.71s/it]
 40%|███▉      | 203/512 [22:54<34:30,  6.70s/it]
                                                 
{'loss': '0.01162', 'grad_norm': '0.7777', 'learning_rate': '7.372e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.48', 'tokens/total': 6136640, 'tokens/trainable': 92842, 'epoch': '0.793'}
-
 40%|███▉      | 203/512 [22:54<34:30,  6.70s/it]
 40%|███▉      | 204/512 [23:01<34:26,  6.71s/it]
                                                 
{'loss': '0.0005008', 'grad_norm': '0.07255', 'learning_rate': '7.345e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 6167184, 'tokens/trainable': 93323, 'epoch': '0.7969'}
-
 40%|███▉      | 204/512 [23:01<34:26,  6.71s/it]
 40%|████      | 205/512 [23:08<34:14,  6.69s/it]
                                                 
{'loss': '0.007061', 'grad_norm': '0.6082', 'learning_rate': '7.319e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.75', 'tokens/total': 6197376, 'tokens/trainable': 93756, 'epoch': '0.8008'}
-
 40%|████      | 205/512 [23:08<34:14,  6.69s/it]
 40%|████      | 206/512 [23:15<34:13,  6.71s/it]
                                                 
{'loss': '0.006006', 'grad_norm': '0.4931', 'learning_rate': '7.292e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.47', 'tokens/total': 6227600, 'tokens/trainable': 94235, 'epoch': '0.8047'}
-
 40%|████      | 206/512 [23:15<34:13,  6.71s/it]
 40%|████      | 207/512 [23:21<34:07,  6.71s/it]
                                                 
{'loss': '0.0006859', 'grad_norm': '0.06813', 'learning_rate': '7.266e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.48', 'tokens/total': 6257760, 'tokens/trainable': 94733, 'epoch': '0.8086'}
-
 40%|████      | 207/512 [23:21<34:07,  6.71s/it]
 41%|████      | 208/512 [23:28<34:00,  6.71s/it]
                                                 
{'loss': '0.001039', 'grad_norm': '0.06704', 'learning_rate': '7.239e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.68', 'tokens/total': 6288048, 'tokens/trainable': 95218, 'epoch': '0.8125'}
-
 41%|████      | 208/512 [23:28<34:00,  6.71s/it]
 41%|████      | 209/512 [23:35<33:48,  6.69s/it]
                                                 
{'loss': '0.003791', 'grad_norm': '0.2953', 'learning_rate': '7.212e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '29.17', 'tokens/total': 6318688, 'tokens/trainable': 95619, 'epoch': '0.8164'}
-
 41%|████      | 209/512 [23:35<33:48,  6.69s/it]
 41%|████      | 210/512 [23:41<33:43,  6.70s/it]
                                                 
{'loss': '0.002394', 'grad_norm': '0.3258', 'learning_rate': '7.185e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 6348944, 'tokens/trainable': 96076, 'epoch': '0.8203'}
-
 41%|████      | 210/512 [23:41<33:43,  6.70s/it]
 41%|████      | 211/512 [23:48<33:33,  6.69s/it]
                                                 
{'loss': '0.006002', 'grad_norm': '0.4735', 'learning_rate': '7.158e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.28', 'tokens/total': 6378960, 'tokens/trainable': 96505, 'epoch': '0.8242'}
-
 41%|████      | 211/512 [23:48<33:33,  6.69s/it]
 41%|████▏     | 212/512 [23:55<33:29,  6.70s/it]
                                                 
{'loss': '0.003804', 'grad_norm': '0.1573', 'learning_rate': '7.131e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.57', 'tokens/total': 6409472, 'tokens/trainable': 96930, 'epoch': '0.8281'}
-
 41%|████▏     | 212/512 [23:55<33:29,  6.70s/it]
 42%|████▏     | 213/512 [24:01<33:24,  6.70s/it]
                                                 
{'loss': '0.005038', 'grad_norm': '0.1833', 'learning_rate': '7.104e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 6439904, 'tokens/trainable': 97361, 'epoch': '0.832'}
-
 42%|████▏     | 213/512 [24:01<33:24,  6.70s/it]
 42%|████▏     | 214/512 [24:08<33:22,  6.72s/it]
                                                 
{'loss': '0.002086', 'grad_norm': '0.1782', 'learning_rate': '7.077e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.26', 'tokens/total': 6470176, 'tokens/trainable': 97858, 'epoch': '0.8359'}
-
 42%|████▏     | 214/512 [24:08<33:22,  6.72s/it]
 42%|████▏     | 215/512 [24:15<33:10,  6.70s/it]
                                                 
{'loss': '0.005029', 'grad_norm': '0.2747', 'learning_rate': '7.05e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.41', 'tokens/total': 6500336, 'tokens/trainable': 98308, 'epoch': '0.8398'}
-
 42%|████▏     | 215/512 [24:15<33:10,  6.70s/it]
 42%|████▏     | 216/512 [24:22<33:02,  6.70s/it]
                                                 
{'loss': '0.005489', 'grad_norm': '0.4509', 'learning_rate': '7.022e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 6530688, 'tokens/trainable': 98781, 'epoch': '0.8438'}
-
 42%|████▏     | 216/512 [24:22<33:02,  6.70s/it]
 42%|████▏     | 217/512 [24:28<32:59,  6.71s/it]
                                                 
{'loss': '0.003419', 'grad_norm': '0.2319', 'learning_rate': '6.995e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.67', 'tokens/total': 6561040, 'tokens/trainable': 99266, 'epoch': '0.8477'}
-
 42%|████▏     | 217/512 [24:28<32:59,  6.71s/it]
 43%|████▎     | 218/512 [24:34<32:06,  6.55s/it]
                                                 
{'loss': '0.0009051', 'grad_norm': '0.06469', 'learning_rate': '6.968e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.6', 'tokens/total': 6589392, 'tokens/trainable': 99731, 'epoch': '0.8516'}
-
 43%|████▎     | 218/512 [24:34<32:06,  6.55s/it]
 43%|████▎     | 219/512 [24:41<32:12,  6.60s/it]
                                                 
{'loss': '0.001119', 'grad_norm': '0.06683', 'learning_rate': '6.94e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.66', 'tokens/total': 6619824, 'tokens/trainable': 100164, 'epoch': '0.8555'}
-
 43%|████▎     | 219/512 [24:41<32:12,  6.60s/it]
 43%|████▎     | 220/512 [24:48<32:19,  6.64s/it]
                                                 
{'loss': '0.001774', 'grad_norm': '0.1526', 'learning_rate': '6.913e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 6650224, 'tokens/trainable': 100620, 'epoch': '0.8594'}
-
 43%|████▎     | 220/512 [24:48<32:19,  6.64s/it]
 43%|████▎     | 221/512 [24:55<32:19,  6.67s/it]
                                                 
{'loss': '0.006786', 'grad_norm': '0.8795', 'learning_rate': '6.885e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 6680384, 'tokens/trainable': 101080, 'epoch': '0.8633'}
-
 43%|████▎     | 221/512 [24:55<32:19,  6.67s/it]
 43%|████▎     | 222/512 [25:01<32:16,  6.68s/it]
                                                 
{'loss': '0.001469', 'grad_norm': '0.1953', 'learning_rate': '6.857e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 6710896, 'tokens/trainable': 101546, 'epoch': '0.8672'}
-
 43%|████▎     | 222/512 [25:01<32:16,  6.68s/it]
 44%|████▎     | 223/512 [25:08<32:08,  6.67s/it]
                                                 
{'loss': '0.0001241', 'grad_norm': '0.01113', 'learning_rate': '6.83e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 6741280, 'tokens/trainable': 101987, 'epoch': '0.8711'}
-
 44%|████▎     | 223/512 [25:08<32:08,  6.67s/it]
 44%|████▍     | 224/512 [25:15<32:05,  6.69s/it]
                                                 
{'loss': '0.004402', 'grad_norm': '0.4161', 'learning_rate': '6.802e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 6771488, 'tokens/trainable': 102435, 'epoch': '0.875'}
-
 44%|████▍     | 224/512 [25:15<32:05,  6.69s/it][2026-08-18 13:04:30,905] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
-
 44%|████▍     | 225/512 [25:23<34:52,  7.29s/it]
                                                 
{'loss': '0.0005714', 'grad_norm': '0.2329', 'learning_rate': '6.774e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '28.91', 'tokens/total': 6801744, 'tokens/trainable': 102843, 'epoch': '0.8789'}
-
 44%|████▍     | 225/512 [25:23<34:52,  7.29s/it]
 44%|████▍     | 226/512 [25:30<34:07,  7.16s/it]
                                                 
{'loss': '0.0001228', 'grad_norm': '0.01683', 'learning_rate': '6.746e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.37', 'tokens/total': 6832432, 'tokens/trainable': 103342, 'epoch': '0.8828'}
-
 44%|████▍     | 226/512 [25:30<34:07,  7.16s/it]
 44%|████▍     | 227/512 [25:37<33:20,  7.02s/it]
                                                 
{'loss': '0.004477', 'grad_norm': '0.3537', 'learning_rate': '6.718e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.71', 'tokens/train_per_sec_per_gpu': '40.68', 'tokens/total': 6862448, 'tokens/trainable': 103844, 'epoch': '0.8867'}
-
 44%|████▍     | 227/512 [25:37<33:20,  7.02s/it]
 45%|████▍     | 228/512 [25:44<32:46,  6.93s/it]
                                                 
{'loss': '0.002163', 'grad_norm': '0.4744', 'learning_rate': '6.69e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 6892832, 'tokens/trainable': 104307, 'epoch': '0.8906'}
-
 45%|████▍     | 228/512 [25:44<32:46,  6.93s/it]
 45%|████▍     | 229/512 [25:50<32:18,  6.85s/it]
                                                 
{'loss': '0.004895', 'grad_norm': '0.4888', 'learning_rate': '6.662e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.43', 'tokens/total': 6922832, 'tokens/trainable': 104772, 'epoch': '0.8945'}
-
 45%|████▍     | 229/512 [25:50<32:18,  6.85s/it]
 45%|████▍     | 230/512 [25:57<32:01,  6.82s/it]
                                                 
{'loss': '7.635e-05', 'grad_norm': '0.01735', 'learning_rate': '6.634e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.83', 'tokens/total': 6953104, 'tokens/trainable': 105240, 'epoch': '0.8984'}
-
 45%|████▍     | 230/512 [25:57<32:01,  6.82s/it]
 45%|████▌     | 231/512 [26:04<31:45,  6.78s/it]
                                                 
{'loss': '0.003428', 'grad_norm': '0.3774', 'learning_rate': '6.606e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.75', 'tokens/total': 6983440, 'tokens/trainable': 105703, 'epoch': '0.9023'}
-
 45%|████▌     | 231/512 [26:04<31:45,  6.78s/it]
 45%|████▌     | 232/512 [26:10<31:31,  6.76s/it]
                                                 
{'loss': '1.835e-05', 'grad_norm': '0.0006717', 'learning_rate': '6.578e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 7013840, 'tokens/trainable': 106179, 'epoch': '0.9062'}
-
 45%|████▌     | 232/512 [26:10<31:31,  6.76s/it]
 46%|████▌     | 233/512 [26:17<31:23,  6.75s/it]
                                                 
{'loss': '2.816e-05', 'grad_norm': '0.001895', 'learning_rate': '6.55e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '37.96', 'tokens/total': 7044240, 'tokens/trainable': 106683, 'epoch': '0.9102'}
-
 46%|████▌     | 233/512 [26:17<31:23,  6.75s/it]
 46%|████▌     | 234/512 [26:24<31:09,  6.72s/it]
                                                 
{'loss': '0.0001036', 'grad_norm': '0.0416', 'learning_rate': '6.522e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 7074224, 'tokens/trainable': 107121, 'epoch': '0.9141'}
-
 46%|████▌     | 234/512 [26:24<31:09,  6.72s/it]
 46%|████▌     | 235/512 [26:31<31:00,  6.72s/it]
                                                 
{'loss': '5.149e-05', 'grad_norm': '0.005845', 'learning_rate': '6.493e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 7104672, 'tokens/trainable': 107561, 'epoch': '0.918'}
-
 46%|████▌     | 235/512 [26:31<31:00,  6.72s/it]
 46%|████▌     | 236/512 [26:37<30:53,  6.72s/it]
                                                 
{'loss': '2.886e-05', 'grad_norm': '0.002267', 'learning_rate': '6.465e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 7134784, 'tokens/trainable': 108018, 'epoch': '0.9219'}
-
 46%|████▌     | 236/512 [26:37<30:53,  6.72s/it]
 46%|████▋     | 237/512 [26:44<30:43,  6.70s/it]
                                                 
{'loss': '0.0009025', 'grad_norm': '0.2714', 'learning_rate': '6.437e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.28', 'tokens/total': 7164960, 'tokens/trainable': 108446, 'epoch': '0.9258'}
-
 46%|████▋     | 237/512 [26:44<30:43,  6.70s/it]
 46%|████▋     | 238/512 [26:51<30:32,  6.69s/it]
                                                 
{'loss': '0.0004671', 'grad_norm': '0.1335', 'learning_rate': '6.408e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.14', 'tokens/total': 7195344, 'tokens/trainable': 108866, 'epoch': '0.9297'}
-
 46%|████▋     | 238/512 [26:51<30:32,  6.69s/it]
 47%|████▋     | 239/512 [26:57<30:31,  6.71s/it]
                                                 
{'loss': '0.0006794', 'grad_norm': '0.1007', 'learning_rate': '6.38e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.8', 'tokens/total': 7225680, 'tokens/trainable': 109316, 'epoch': '0.9336'}
-
 47%|████▋     | 239/512 [26:57<30:31,  6.71s/it]
 47%|████▋     | 240/512 [27:04<30:26,  6.72s/it]
                                                 
{'loss': '0.0005133', 'grad_norm': '0.1203', 'learning_rate': '6.351e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.26', 'tokens/total': 7256112, 'tokens/trainable': 109787, 'epoch': '0.9375'}
-
 47%|████▋     | 240/512 [27:04<30:26,  6.72s/it]
 47%|████▋     | 241/512 [27:11<30:15,  6.70s/it]
                                                 
{'loss': '3.099e-05', 'grad_norm': '0.006275', 'learning_rate': '6.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.3', 'tokens/total': 7286384, 'tokens/trainable': 110246, 'epoch': '0.9414'}
-
 47%|████▋     | 241/512 [27:11<30:15,  6.70s/it]
 47%|████▋     | 242/512 [27:17<30:11,  6.71s/it]
                                                 
{'loss': '0.000395', 'grad_norm': '0.05609', 'learning_rate': '6.294e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.14', 'tokens/total': 7316688, 'tokens/trainable': 110667, 'epoch': '0.9453'}
-
 47%|████▋     | 242/512 [27:17<30:11,  6.71s/it]
 47%|████▋     | 243/512 [27:24<30:06,  6.71s/it]
                                                 
{'loss': '0.0002062', 'grad_norm': '0.05409', 'learning_rate': '6.266e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.09', 'tokens/total': 7346992, 'tokens/trainable': 111154, 'epoch': '0.9492'}
-
 47%|████▋     | 243/512 [27:24<30:06,  6.71s/it]
 48%|████▊     | 244/512 [27:31<30:03,  6.73s/it]
                                                 
{'loss': '0.0005747', 'grad_norm': '0.05567', 'learning_rate': '6.237e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 7377632, 'tokens/trainable': 111611, 'epoch': '0.9531'}
-
 48%|████▊     | 244/512 [27:31<30:03,  6.73s/it]
 48%|████▊     | 245/512 [27:38<29:52,  6.72s/it]
                                                 
{'loss': '0.0008788', 'grad_norm': '0.1613', 'learning_rate': '6.208e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 7407872, 'tokens/trainable': 112102, 'epoch': '0.957'}
-
 48%|████▊     | 245/512 [27:38<29:52,  6.72s/it]
 48%|████▊     | 246/512 [27:44<29:44,  6.71s/it]
                                                 
{'loss': '4.069e-05', 'grad_norm': '0.005659', 'learning_rate': '6.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.16', 'tokens/total': 7438448, 'tokens/trainable': 112569, 'epoch': '0.9609'}
-
 48%|████▊     | 246/512 [27:44<29:44,  6.71s/it]
 48%|████▊     | 247/512 [27:51<29:00,  6.57s/it]
                                                 
{'loss': '0.0005083', 'grad_norm': '0.1775', 'learning_rate': '6.151e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '37.68', 'tokens/total': 7466768, 'tokens/trainable': 113023, 'epoch': '0.9648'}
-
 48%|████▊     | 247/512 [27:51<29:00,  6.57s/it]
 48%|████▊     | 248/512 [27:57<29:09,  6.63s/it]
                                                 
{'loss': '3.471e-05', 'grad_norm': '0.002822', 'learning_rate': '6.122e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.17', 'tokens/total': 7497360, 'tokens/trainable': 113464, 'epoch': '0.9688'}
-
 48%|████▊     | 248/512 [27:57<29:09,  6.63s/it]
 49%|████▊     | 249/512 [28:04<29:07,  6.64s/it]
                                                 
{'loss': '8.433e-05', 'grad_norm': '0.03676', 'learning_rate': '6.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 7527504, 'tokens/trainable': 113908, 'epoch': '0.9727'}
-
 49%|████▊     | 249/512 [28:04<29:07,  6.64s/it]
 49%|████▉     | 250/512 [28:10<28:26,  6.51s/it]
                                                 
{'loss': '0.0006511', 'grad_norm': '0.1124', 'learning_rate': '6.065e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 7555808, 'tokens/trainable': 114348, 'epoch': '0.9766'}
-
 49%|████▉     | 250/512 [28:10<28:26,  6.51s/it]
 49%|████▉     | 251/512 [28:17<28:36,  6.58s/it]
                                                 
{'loss': '0.0002415', 'grad_norm': '0.04923', 'learning_rate': '6.036e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.15', 'tokens/total': 7586320, 'tokens/trainable': 114830, 'epoch': '0.9805'}
-
 49%|████▉     | 251/512 [28:17<28:36,  6.58s/it]
 49%|████▉     | 252/512 [28:24<28:41,  6.62s/it]
                                                 
{'loss': '0.005054', 'grad_norm': '0.4834', 'learning_rate': '6.007e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 7616576, 'tokens/trainable': 115278, 'epoch': '0.9844'}
-
 49%|████▉     | 252/512 [28:24<28:41,  6.62s/it]
 49%|████▉     | 253/512 [28:30<28:46,  6.67s/it]
                                                 
{'loss': '8.867e-05', 'grad_norm': '0.01336', 'learning_rate': '5.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 7647152, 'tokens/trainable': 115699, 'epoch': '0.9883'}
-
 49%|████▉     | 253/512 [28:30<28:46,  6.67s/it]
 50%|████▉     | 254/512 [28:37<28:42,  6.68s/it]
                                                 
{'loss': '0.0006354', 'grad_norm': '0.09679', 'learning_rate': '5.949e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '30.73', 'tokens/total': 7677584, 'tokens/trainable': 116133, 'epoch': '0.9922'}
-
 50%|████▉     | 254/512 [28:37<28:42,  6.68s/it]
 50%|████▉     | 255/512 [28:44<28:37,  6.68s/it]
                                                 
{'loss': '0.001129', 'grad_norm': '0.514', 'learning_rate': '5.92e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 7708016, 'tokens/trainable': 116591, 'epoch': '0.9961'}
-
 50%|████▉     | 255/512 [28:44<28:37,  6.68s/it]
 50%|█████     | 256/512 [28:51<28:46,  6.74s/it]
                                                 
{'loss': '7.111e-06', 'grad_norm': '0.000276', 'learning_rate': '5.891e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.04', 'tokens/total': 7738608, 'tokens/trainable': 117039, 'epoch': '1'}
-
 50%|█████     | 256/512 [28:51<28:46,  6.74s/it][2026-08-18 13:08:06,959] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
-
 50%|█████     | 257/512 [29:00<31:40,  7.45s/it]
                                                 
{'loss': '2.959e-05', 'grad_norm': '0.00545', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.24', 'tokens/total': 7768944, 'tokens/trainable': 117495, 'epoch': '1.004'}
-
 50%|█████     | 257/512 [29:00<31:40,  7.45s/it]
 50%|█████     | 258/512 [29:07<30:36,  7.23s/it]
                                                 
{'loss': '2.851e-05', 'grad_norm': '0.003981', 'learning_rate': '5.834e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.44', 'tokens/total': 7799232, 'tokens/trainable': 117925, 'epoch': '1.008'}
-
 50%|█████     | 258/512 [29:07<30:36,  7.23s/it]
 51%|█████     | 259/512 [29:13<29:50,  7.08s/it]
                                                 
{'loss': '1.161e-05', 'grad_norm': '0.001549', 'learning_rate': '5.805e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.58', 'tokens/total': 7829568, 'tokens/trainable': 118373, 'epoch': '1.012'}
-
 51%|█████     | 259/512 [29:13<29:50,  7.08s/it]
 51%|█████     | 260/512 [29:20<29:15,  6.97s/it]
                                                 
{'loss': '9.4e-06', 'grad_norm': '0.0003936', 'learning_rate': '5.776e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 7860000, 'tokens/trainable': 118816, 'epoch': '1.016'}
-
 51%|█████     | 260/512 [29:20<29:15,  6.97s/it]
 51%|█████     | 261/512 [29:27<28:47,  6.88s/it]
                                                 
{'loss': '0.000134', 'grad_norm': '0.02447', 'learning_rate': '5.747e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.8', 'tokens/total': 7890464, 'tokens/trainable': 119263, 'epoch': '1.02'}
-
 51%|█████     | 261/512 [29:27<28:47,  6.88s/it]
 51%|█████     | 262/512 [29:33<28:33,  6.85s/it]
                                                 
{'loss': '0.002393', 'grad_norm': '0.3619', 'learning_rate': '5.718e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.66', 'tokens/total': 7920944, 'tokens/trainable': 119772, 'epoch': '1.023'}
-
 51%|█████     | 262/512 [29:33<28:33,  6.85s/it]
 51%|█████▏    | 263/512 [29:40<28:18,  6.82s/it]
                                                 
{'loss': '0.000525', 'grad_norm': '0.1766', 'learning_rate': '5.689e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 7951296, 'tokens/trainable': 120269, 'epoch': '1.027'}
-
 51%|█████▏    | 263/512 [29:40<28:18,  6.82s/it]
 52%|█████▏    | 264/512 [29:47<28:07,  6.81s/it]
                                                 
{'loss': '1.115e-05', 'grad_norm': '0.0007879', 'learning_rate': '5.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.96', 'tokens/total': 7981856, 'tokens/trainable': 120746, 'epoch': '1.031'}
-
 52%|█████▏    | 264/512 [29:47<28:07,  6.81s/it]
 52%|█████▏    | 265/512 [29:54<27:51,  6.77s/it]
                                                 
{'loss': '4.71e-05', 'grad_norm': '0.008799', 'learning_rate': '5.631e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.22', 'tokens/total': 8012192, 'tokens/trainable': 121204, 'epoch': '1.035'}
-
 52%|█████▏    | 265/512 [29:54<27:51,  6.77s/it]
 52%|█████▏    | 266/512 [30:00<27:43,  6.76s/it]
                                                 
{'loss': '8.791e-06', 'grad_norm': '0.0005215', 'learning_rate': '5.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.23', 'tokens/total': 8042736, 'tokens/trainable': 121670, 'epoch': '1.039'}
-
 52%|█████▏    | 266/512 [30:00<27:43,  6.76s/it]
 52%|█████▏    | 267/512 [30:07<27:35,  6.76s/it]
                                                 
{'loss': '0.0008057', 'grad_norm': '0.1482', 'learning_rate': '5.573e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '40.04', 'tokens/total': 8073024, 'tokens/trainable': 122183, 'epoch': '1.043'}
-
 52%|█████▏    | 267/512 [30:07<27:35,  6.76s/it]
 52%|█████▏    | 268/512 [30:14<27:24,  6.74s/it]
                                                 
{'loss': '0.000323', 'grad_norm': '0.08615', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 8103280, 'tokens/trainable': 122645, 'epoch': '1.047'}
-
 52%|█████▏    | 268/512 [30:14<27:24,  6.74s/it]
 53%|█████▎    | 269/512 [30:21<27:15,  6.73s/it]
                                                 
{'loss': '1.134e-05', 'grad_norm': '0.0005172', 'learning_rate': '5.515e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 8133440, 'tokens/trainable': 123105, 'epoch': '1.051'}
-
 53%|█████▎    | 269/512 [30:21<27:15,  6.73s/it]
 53%|█████▎    | 270/512 [30:27<27:09,  6.73s/it]
                                                 
{'loss': '1.195e-05', 'grad_norm': '0.001041', 'learning_rate': '5.485e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 8163920, 'tokens/trainable': 123595, 'epoch': '1.055'}
-
 53%|█████▎    | 270/512 [30:27<27:09,  6.73s/it]
 53%|█████▎    | 271/512 [30:34<27:00,  6.73s/it]
                                                 
{'loss': '1.726e-05', 'grad_norm': '0.005696', 'learning_rate': '5.456e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.59', 'tokens/total': 8194032, 'tokens/trainable': 124077, 'epoch': '1.059'}
-
 53%|█████▎    | 271/512 [30:34<27:00,  6.73s/it]
 53%|█████▎    | 272/512 [30:41<26:56,  6.73s/it]
                                                 
{'loss': '1.061e-05', 'grad_norm': '0.0005264', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.65', 'tokens/total': 8224320, 'tokens/trainable': 124547, 'epoch': '1.062'}
-
 53%|█████▎    | 272/512 [30:41<26:56,  6.73s/it]
 53%|█████▎    | 273/512 [30:47<26:46,  6.72s/it]
                                                 
{'loss': '0.0003628', 'grad_norm': '0.08169', 'learning_rate': '5.398e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 8254368, 'tokens/trainable': 124977, 'epoch': '1.066'}
-
 53%|█████▎    | 273/512 [30:47<26:46,  6.72s/it]
 54%|█████▎    | 274/512 [30:54<26:37,  6.71s/it]
                                                 
{'loss': '1.123e-05', 'grad_norm': '0.0005203', 'learning_rate': '5.369e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.51', 'tokens/total': 8284944, 'tokens/trainable': 125438, 'epoch': '1.07'}
-
 54%|█████▎    | 274/512 [30:54<26:37,  6.71s/it]
 54%|█████▎    | 275/512 [31:01<26:32,  6.72s/it]
                                                 
{'loss': '1.476e-05', 'grad_norm': '0.00311', 'learning_rate': '5.34e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.47', 'tokens/total': 8315584, 'tokens/trainable': 125889, 'epoch': '1.074'}
-
 54%|█████▎    | 275/512 [31:01<26:32,  6.72s/it]
 54%|█████▍    | 276/512 [31:08<26:26,  6.72s/it]
                                                 
{'loss': '9.164e-05', 'grad_norm': '0.01906', 'learning_rate': '5.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 8345952, 'tokens/trainable': 126363, 'epoch': '1.078'}
-
 54%|█████▍    | 276/512 [31:08<26:26,  6.72s/it]
 54%|█████▍    | 277/512 [31:14<26:23,  6.74s/it]
                                                 
{'loss': '1.797e-05', 'grad_norm': '0.002002', 'learning_rate': '5.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 8376416, 'tokens/trainable': 126828, 'epoch': '1.082'}
-
 54%|█████▍    | 277/512 [31:14<26:23,  6.74s/it]
 54%|█████▍    | 278/512 [31:21<26:14,  6.73s/it]
                                                 
{'loss': '1.279e-05', 'grad_norm': '0.001124', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.16', 'tokens/total': 8406640, 'tokens/trainable': 127331, 'epoch': '1.086'}
-
 54%|█████▍    | 278/512 [31:21<26:14,  6.73s/it]
 54%|█████▍    | 279/512 [31:28<26:05,  6.72s/it]
                                                 
{'loss': '0.0005633', 'grad_norm': '0.07841', 'learning_rate': '5.224e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 8436784, 'tokens/trainable': 127798, 'epoch': '1.09'}
-
 54%|█████▍    | 279/512 [31:28<26:05,  6.72s/it]
 55%|█████▍    | 280/512 [31:35<25:59,  6.72s/it]
                                                 
{'loss': '4.469e-05', 'grad_norm': '0.006657', 'learning_rate': '5.195e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 8467152, 'tokens/trainable': 128247, 'epoch': '1.094'}
-
 55%|█████▍    | 280/512 [31:35<25:59,  6.72s/it]
 55%|█████▍    | 281/512 [31:41<25:54,  6.73s/it]
                                                 
{'loss': '9.423e-06', 'grad_norm': '0.0006085', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 8497424, 'tokens/trainable': 128729, 'epoch': '1.098'}
-
 55%|█████▍    | 281/512 [31:41<25:54,  6.73s/it]
 55%|█████▌    | 282/512 [31:48<25:48,  6.73s/it]
                                                 
{'loss': '0.002226', 'grad_norm': '0.4024', 'learning_rate': '5.138e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 8527616, 'tokens/trainable': 129203, 'epoch': '1.102'}
-
 55%|█████▌    | 282/512 [31:48<25:48,  6.73s/it]
 55%|█████▌    | 283/512 [31:55<25:42,  6.74s/it]
                                                 
{'loss': '9.101e-06', 'grad_norm': '0.0008598', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 8557984, 'tokens/trainable': 129653, 'epoch': '1.105'}
-
 55%|█████▌    | 283/512 [31:55<25:42,  6.74s/it]
 55%|█████▌    | 284/512 [32:02<25:37,  6.74s/it]
                                                 
{'loss': '0.0004534', 'grad_norm': '0.1385', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.56', 'tokens/total': 8588384, 'tokens/trainable': 130100, 'epoch': '1.109'}
-
 55%|█████▌    | 284/512 [32:02<25:37,  6.74s/it]
 56%|█████▌    | 285/512 [32:08<25:28,  6.73s/it]
                                                 
{'loss': '0.0002941', 'grad_norm': '0.08453', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.44', 'tokens/total': 8618768, 'tokens/trainable': 130543, 'epoch': '1.113'}
-
 56%|█████▌    | 285/512 [32:08<25:28,  6.73s/it]
 56%|█████▌    | 286/512 [32:15<25:21,  6.73s/it]
                                                 
{'loss': '0.0001019', 'grad_norm': '0.01867', 'learning_rate': '5.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 8648976, 'tokens/trainable': 131000, 'epoch': '1.117'}
-
 56%|█████▌    | 286/512 [32:15<25:21,  6.73s/it]
 56%|█████▌    | 287/512 [32:22<25:13,  6.73s/it]
                                                 
{'loss': '1.198e-05', 'grad_norm': '0.00159', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 8679152, 'tokens/trainable': 131449, 'epoch': '1.121'}
-
 56%|█████▌    | 287/512 [32:22<25:13,  6.73s/it]
 56%|█████▋    | 288/512 [32:28<25:03,  6.71s/it]
                                                 
{'loss': '7.07e-06', 'grad_norm': '0.0004093', 'learning_rate': '4.964e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '35.94', 'tokens/total': 8709504, 'tokens/trainable': 131923, 'epoch': '1.125'}
-
 56%|█████▋    | 288/512 [32:28<25:03,  6.71s/it][2026-08-18 13:11:44,549] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
-
 56%|█████▋    | 289/512 [32:37<27:25,  7.38s/it]
                                                 
{'loss': '0.0006339', 'grad_norm': '0.1226', 'learning_rate': '4.935e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '33.11', 'tokens/total': 8739680, 'tokens/trainable': 132397, 'epoch': '1.129'}
-
 56%|█████▋    | 289/512 [32:37<27:25,  7.38s/it]
 57%|█████▋    | 290/512 [32:44<26:34,  7.18s/it]
                                                 
{'loss': '5.397e-06', 'grad_norm': '0.0002119', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 8770112, 'tokens/trainable': 132846, 'epoch': '1.133'}
-
 57%|█████▋    | 290/512 [32:44<26:34,  7.18s/it]
 57%|█████▋    | 291/512 [32:51<25:57,  7.05s/it]
                                                 
{'loss': '0.00135', 'grad_norm': '0.2777', 'learning_rate': '4.878e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '35.65', 'tokens/total': 8800368, 'tokens/trainable': 133347, 'epoch': '1.137'}
-
 57%|█████▋    | 291/512 [32:51<25:57,  7.05s/it]
 57%|█████▋    | 292/512 [32:57<25:27,  6.94s/it]
                                                 
{'loss': '8.635e-06', 'grad_norm': '0.001235', 'learning_rate': '4.849e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 8830688, 'tokens/trainable': 133779, 'epoch': '1.141'}
-
 57%|█████▋    | 292/512 [32:57<25:27,  6.94s/it]
 57%|█████▋    | 293/512 [33:04<25:08,  6.89s/it]
                                                 
{'loss': '4.307e-06', 'grad_norm': '0.0002618', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 8861008, 'tokens/trainable': 134251, 'epoch': '1.145'}
-
 57%|█████▋    | 293/512 [33:04<25:08,  6.89s/it]
 57%|█████▋    | 294/512 [33:11<24:56,  6.86s/it]
                                                 
{'loss': '4.467e-06', 'grad_norm': '0.0004677', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 8891760, 'tokens/trainable': 134683, 'epoch': '1.148'}
-
 57%|█████▋    | 294/512 [33:11<24:56,  6.86s/it]
 58%|█████▊    | 295/512 [33:18<24:40,  6.82s/it]
                                                 
{'loss': '4.914e-06', 'grad_norm': '0.0002164', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 8922192, 'tokens/trainable': 135165, 'epoch': '1.152'}
-
 58%|█████▊    | 295/512 [33:18<24:40,  6.82s/it]
 58%|█████▊    | 296/512 [33:24<24:23,  6.78s/it]
                                                 
{'loss': '4.242e-06', 'grad_norm': '0.0002447', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 8952496, 'tokens/trainable': 135599, 'epoch': '1.156'}
-
 58%|█████▊    | 296/512 [33:24<24:23,  6.78s/it]
 58%|█████▊    | 297/512 [33:31<24:11,  6.75s/it]
                                                 
{'loss': '4.674e-06', 'grad_norm': '0.0002878', 'learning_rate': '4.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.56', 'tokens/total': 8982768, 'tokens/trainable': 136038, 'epoch': '1.16'}
-
 58%|█████▊    | 297/512 [33:31<24:11,  6.75s/it]
 58%|█████▊    | 298/512 [33:38<24:03,  6.74s/it]
                                                 
{'loss': '8.829e-06', 'grad_norm': '0.001556', 'learning_rate': '4.677e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 9013248, 'tokens/trainable': 136495, 'epoch': '1.164'}
-
 58%|█████▊    | 298/512 [33:38<24:03,  6.74s/it]
 58%|█████▊    | 299/512 [33:45<23:56,  6.75s/it]
                                                 
{'loss': '5.587e-06', 'grad_norm': '0.0002536', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 9043632, 'tokens/trainable': 136953, 'epoch': '1.168'}
-
 58%|█████▊    | 299/512 [33:45<23:56,  6.75s/it]
 59%|█████▊    | 300/512 [33:51<23:45,  6.73s/it]
                                                 
{'loss': '5.917e-06', 'grad_norm': '0.000319', 'learning_rate': '4.62e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 9073936, 'tokens/trainable': 137412, 'epoch': '1.172'}
-
 59%|█████▊    | 300/512 [33:51<23:45,  6.73s/it]
 59%|█████▉    | 301/512 [33:58<23:38,  6.72s/it]
                                                 
{'loss': '9.368e-06', 'grad_norm': '0.0005231', 'learning_rate': '4.592e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.62', 'tokens/total': 9104416, 'tokens/trainable': 137907, 'epoch': '1.176'}
-
 59%|█████▉    | 301/512 [33:58<23:38,  6.72s/it]
 59%|█████▉    | 302/512 [34:05<23:33,  6.73s/it]
                                                 
{'loss': '9.613e-06', 'grad_norm': '0.001005', 'learning_rate': '4.563e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 9134688, 'tokens/trainable': 138383, 'epoch': '1.18'}
-
 59%|█████▉    | 302/512 [34:05<23:33,  6.73s/it]
 59%|█████▉    | 303/512 [34:11<23:29,  6.74s/it]
                                                 
{'loss': '2.478e-05', 'grad_norm': '0.005482', 'learning_rate': '4.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '36.35', 'tokens/total': 9165088, 'tokens/trainable': 138842, 'epoch': '1.184'}
-
 59%|█████▉    | 303/512 [34:11<23:29,  6.74s/it]
 59%|█████▉    | 304/512 [34:18<23:17,  6.72s/it]
                                                 
{'loss': '5.224e-06', 'grad_norm': '0.0005683', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '30.61', 'tokens/total': 9195584, 'tokens/trainable': 139269, 'epoch': '1.188'}
-
 59%|█████▉    | 304/512 [34:18<23:17,  6.72s/it]
 60%|█████▉    | 305/512 [34:25<23:11,  6.72s/it]
                                                 
{'loss': '2.194e-05', 'grad_norm': '0.003497', 'learning_rate': '4.478e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.36', 'tokens/total': 9225760, 'tokens/trainable': 139736, 'epoch': '1.191'}
-
 60%|█████▉    | 305/512 [34:25<23:11,  6.72s/it]
 60%|█████▉    | 306/512 [34:32<23:04,  6.72s/it]
                                                 
{'loss': '4.178e-06', 'grad_norm': '0.000202', 'learning_rate': '4.45e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.91', 'tokens/total': 9256352, 'tokens/trainable': 140226, 'epoch': '1.195'}
-
 60%|█████▉    | 306/512 [34:32<23:04,  6.72s/it]
 60%|█████▉    | 307/512 [34:38<22:20,  6.54s/it]
                                                 
{'loss': '5.606e-06', 'grad_norm': '0.0004581', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.19', 'memory/max_allocated (GiB)': '33.19', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 9284256, 'tokens/trainable': 140664, 'epoch': '1.199'}
-
 60%|█████▉    | 307/512 [34:38<22:20,  6.54s/it]
 60%|██████    | 308/512 [34:44<22:29,  6.62s/it]
                                                 
{'loss': '7.164e-06', 'grad_norm': '0.001661', 'learning_rate': '4.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '29.85', 'tokens/total': 9314768, 'tokens/trainable': 141115, 'epoch': '1.203'}
-
 60%|██████    | 308/512 [34:45<22:29,  6.62s/it]
 60%|██████    | 309/512 [34:51<22:29,  6.65s/it]
                                                 
{'loss': '3.827e-06', 'grad_norm': '0.0001901', 'learning_rate': '4.366e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '28.88', 'tokens/total': 9344976, 'tokens/trainable': 141530, 'epoch': '1.207'}
-
 60%|██████    | 309/512 [34:51<22:29,  6.65s/it]
 61%|██████    | 310/512 [34:58<22:24,  6.66s/it]
                                                 
{'loss': '2.814e-05', 'grad_norm': '0.003887', 'learning_rate': '4.338e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 9375280, 'tokens/trainable': 141964, 'epoch': '1.211'}
-
 61%|██████    | 310/512 [34:58<22:24,  6.66s/it]
 61%|██████    | 311/512 [35:05<22:19,  6.67s/it]
                                                 
{'loss': '0.0001199', 'grad_norm': '0.02483', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 9405536, 'tokens/trainable': 142428, 'epoch': '1.215'}
-
 61%|██████    | 311/512 [35:05<22:19,  6.67s/it]
 61%|██████    | 312/512 [35:11<22:12,  6.66s/it]
                                                 
{'loss': '9.509e-06', 'grad_norm': '0.001796', 'learning_rate': '4.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '38.07', 'tokens/total': 9435696, 'tokens/trainable': 142894, 'epoch': '1.219'}
-
 61%|██████    | 312/512 [35:11<22:12,  6.66s/it]
 61%|██████    | 313/512 [35:18<22:09,  6.68s/it]
                                                 
{'loss': '8.858e-06', 'grad_norm': '0.0007182', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 9466080, 'tokens/trainable': 143368, 'epoch': '1.223'}
-
 61%|██████    | 313/512 [35:18<22:09,  6.68s/it]
 61%|██████▏   | 314/512 [35:25<22:04,  6.69s/it]
                                                 
{'loss': '0.0151', 'grad_norm': '0.7545', 'learning_rate': '4.226e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 9496400, 'tokens/trainable': 143824, 'epoch': '1.227'}
-
 61%|██████▏   | 314/512 [35:25<22:04,  6.69s/it]
 62%|██████▏   | 315/512 [35:31<22:05,  6.73s/it]
                                                 
{'loss': '0.0007513', 'grad_norm': '0.281', 'learning_rate': '4.198e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 9526880, 'tokens/trainable': 144306, 'epoch': '1.23'}
-
 62%|██████▏   | 315/512 [35:31<22:05,  6.73s/it]
 62%|██████▏   | 316/512 [35:38<21:58,  6.73s/it]
                                                 
{'loss': '0.0005616', 'grad_norm': '0.7471', 'learning_rate': '4.17e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 9557280, 'tokens/trainable': 144763, 'epoch': '1.234'}
-
 62%|██████▏   | 316/512 [35:38<21:58,  6.73s/it]
 62%|██████▏   | 317/512 [35:45<21:52,  6.73s/it]
                                                 
{'loss': '5.734e-06', 'grad_norm': '0.0004947', 'learning_rate': '4.143e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.4', 'tokens/total': 9587712, 'tokens/trainable': 145255, 'epoch': '1.238'}
-
 62%|██████▏   | 317/512 [35:45<21:52,  6.73s/it]
 62%|██████▏   | 318/512 [35:52<21:45,  6.73s/it]
                                                 
{'loss': '2.625e-05', 'grad_norm': '0.003202', 'learning_rate': '4.115e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 9618064, 'tokens/trainable': 145714, 'epoch': '1.242'}
-
 62%|██████▏   | 318/512 [35:52<21:45,  6.73s/it]
 62%|██████▏   | 319/512 [35:58<21:39,  6.73s/it]
                                                 
{'loss': '0.0263', 'grad_norm': '1.172', 'learning_rate': '4.087e-05', 'ppl': '1.027', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.96', 'tokens/total': 9648336, 'tokens/trainable': 146200, 'epoch': '1.246'}
-
 62%|██████▏   | 319/512 [35:58<21:39,  6.73s/it]
 62%|██████▎   | 320/512 [36:05<21:31,  6.73s/it]
                                                 
{'loss': '8.492e-06', 'grad_norm': '0.00046', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 9678688, 'tokens/trainable': 146668, 'epoch': '1.25'}
-
 62%|██████▎   | 320/512 [36:05<21:31,  6.73s/it][2026-08-18 13:15:21,351] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
-
 63%|██████▎   | 321/512 [36:14<23:15,  7.31s/it]
                                                 
{'loss': '6.043e-05', 'grad_norm': '0.006735', 'learning_rate': '4.032e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 9708848, 'tokens/trainable': 147114, 'epoch': '1.254'}
-
 63%|██████▎   | 321/512 [36:14<23:15,  7.31s/it]
 63%|██████▎   | 322/512 [36:20<22:33,  7.12s/it]
                                                 
{'loss': '0.0002499', 'grad_norm': '0.05488', 'learning_rate': '4.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 9739360, 'tokens/trainable': 147564, 'epoch': '1.258'}
-
 63%|██████▎   | 322/512 [36:20<22:33,  7.12s/it]
 63%|██████▎   | 323/512 [36:27<22:03,  7.00s/it]
                                                 
{'loss': '1.1e-05', 'grad_norm': '0.001182', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 9769776, 'tokens/trainable': 148013, 'epoch': '1.262'}
-
 63%|██████▎   | 323/512 [36:27<22:03,  7.00s/it]
 63%|██████▎   | 324/512 [36:34<21:44,  6.94s/it]
                                                 
{'loss': '0.0002116', 'grad_norm': '0.01802', 'learning_rate': '3.95e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.99', 'tokens/total': 9800096, 'tokens/trainable': 148470, 'epoch': '1.266'}
-
 63%|██████▎   | 324/512 [36:34<21:44,  6.94s/it]
 63%|██████▎   | 325/512 [36:41<21:22,  6.86s/it]
                                                 
{'loss': '0.01622', 'grad_norm': '1.591', 'learning_rate': '3.923e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 9830304, 'tokens/trainable': 148926, 'epoch': '1.27'}
-
 63%|██████▎   | 325/512 [36:41<21:22,  6.86s/it]
 64%|██████▎   | 326/512 [36:47<21:10,  6.83s/it]
                                                 
{'loss': '3.801e-05', 'grad_norm': '0.004154', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.72', 'tokens/total': 9860608, 'tokens/trainable': 149420, 'epoch': '1.273'}
-
 64%|██████▎   | 326/512 [36:47<21:10,  6.83s/it]
 64%|██████▍   | 327/512 [36:54<20:57,  6.80s/it]
                                                 
{'loss': '5.527e-05', 'grad_norm': '0.002978', 'learning_rate': '3.869e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '39.32', 'tokens/total': 9890560, 'tokens/trainable': 149873, 'epoch': '1.277'}
-
 64%|██████▍   | 327/512 [36:54<20:57,  6.80s/it]
 64%|██████▍   | 328/512 [37:01<20:47,  6.78s/it]
                                                 
{'loss': '0.0003906', 'grad_norm': '0.02816', 'learning_rate': '3.842e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 9920880, 'tokens/trainable': 150347, 'epoch': '1.281'}
-
 64%|██████▍   | 328/512 [37:01<20:47,  6.78s/it]
 64%|██████▍   | 329/512 [37:08<20:37,  6.76s/it]
                                                 
{'loss': '0.0005861', 'grad_norm': '0.04141', 'learning_rate': '3.815e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.72', 'tokens/total': 9951280, 'tokens/trainable': 150822, 'epoch': '1.285'}
-
 64%|██████▍   | 329/512 [37:08<20:37,  6.76s/it]
 64%|██████▍   | 330/512 [37:14<20:29,  6.76s/it]
                                                 
{'loss': '9.081e-05', 'grad_norm': '0.004759', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 9981600, 'tokens/trainable': 151266, 'epoch': '1.289'}
-
 64%|██████▍   | 330/512 [37:14<20:29,  6.76s/it]
 65%|██████▍   | 331/512 [37:21<20:20,  6.74s/it]
                                                 
{'loss': '0.0005758', 'grad_norm': '0.04044', 'learning_rate': '3.761e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.11', 'tokens/total': 10012016, 'tokens/trainable': 151734, 'epoch': '1.293'}
-
 65%|██████▍   | 331/512 [37:21<20:20,  6.74s/it]
 65%|██████▍   | 332/512 [37:28<20:10,  6.72s/it]
                                                 
{'loss': '0.0004353', 'grad_norm': '0.06457', 'learning_rate': '3.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.99', 'tokens/total': 10042448, 'tokens/trainable': 152184, 'epoch': '1.297'}
-
 65%|██████▍   | 332/512 [37:28<20:10,  6.72s/it]
 65%|██████▌   | 333/512 [37:34<20:04,  6.73s/it]
                                                 
{'loss': '0.0005283', 'grad_norm': '0.05121', 'learning_rate': '3.708e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 10073024, 'tokens/trainable': 152623, 'epoch': '1.301'}
-
 65%|██████▌   | 333/512 [37:34<20:04,  6.73s/it]
 65%|██████▌   | 334/512 [37:41<19:53,  6.71s/it]
                                                 
{'loss': '0.0001399', 'grad_norm': '0.006798', 'learning_rate': '3.681e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.34', 'tokens/total': 10103504, 'tokens/trainable': 153040, 'epoch': '1.305'}
-
 65%|██████▌   | 334/512 [37:41<19:53,  6.71s/it]
 65%|██████▌   | 335/512 [37:47<19:21,  6.56s/it]
                                                 
{'loss': '0.003549', 'grad_norm': '0.2877', 'learning_rate': '3.655e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.7', 'tokens/total': 10131632, 'tokens/trainable': 153463, 'epoch': '1.309'}
-
 65%|██████▌   | 335/512 [37:47<19:21,  6.56s/it]
 66%|██████▌   | 336/512 [37:54<19:21,  6.60s/it]
                                                 
{'loss': '0.0007546', 'grad_norm': '0.04293', 'learning_rate': '3.628e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 10161824, 'tokens/trainable': 153908, 'epoch': '1.312'}
-
 66%|██████▌   | 336/512 [37:54<19:21,  6.60s/it]
 66%|██████▌   | 337/512 [38:01<19:25,  6.66s/it]
                                                 
{'loss': '0.001123', 'grad_norm': '0.1319', 'learning_rate': '3.602e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.63', 'tokens/total': 10192448, 'tokens/trainable': 154387, 'epoch': '1.316'}
-
 66%|██████▌   | 337/512 [38:01<19:25,  6.66s/it]
 66%|██████▌   | 338/512 [38:08<19:23,  6.69s/it]
                                                 
{'loss': '0.0004925', 'grad_norm': '0.06429', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 10222704, 'tokens/trainable': 154836, 'epoch': '1.32'}
-
 66%|██████▌   | 338/512 [38:08<19:23,  6.69s/it]
 66%|██████▌   | 339/512 [38:14<19:22,  6.72s/it]
                                                 
{'loss': '8.633e-05', 'grad_norm': '0.006817', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.73', 'tokens/total': 10253168, 'tokens/trainable': 155261, 'epoch': '1.324'}
-
 66%|██████▌   | 339/512 [38:14<19:22,  6.72s/it]
 66%|██████▋   | 340/512 [38:21<19:12,  6.70s/it]
                                                 
{'loss': '0.0003515', 'grad_norm': '0.04607', 'learning_rate': '3.523e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.14', 'tokens/total': 10283552, 'tokens/trainable': 155674, 'epoch': '1.328'}
-
 66%|██████▋   | 340/512 [38:21<19:12,  6.70s/it]
 67%|██████▋   | 341/512 [38:28<19:08,  6.72s/it]
                                                 
{'loss': '0.0007426', 'grad_norm': '0.2491', 'learning_rate': '3.497e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.67', 'tokens/total': 10313872, 'tokens/trainable': 156164, 'epoch': '1.332'}
-
 67%|██████▋   | 341/512 [38:28<19:08,  6.72s/it]
 67%|██████▋   | 342/512 [38:35<19:02,  6.72s/it]
                                                 
{'loss': '0.0001773', 'grad_norm': '0.01124', 'learning_rate': '3.471e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.47', 'tokens/total': 10344256, 'tokens/trainable': 156611, 'epoch': '1.336'}
-
 67%|██████▋   | 342/512 [38:35<19:02,  6.72s/it]
 67%|██████▋   | 343/512 [38:41<18:54,  6.71s/it]
                                                 
{'loss': '0.0001897', 'grad_norm': '0.009496', 'learning_rate': '3.445e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.58', 'tokens/total': 10374544, 'tokens/trainable': 157073, 'epoch': '1.34'}
-
 67%|██████▋   | 343/512 [38:41<18:54,  6.71s/it]
 67%|██████▋   | 344/512 [38:48<18:49,  6.72s/it]
                                                 
{'loss': '0.0007583', 'grad_norm': '0.1076', 'learning_rate': '3.42e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 10405040, 'tokens/trainable': 157523, 'epoch': '1.344'}
-
 67%|██████▋   | 344/512 [38:48<18:49,  6.72s/it]
 67%|██████▋   | 345/512 [38:55<18:41,  6.72s/it]
                                                 
{'loss': '0.0015', 'grad_norm': '0.1594', 'learning_rate': '3.394e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.07', 'tokens/total': 10435424, 'tokens/trainable': 157964, 'epoch': '1.348'}
-
 67%|██████▋   | 345/512 [38:55<18:41,  6.72s/it]
 68%|██████▊   | 346/512 [39:01<18:33,  6.71s/it]
                                                 
{'loss': '3.834e-05', 'grad_norm': '0.001656', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '30.35', 'tokens/total': 10465760, 'tokens/trainable': 158390, 'epoch': '1.352'}
-
 68%|██████▊   | 346/512 [39:01<18:33,  6.71s/it]
 68%|██████▊   | 347/512 [39:08<18:24,  6.69s/it]
                                                 
{'loss': '0.0003919', 'grad_norm': '0.02406', 'learning_rate': '3.343e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.47', 'tokens/total': 10495904, 'tokens/trainable': 158848, 'epoch': '1.355'}
-
 68%|██████▊   | 347/512 [39:08<18:24,  6.69s/it]
 68%|██████▊   | 348/512 [39:15<18:21,  6.72s/it]
                                                 
{'loss': '0.0006162', 'grad_norm': '0.04579', 'learning_rate': '3.317e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.63', 'tokens/total': 10526624, 'tokens/trainable': 159270, 'epoch': '1.359'}
-
 68%|██████▊   | 348/512 [39:15<18:21,  6.72s/it]
 68%|██████▊   | 349/512 [39:21<18:13,  6.71s/it]
                                                 
{'loss': '0.001287', 'grad_norm': '0.1865', 'learning_rate': '3.292e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.46', 'tokens/total': 10556752, 'tokens/trainable': 159720, 'epoch': '1.363'}
-
 68%|██████▊   | 349/512 [39:21<18:13,  6.71s/it]
 68%|██████▊   | 350/512 [39:28<18:12,  6.74s/it]
                                                 
{'loss': '0.0002092', 'grad_norm': '0.01593', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.09', 'tokens/total': 10587264, 'tokens/trainable': 160187, 'epoch': '1.367'}
-
 68%|██████▊   | 350/512 [39:28<18:12,  6.74s/it]
 69%|██████▊   | 351/512 [39:35<18:04,  6.74s/it]
                                                 
{'loss': '0.0001318', 'grad_norm': '0.008725', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 10617504, 'tokens/trainable': 160645, 'epoch': '1.371'}
-
 69%|██████▊   | 351/512 [39:35<18:04,  6.74s/it]
 69%|██████▉   | 352/512 [39:42<17:56,  6.73s/it]
                                                 
{'loss': '0.0008062', 'grad_norm': '0.06378', 'learning_rate': '3.217e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.42', 'tokens/total': 10647760, 'tokens/trainable': 161102, 'epoch': '1.375'}
-
 69%|██████▉   | 352/512 [39:42<17:56,  6.73s/it][2026-08-18 13:18:57,958] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
-
 69%|██████▉   | 353/512 [39:50<19:17,  7.28s/it]
                                                 
{'loss': '7.239e-05', 'grad_norm': '0.00372', 'learning_rate': '3.192e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.86', 'tokens/total': 10678080, 'tokens/trainable': 161555, 'epoch': '1.379'}
-
 69%|██████▉   | 353/512 [39:50<19:17,  7.28s/it]
 69%|██████▉   | 354/512 [39:57<18:43,  7.11s/it]
                                                 
{'loss': '3.681e-05', 'grad_norm': '0.002151', 'learning_rate': '3.167e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.41', 'tokens/train_per_sec_per_gpu': '36.65', 'tokens/total': 10708544, 'tokens/trainable': 162020, 'epoch': '1.383'}
-
 69%|██████▉   | 354/512 [39:57<18:43,  7.11s/it]
 69%|██████▉   | 355/512 [40:04<18:17,  6.99s/it]
                                                 
{'loss': '4.372e-05', 'grad_norm': '0.002642', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 10739024, 'tokens/trainable': 162448, 'epoch': '1.387'}
-
 69%|██████▉   | 355/512 [40:04<18:17,  6.99s/it]
 70%|██████▉   | 356/512 [40:10<17:53,  6.88s/it]
                                                 
{'loss': '7.116e-05', 'grad_norm': '0.003865', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.61', 'memory/max_allocated (GiB)': '33.61', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '28.2', 'tokens/total': 10769024, 'tokens/trainable': 162863, 'epoch': '1.391'}
-
 70%|██████▉   | 356/512 [40:10<17:53,  6.88s/it]
 70%|██████▉   | 357/512 [40:17<17:38,  6.83s/it]
                                                 
{'loss': '5.407e-05', 'grad_norm': '0.0179', 'learning_rate': '3.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 10799328, 'tokens/trainable': 163318, 'epoch': '1.395'}
-
 70%|██████▉   | 357/512 [40:17<17:38,  6.83s/it]
 70%|██████▉   | 358/512 [40:24<17:28,  6.81s/it]
                                                 
{'loss': '0.01498', 'grad_norm': '0.8479', 'learning_rate': '3.068e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 10829488, 'tokens/trainable': 163767, 'epoch': '1.398'}
-
 70%|██████▉   | 358/512 [40:24<17:28,  6.81s/it]
 70%|███████   | 359/512 [40:31<17:16,  6.78s/it]
                                                 
{'loss': '3.039e-05', 'grad_norm': '0.00157', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '31.78', 'tokens/total': 10859936, 'tokens/trainable': 164208, 'epoch': '1.402'}
-
 70%|███████   | 359/512 [40:31<17:16,  6.78s/it]
 70%|███████   | 360/512 [40:37<17:07,  6.76s/it]
                                                 
{'loss': '0.0001003', 'grad_norm': '0.01467', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.08', 'tokens/total': 10890400, 'tokens/trainable': 164655, 'epoch': '1.406'}
-
 70%|███████   | 360/512 [40:37<17:07,  6.76s/it]
 71%|███████   | 361/512 [40:44<16:56,  6.73s/it]
                                                 
{'loss': '0.00261', 'grad_norm': '0.3479', 'learning_rate': '2.995e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 10920624, 'tokens/trainable': 165100, 'epoch': '1.41'}
-
 71%|███████   | 361/512 [40:44<16:56,  6.73s/it]
 71%|███████   | 362/512 [40:51<16:51,  6.74s/it]
                                                 
{'loss': '6.117e-05', 'grad_norm': '0.004394', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.84', 'tokens/total': 10951136, 'tokens/trainable': 165587, 'epoch': '1.414'}
-
 71%|███████   | 362/512 [40:51<16:51,  6.74s/it]
 71%|███████   | 363/512 [40:57<16:41,  6.72s/it]
                                                 
{'loss': '0.0001775', 'grad_norm': '0.0162', 'learning_rate': '2.947e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 10981344, 'tokens/trainable': 166034, 'epoch': '1.418'}
-
 71%|███████   | 363/512 [40:57<16:41,  6.72s/it]
 71%|███████   | 364/512 [41:04<16:34,  6.72s/it]
                                                 
{'loss': '0.0005034', 'grad_norm': '0.09478', 'learning_rate': '2.924e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 11011552, 'tokens/trainable': 166485, 'epoch': '1.422'}
-
 71%|███████   | 364/512 [41:04<16:34,  6.72s/it]
 71%|███████▏  | 365/512 [41:11<16:28,  6.72s/it]
                                                 
{'loss': '0.0001905', 'grad_norm': '0.0149', 'learning_rate': '2.9e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 11041872, 'tokens/trainable': 166973, 'epoch': '1.426'}
-
 71%|███████▏  | 365/512 [41:11<16:28,  6.72s/it]
 71%|███████▏  | 366/512 [41:18<16:20,  6.72s/it]
                                                 
{'loss': '0.0001006', 'grad_norm': '0.005556', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '29.1', 'tokens/total': 11072192, 'tokens/trainable': 167414, 'epoch': '1.43'}
-
 71%|███████▏  | 366/512 [41:18<16:20,  6.72s/it]
 72%|███████▏  | 367/512 [41:24<16:15,  6.72s/it]
                                                 
{'loss': '0.0005889', 'grad_norm': '0.2071', 'learning_rate': '2.853e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.11', 'tokens/total': 11102736, 'tokens/trainable': 167851, 'epoch': '1.434'}
-
 72%|███████▏  | 367/512 [41:24<16:15,  6.72s/it]
 72%|███████▏  | 368/512 [41:31<16:07,  6.72s/it]
                                                 
{'loss': '1.23e-05', 'grad_norm': '0.0007192', 'learning_rate': '2.829e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.48', 'tokens/total': 11133056, 'tokens/trainable': 168294, 'epoch': '1.438'}
-
 72%|███████▏  | 368/512 [41:31<16:07,  6.72s/it]
 72%|███████▏  | 369/512 [41:38<16:04,  6.74s/it]
                                                 
{'loss': '0.0002043', 'grad_norm': '0.03088', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.08', 'tokens/total': 11163568, 'tokens/trainable': 168769, 'epoch': '1.441'}
-
 72%|███████▏  | 369/512 [41:38<16:04,  6.74s/it]
 72%|███████▏  | 370/512 [41:45<15:59,  6.75s/it]
                                                 
{'loss': '0.0001146', 'grad_norm': '0.008817', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 11193936, 'tokens/trainable': 169270, 'epoch': '1.445'}
-
 72%|███████▏  | 370/512 [41:45<15:59,  6.75s/it]
 72%|███████▏  | 371/512 [41:51<15:53,  6.77s/it]
                                                 
{'loss': '0.0001343', 'grad_norm': '0.02386', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 11224480, 'tokens/trainable': 169711, 'epoch': '1.449'}
-
 72%|███████▏  | 371/512 [41:51<15:53,  6.77s/it]
 73%|███████▎  | 372/512 [41:58<15:47,  6.77s/it]
                                                 
{'loss': '0.000151', 'grad_norm': '0.0105', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 11254912, 'tokens/trainable': 170185, 'epoch': '1.453'}
-
 73%|███████▎  | 372/512 [41:58<15:47,  6.77s/it]
 73%|███████▎  | 373/512 [42:05<15:39,  6.76s/it]
                                                 
{'loss': '0.0002608', 'grad_norm': '0.02697', 'learning_rate': '2.714e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 11285104, 'tokens/trainable': 170650, 'epoch': '1.457'}
-
 73%|███████▎  | 373/512 [42:05<15:39,  6.76s/it]
 73%|███████▎  | 374/512 [42:12<15:31,  6.75s/it]
                                                 
{'loss': '0.0001453', 'grad_norm': '0.02033', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '31.98', 'tokens/total': 11315600, 'tokens/trainable': 171126, 'epoch': '1.461'}
-
 73%|███████▎  | 374/512 [42:12<15:31,  6.75s/it]
 73%|███████▎  | 375/512 [42:18<15:21,  6.72s/it]
                                                 
{'loss': '0.0004557', 'grad_norm': '0.03326', 'learning_rate': '2.668e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 11345776, 'tokens/trainable': 171600, 'epoch': '1.465'}
-
 73%|███████▎  | 375/512 [42:18<15:21,  6.72s/it]
 73%|███████▎  | 376/512 [42:25<15:13,  6.71s/it]
                                                 
{'loss': '0.0001491', 'grad_norm': '0.01729', 'learning_rate': '2.646e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '28.09', 'tokens/total': 11376208, 'tokens/trainable': 172017, 'epoch': '1.469'}
-
 73%|███████▎  | 376/512 [42:25<15:13,  6.71s/it]
 74%|███████▎  | 377/512 [42:32<15:06,  6.72s/it]
                                                 
{'loss': '6.834e-05', 'grad_norm': '0.04457', 'learning_rate': '2.624e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.99', 'tokens/total': 11406512, 'tokens/trainable': 172493, 'epoch': '1.473'}
-
 74%|███████▎  | 377/512 [42:32<15:06,  6.72s/it]
 74%|███████▍  | 378/512 [42:38<15:00,  6.72s/it]
                                                 
{'loss': '3.245e-05', 'grad_norm': '0.002269', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.27', 'tokens/total': 11436544, 'tokens/trainable': 172951, 'epoch': '1.477'}
-
 74%|███████▍  | 378/512 [42:38<15:00,  6.72s/it]
 74%|███████▍  | 379/512 [42:45<14:53,  6.71s/it]
                                                 
{'loss': '5.487e-05', 'grad_norm': '0.003817', 'learning_rate': '2.579e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.63', 'tokens/total': 11467008, 'tokens/trainable': 173435, 'epoch': '1.48'}
-
 74%|███████▍  | 379/512 [42:45<14:53,  6.71s/it]
 74%|███████▍  | 380/512 [42:52<14:46,  6.71s/it]
                                                 
{'loss': '8.241e-05', 'grad_norm': '0.005269', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.4', 'tokens/total': 11497184, 'tokens/trainable': 173900, 'epoch': '1.484'}
-
 74%|███████▍  | 380/512 [42:52<14:46,  6.71s/it]
 74%|███████▍  | 381/512 [42:59<14:40,  6.72s/it]
                                                 
{'loss': '0.0001424', 'grad_norm': '0.01606', 'learning_rate': '2.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 11527600, 'tokens/trainable': 174371, 'epoch': '1.488'}
-
 74%|███████▍  | 381/512 [42:59<14:40,  6.72s/it]
 75%|███████▍  | 382/512 [43:05<14:33,  6.72s/it]
                                                 
{'loss': '5.001e-05', 'grad_norm': '0.005853', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 11557808, 'tokens/trainable': 174826, 'epoch': '1.492'}
-
 75%|███████▍  | 382/512 [43:05<14:33,  6.72s/it]
 75%|███████▍  | 383/512 [43:12<14:27,  6.73s/it]
                                                 
{'loss': '4.21e-05', 'grad_norm': '0.003166', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 11588352, 'tokens/trainable': 175276, 'epoch': '1.496'}
-
 75%|███████▍  | 383/512 [43:12<14:27,  6.73s/it]
 75%|███████▌  | 384/512 [43:19<14:21,  6.73s/it]
                                                 
{'loss': '0.001039', 'grad_norm': '0.08839', 'learning_rate': '2.47e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.69', 'tokens/total': 11618640, 'tokens/trainable': 175720, 'epoch': '1.5'}
-
 75%|███████▌  | 384/512 [43:19<14:21,  6.73s/it][2026-08-18 13:22:34,933] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
-
 75%|███████▌  | 385/512 [43:27<15:26,  7.30s/it]
                                                 
{'loss': '0.000537', 'grad_norm': '0.09447', 'learning_rate': '2.449e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.23', 'tokens/total': 11649072, 'tokens/trainable': 176203, 'epoch': '1.504'}
-
 75%|███████▌  | 385/512 [43:27<15:26,  7.30s/it]
 75%|███████▌  | 386/512 [43:34<14:57,  7.12s/it]
                                                 
{'loss': '1.903e-05', 'grad_norm': '0.001336', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 11679552, 'tokens/trainable': 176651, 'epoch': '1.508'}
-
 75%|███████▌  | 386/512 [43:34<14:57,  7.12s/it]
 76%|███████▌  | 387/512 [43:41<14:35,  7.00s/it]
                                                 
{'loss': '7.681e-05', 'grad_norm': '0.01169', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 11709760, 'tokens/trainable': 177115, 'epoch': '1.512'}
-
 76%|███████▌  | 387/512 [43:41<14:35,  7.00s/it]
 76%|███████▌  | 388/512 [43:47<14:17,  6.92s/it]
                                                 
{'loss': '0.0007352', 'grad_norm': '0.09523', 'learning_rate': '2.385e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 11740288, 'tokens/trainable': 177601, 'epoch': '1.516'}
-
 76%|███████▌  | 388/512 [43:47<14:17,  6.92s/it]
 76%|███████▌  | 389/512 [43:54<14:02,  6.85s/it]
                                                 
{'loss': '6.468e-05', 'grad_norm': '0.005972', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 11770592, 'tokens/trainable': 178050, 'epoch': '1.52'}
-
 76%|███████▌  | 389/512 [43:54<14:02,  6.85s/it]
 76%|███████▌  | 390/512 [44:01<13:51,  6.82s/it]
                                                 
{'loss': '1.599e-05', 'grad_norm': '0.001025', 'learning_rate': '2.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.38', 'tokens/total': 11800752, 'tokens/trainable': 178531, 'epoch': '1.523'}
-
 76%|███████▌  | 390/512 [44:01<13:51,  6.82s/it]
 76%|███████▋  | 391/512 [44:08<13:42,  6.80s/it]
                                                 
{'loss': '0.0004187', 'grad_norm': '0.05442', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 11831360, 'tokens/trainable': 179000, 'epoch': '1.527'}
-
 76%|███████▋  | 391/512 [44:08<13:42,  6.80s/it]
 77%|███████▋  | 392/512 [44:14<13:30,  6.75s/it]
                                                 
{'loss': '1.532e-05', 'grad_norm': '0.002925', 'learning_rate': '2.303e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 11861552, 'tokens/trainable': 179457, 'epoch': '1.531'}
-
 77%|███████▋  | 392/512 [44:14<13:30,  6.75s/it]
 77%|███████▋  | 393/512 [44:21<13:20,  6.73s/it]
                                                 
{'loss': '0.002704', 'grad_norm': '1.64', 'learning_rate': '2.282e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 11891904, 'tokens/trainable': 179890, 'epoch': '1.535'}
-
 77%|███████▋  | 393/512 [44:21<13:20,  6.73s/it]
 77%|███████▋  | 394/512 [44:28<13:12,  6.72s/it]
                                                 
{'loss': '3.968e-05', 'grad_norm': '0.003221', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 11922208, 'tokens/trainable': 180341, 'epoch': '1.539'}
-
 77%|███████▋  | 394/512 [44:28<13:12,  6.72s/it]
 77%|███████▋  | 395/512 [44:34<13:03,  6.70s/it]
                                                 
{'loss': '0.0001106', 'grad_norm': '0.0131', 'learning_rate': '2.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 11952672, 'tokens/trainable': 180762, 'epoch': '1.543'}
-
 77%|███████▋  | 395/512 [44:34<13:03,  6.70s/it]
 77%|███████▋  | 396/512 [44:41<12:58,  6.71s/it]
                                                 
{'loss': '8.803e-05', 'grad_norm': '0.01553', 'learning_rate': '2.222e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.04', 'tokens/total': 11983088, 'tokens/trainable': 181226, 'epoch': '1.547'}
-
 77%|███████▋  | 396/512 [44:41<12:58,  6.71s/it]
 78%|███████▊  | 397/512 [44:48<12:52,  6.72s/it]
                                                 
{'loss': '8.931e-05', 'grad_norm': '0.01608', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.15', 'tokens/total': 12013488, 'tokens/trainable': 181714, 'epoch': '1.551'}
-
 78%|███████▊  | 397/512 [44:48<12:52,  6.72s/it]
 78%|███████▊  | 398/512 [44:54<12:44,  6.70s/it]
                                                 
{'loss': '0.0001537', 'grad_norm': '0.05752', 'learning_rate': '2.183e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.52', 'tokens/total': 12043968, 'tokens/trainable': 182142, 'epoch': '1.555'}
-
 78%|███████▊  | 398/512 [44:54<12:44,  6.70s/it]
 78%|███████▊  | 399/512 [45:01<12:38,  6.71s/it]
                                                 
{'loss': '1.811e-05', 'grad_norm': '0.001174', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.82', 'tokens/total': 12074160, 'tokens/trainable': 182604, 'epoch': '1.559'}
-
 78%|███████▊  | 399/512 [45:01<12:38,  6.71s/it]
 78%|███████▊  | 400/512 [45:08<12:30,  6.70s/it]
                                                 
{'loss': '5.335e-05', 'grad_norm': '0.004808', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '25.28', 'tokens/total': 12104352, 'tokens/trainable': 183011, 'epoch': '1.562'}
-
 78%|███████▊  | 400/512 [45:08<12:30,  6.70s/it]
 78%|███████▊  | 401/512 [45:15<12:23,  6.70s/it]
                                                 
{'loss': '6.45e-05', 'grad_norm': '0.00652', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.69', 'tokens/total': 12134240, 'tokens/trainable': 183459, 'epoch': '1.566'}
-
 78%|███████▊  | 401/512 [45:15<12:23,  6.70s/it]
 79%|███████▊  | 402/512 [45:21<12:19,  6.72s/it]
                                                 
{'loss': '6.929e-05', 'grad_norm': '0.006228', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.61', 'tokens/total': 12164480, 'tokens/trainable': 183959, 'epoch': '1.57'}
-
 79%|███████▊  | 402/512 [45:21<12:19,  6.72s/it]
 79%|███████▊  | 403/512 [45:28<12:13,  6.73s/it]
                                                 
{'loss': '0.002115', 'grad_norm': '0.3261', 'learning_rate': '2.086e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.46', 'tokens/total': 12194640, 'tokens/trainable': 184431, 'epoch': '1.574'}
-
 79%|███████▊  | 403/512 [45:28<12:13,  6.73s/it]
 79%|███████▉  | 404/512 [45:35<12:06,  6.73s/it]
                                                 
{'loss': '4.601e-05', 'grad_norm': '0.007742', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.34', 'tokens/total': 12224960, 'tokens/trainable': 184867, 'epoch': '1.578'}
-
 79%|███████▉  | 404/512 [45:35<12:06,  6.73s/it]
 79%|███████▉  | 405/512 [45:42<11:58,  6.71s/it]
                                                 
{'loss': '5.004e-05', 'grad_norm': '0.004995', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.25', 'tokens/total': 12255008, 'tokens/trainable': 185300, 'epoch': '1.582'}
-
 79%|███████▉  | 405/512 [45:42<11:58,  6.71s/it]
 79%|███████▉  | 406/512 [45:48<11:51,  6.71s/it]
                                                 
{'loss': '3.442e-05', 'grad_norm': '0.004599', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 12285248, 'tokens/trainable': 185754, 'epoch': '1.586'}
-
 79%|███████▉  | 406/512 [45:48<11:51,  6.71s/it]
 79%|███████▉  | 407/512 [45:55<11:43,  6.70s/it]
                                                 
{'loss': '1.92e-05', 'grad_norm': '0.001146', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.8', 'tokens/total': 12315552, 'tokens/trainable': 186213, 'epoch': '1.59'}
-
 79%|███████▉  | 407/512 [45:55<11:43,  6.70s/it]
 80%|███████▉  | 408/512 [46:02<11:37,  6.70s/it]
                                                 
{'loss': '5.249e-05', 'grad_norm': '0.004688', 'learning_rate': '1.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.72', 'tokens/total': 12345904, 'tokens/trainable': 186659, 'epoch': '1.594'}
-
 80%|███████▉  | 408/512 [46:02<11:37,  6.70s/it]
 80%|███████▉  | 409/512 [46:08<11:30,  6.71s/it]
                                                 
{'loss': '3.172e-05', 'grad_norm': '0.002638', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 12376336, 'tokens/trainable': 187083, 'epoch': '1.598'}
-
 80%|███████▉  | 409/512 [46:08<11:30,  6.71s/it]
 80%|████████  | 410/512 [46:15<11:25,  6.72s/it]
                                                 
{'loss': '5.453e-05', 'grad_norm': '0.007202', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 12406880, 'tokens/trainable': 187530, 'epoch': '1.602'}
-
 80%|████████  | 410/512 [46:15<11:25,  6.72s/it]
 80%|████████  | 411/512 [46:22<11:18,  6.71s/it]
                                                 
{'loss': '3.531e-05', 'grad_norm': '0.00717', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 12437088, 'tokens/trainable': 187972, 'epoch': '1.605'}
-
 80%|████████  | 411/512 [46:22<11:18,  6.71s/it]
 80%|████████  | 412/512 [46:28<10:55,  6.55s/it]
                                                 
{'loss': '3.509e-05', 'grad_norm': '0.00389', 'learning_rate': '1.922e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '39.89', 'tokens/total': 12465440, 'tokens/trainable': 188417, 'epoch': '1.609'}
-
 80%|████████  | 412/512 [46:28<10:55,  6.55s/it]
 81%|████████  | 413/512 [46:35<10:54,  6.62s/it]
                                                 
{'loss': '0.005698', 'grad_norm': '0.5135', 'learning_rate': '1.904e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 12495968, 'tokens/trainable': 188868, 'epoch': '1.613'}
-
 81%|████████  | 413/512 [46:35<10:54,  6.62s/it]
 81%|████████  | 414/512 [46:41<10:51,  6.65s/it]
                                                 
{'loss': '1.132e-05', 'grad_norm': '0.00101', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.85', 'tokens/total': 12526128, 'tokens/trainable': 189337, 'epoch': '1.617'}
-
 81%|████████  | 414/512 [46:41<10:51,  6.65s/it]
 81%|████████  | 415/512 [46:48<10:45,  6.65s/it]
                                                 
{'loss': '7.355e-06', 'grad_norm': '0.0004971', 'learning_rate': '1.87e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.82', 'tokens/total': 12556256, 'tokens/trainable': 189750, 'epoch': '1.621'}
-
 81%|████████  | 415/512 [46:48<10:45,  6.65s/it]
 81%|████████▏ | 416/512 [46:55<10:41,  6.68s/it]
                                                 
{'loss': '1.419e-05', 'grad_norm': '0.0008849', 'learning_rate': '1.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 12586736, 'tokens/trainable': 190216, 'epoch': '1.625'}
-
 81%|████████▏ | 416/512 [46:55<10:41,  6.68s/it][2026-08-18 13:26:11,049] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
-
 81%|████████▏ | 417/512 [47:03<11:29,  7.26s/it]
                                                 
{'loss': '2.454e-05', 'grad_norm': '0.002204', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 12617168, 'tokens/trainable': 190708, 'epoch': '1.629'}
-
 81%|████████▏ | 417/512 [47:03<11:29,  7.26s/it]
 82%|████████▏ | 418/512 [47:10<11:05,  7.08s/it]
                                                 
{'loss': '0.0002044', 'grad_norm': '0.02701', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.85', 'tokens/total': 12647680, 'tokens/trainable': 191126, 'epoch': '1.633'}
-
 82%|████████▏ | 418/512 [47:10<11:05,  7.08s/it]
 82%|████████▏ | 419/512 [47:17<10:46,  6.95s/it]
                                                 
{'loss': '0.0007251', 'grad_norm': '0.1568', 'learning_rate': '1.802e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.3', 'tokens/total': 12677952, 'tokens/trainable': 191540, 'epoch': '1.637'}
-
 82%|████████▏ | 419/512 [47:17<10:46,  6.95s/it]
 82%|████████▏ | 420/512 [47:23<10:32,  6.87s/it]
                                                 
{'loss': '3.131e-05', 'grad_norm': '0.00331', 'learning_rate': '1.786e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '36.65', 'tokens/total': 12708416, 'tokens/trainable': 192014, 'epoch': '1.641'}
-
 82%|████████▏ | 420/512 [47:23<10:32,  6.87s/it]
 82%|████████▏ | 421/512 [47:30<10:21,  6.83s/it]
                                                 
{'loss': '1.931e-05', 'grad_norm': '0.001256', 'learning_rate': '1.77e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '31.28', 'tokens/total': 12738640, 'tokens/trainable': 192463, 'epoch': '1.645'}
-
 82%|████████▏ | 421/512 [47:30<10:21,  6.83s/it]
 82%|████████▏ | 422/512 [47:37<10:10,  6.79s/it]
                                                 
{'loss': '9.236e-05', 'grad_norm': '0.016', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 12769232, 'tokens/trainable': 192918, 'epoch': '1.648'}
-
 82%|████████▏ | 422/512 [47:37<10:10,  6.79s/it]
 83%|████████▎ | 423/512 [47:44<10:03,  6.78s/it]
                                                 
{'loss': '0.0002709', 'grad_norm': '0.04851', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.35', 'tokens/total': 12799648, 'tokens/trainable': 193425, 'epoch': '1.652'}
-
 83%|████████▎ | 423/512 [47:44<10:03,  6.78s/it]
 83%|████████▎ | 424/512 [47:50<09:55,  6.76s/it]
                                                 
{'loss': '6.991e-05', 'grad_norm': '0.01011', 'learning_rate': '1.722e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.23', 'tokens/total': 12829920, 'tokens/trainable': 193898, 'epoch': '1.656'}
-
 83%|████████▎ | 424/512 [47:50<09:55,  6.76s/it]
 83%|████████▎ | 425/512 [47:57<09:46,  6.74s/it]
                                                 
{'loss': '5.418e-05', 'grad_norm': '0.005467', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.24', 'tokens/total': 12860112, 'tokens/trainable': 194315, 'epoch': '1.66'}
-
 83%|████████▎ | 425/512 [47:57<09:46,  6.74s/it]
 83%|████████▎ | 426/512 [48:04<09:39,  6.74s/it]
                                                 
{'loss': '2.498e-05', 'grad_norm': '0.00158', 'learning_rate': '1.69e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 12890320, 'tokens/trainable': 194796, 'epoch': '1.664'}
-
 83%|████████▎ | 426/512 [48:04<09:39,  6.74s/it]
 83%|████████▎ | 427/512 [48:11<09:33,  6.74s/it]
                                                 
{'loss': '0.000107', 'grad_norm': '0.02416', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 12920656, 'tokens/trainable': 195262, 'epoch': '1.668'}
-
 83%|████████▎ | 427/512 [48:11<09:33,  6.74s/it]
 84%|████████▎ | 428/512 [48:17<09:26,  6.74s/it]
                                                 
{'loss': '7.484e-06', 'grad_norm': '0.0002809', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 12951040, 'tokens/trainable': 195680, 'epoch': '1.672'}
-
 84%|████████▎ | 428/512 [48:17<09:26,  6.74s/it]
 84%|████████▍ | 429/512 [48:24<09:17,  6.72s/it]
                                                 
{'loss': '1.992e-05', 'grad_norm': '0.001109', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.83', 'tokens/total': 12981232, 'tokens/trainable': 196145, 'epoch': '1.676'}
-
 84%|████████▍ | 429/512 [48:24<09:17,  6.72s/it]
 84%|████████▍ | 430/512 [48:31<09:11,  6.72s/it]
                                                 
{'loss': '0.0001097', 'grad_norm': '0.02079', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.39', 'tokens/total': 13011616, 'tokens/trainable': 196626, 'epoch': '1.68'}
-
 84%|████████▍ | 430/512 [48:31<09:11,  6.72s/it]
 84%|████████▍ | 431/512 [48:38<09:07,  6.76s/it]
                                                 
{'loss': '0.002679', 'grad_norm': '0.4408', 'learning_rate': '1.615e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '31.14', 'tokens/total': 13042208, 'tokens/trainable': 197067, 'epoch': '1.684'}
-
 84%|████████▍ | 431/512 [48:38<09:07,  6.76s/it]
 84%|████████▍ | 432/512 [48:44<09:00,  6.75s/it]
                                                 
{'loss': '2.082e-05', 'grad_norm': '0.002634', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.61', 'tokens/total': 13072800, 'tokens/trainable': 197565, 'epoch': '1.688'}
-
 84%|████████▍ | 432/512 [48:44<09:00,  6.75s/it]
 85%|████████▍ | 433/512 [48:51<08:52,  6.73s/it]
                                                 
{'loss': '1.44e-05', 'grad_norm': '0.004509', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 13103328, 'tokens/trainable': 198015, 'epoch': '1.691'}
-
 85%|████████▍ | 433/512 [48:51<08:52,  6.73s/it]
 85%|████████▍ | 434/512 [48:58<08:44,  6.73s/it]
                                                 
{'loss': '0.0001532', 'grad_norm': '0.01481', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 13133472, 'tokens/trainable': 198442, 'epoch': '1.695'}
-
 85%|████████▍ | 434/512 [48:58<08:44,  6.73s/it]
 85%|████████▍ | 435/512 [49:04<08:37,  6.72s/it]
                                                 
{'loss': '1.444e-05', 'grad_norm': '0.003779', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.3', 'tokens/total': 13164064, 'tokens/trainable': 198914, 'epoch': '1.699'}
-
 85%|████████▍ | 435/512 [49:04<08:37,  6.72s/it]
 85%|████████▌ | 436/512 [49:11<08:31,  6.73s/it]
                                                 
{'loss': '0.001711', 'grad_norm': '0.2497', 'learning_rate': '1.544e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.35', 'tokens/total': 13194464, 'tokens/trainable': 199355, 'epoch': '1.703'}
-
 85%|████████▌ | 436/512 [49:11<08:31,  6.73s/it]
 85%|████████▌ | 437/512 [49:18<08:24,  6.73s/it]
                                                 
{'loss': '2.94e-05', 'grad_norm': '0.00248', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 13224992, 'tokens/trainable': 199830, 'epoch': '1.707'}
-
 85%|████████▌ | 437/512 [49:18<08:24,  6.73s/it]
 86%|████████▌ | 438/512 [49:25<08:18,  6.74s/it]
                                                 
{'loss': '1.559e-05', 'grad_norm': '0.001478', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 13255296, 'tokens/trainable': 200338, 'epoch': '1.711'}
-
 86%|████████▌ | 438/512 [49:25<08:18,  6.74s/it]
 86%|████████▌ | 439/512 [49:31<08:09,  6.71s/it]
                                                 
{'loss': '1.468e-05', 'grad_norm': '0.001041', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 13285344, 'tokens/trainable': 200785, 'epoch': '1.715'}
-
 86%|████████▌ | 439/512 [49:31<08:09,  6.71s/it]
 86%|████████▌ | 440/512 [49:38<08:04,  6.72s/it]
                                                 
{'loss': '8.095e-06', 'grad_norm': '0.0004247', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 13315632, 'tokens/trainable': 201267, 'epoch': '1.719'}
-
 86%|████████▌ | 440/512 [49:38<08:04,  6.72s/it]
 86%|████████▌ | 441/512 [49:45<07:57,  6.73s/it]
                                                 
{'loss': '1.743e-05', 'grad_norm': '0.00263', 'learning_rate': '1.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 13346048, 'tokens/trainable': 201734, 'epoch': '1.723'}
-
 86%|████████▌ | 441/512 [49:45<07:57,  6.73s/it]
 86%|████████▋ | 442/512 [49:51<07:50,  6.72s/it]
                                                 
{'loss': '8.656e-06', 'grad_norm': '0.0005576', 'learning_rate': '1.464e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 13376304, 'tokens/trainable': 202174, 'epoch': '1.727'}
-
 86%|████████▋ | 442/512 [49:51<07:50,  6.72s/it]
 87%|████████▋ | 443/512 [49:58<07:44,  6.73s/it]
                                                 
{'loss': '1.409e-05', 'grad_norm': '0.0009604', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.13', 'tokens/total': 13406720, 'tokens/trainable': 202612, 'epoch': '1.73'}
-
 87%|████████▋ | 443/512 [49:58<07:44,  6.73s/it]
 87%|████████▋ | 444/512 [50:05<07:33,  6.68s/it]
                                                 
{'loss': '4.584e-06', 'grad_norm': '0.0001988', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.96', 'tokens/total': 13436464, 'tokens/trainable': 203020, 'epoch': '1.734'}
-
 87%|████████▋ | 444/512 [50:05<07:33,  6.68s/it]
 87%|████████▋ | 445/512 [50:11<07:27,  6.68s/it]
                                                 
{'loss': '0.000278', 'grad_norm': '0.05945', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 13466672, 'tokens/trainable': 203479, 'epoch': '1.738'}
-
 87%|████████▋ | 445/512 [50:11<07:27,  6.68s/it]
 87%|████████▋ | 446/512 [50:18<07:21,  6.69s/it]
                                                 
{'loss': '3.424e-05', 'grad_norm': '0.005872', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 13496736, 'tokens/trainable': 203946, 'epoch': '1.742'}
-
 87%|████████▋ | 446/512 [50:18<07:21,  6.69s/it]
 87%|████████▋ | 447/512 [50:25<07:14,  6.69s/it]
                                                 
{'loss': '3.27e-05', 'grad_norm': '0.006014', 'learning_rate': '1.402e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '37.43', 'tokens/total': 13526928, 'tokens/trainable': 204382, 'epoch': '1.746'}
-
 87%|████████▋ | 447/512 [50:25<07:14,  6.69s/it]
 88%|████████▊ | 448/512 [50:31<07:08,  6.69s/it]
                                                 
{'loss': '0.0003225', 'grad_norm': '0.05103', 'learning_rate': '1.39e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 13557392, 'tokens/trainable': 204830, 'epoch': '1.75'}
-
 88%|████████▊ | 448/512 [50:31<07:08,  6.69s/it][2026-08-18 13:29:47,708] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
-
 88%|████████▊ | 449/512 [50:40<07:41,  7.32s/it]
                                                 
{'loss': '0.0008941', 'grad_norm': '0.1325', 'learning_rate': '1.378e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.51', 'tokens/total': 13587776, 'tokens/trainable': 205300, 'epoch': '1.754'}
-
 88%|████████▊ | 449/512 [50:40<07:41,  7.32s/it]
 88%|████████▊ | 450/512 [50:47<07:22,  7.14s/it]
                                                 
{'loss': '0.001293', 'grad_norm': '0.2464', 'learning_rate': '1.367e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '30.62', 'tokens/total': 13618112, 'tokens/trainable': 205726, 'epoch': '1.758'}
-
 88%|████████▊ | 450/512 [50:47<07:22,  7.14s/it]
 88%|████████▊ | 451/512 [50:54<07:08,  7.03s/it]
                                                 
{'loss': '6.57e-05', 'grad_norm': '0.01153', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '31.55', 'tokens/total': 13648512, 'tokens/trainable': 206163, 'epoch': '1.762'}
-
 88%|████████▊ | 451/512 [50:54<07:08,  7.03s/it]
 88%|████████▊ | 452/512 [51:00<06:55,  6.93s/it]
                                                 
{'loss': '0.0156', 'grad_norm': '1.082', 'learning_rate': '1.344e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.41', 'tokens/total': 13678704, 'tokens/trainable': 206619, 'epoch': '1.766'}
-
 88%|████████▊ | 452/512 [51:00<06:55,  6.93s/it]
 88%|████████▊ | 453/512 [51:07<06:46,  6.88s/it]
                                                 
{'loss': '1.203e-05', 'grad_norm': '0.0007327', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.18', 'tokens/total': 13709232, 'tokens/trainable': 207101, 'epoch': '1.77'}
-
 88%|████████▊ | 453/512 [51:07<06:46,  6.88s/it]
 89%|████████▊ | 454/512 [51:14<06:35,  6.83s/it]
                                                 
{'loss': '1.262e-05', 'grad_norm': '0.002392', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.69', 'tokens/total': 13739600, 'tokens/trainable': 207546, 'epoch': '1.773'}
-
 89%|████████▊ | 454/512 [51:14<06:35,  6.83s/it]
 89%|████████▉ | 455/512 [51:21<06:27,  6.79s/it]
                                                 
{'loss': '2.477e-05', 'grad_norm': '0.002097', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '34.25', 'tokens/total': 13769936, 'tokens/trainable': 207989, 'epoch': '1.777'}
-
 89%|████████▉ | 455/512 [51:21<06:27,  6.79s/it]
 89%|████████▉ | 456/512 [51:27<06:18,  6.76s/it]
                                                 
{'loss': '1.457e-05', 'grad_norm': '0.0007976', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 13800240, 'tokens/trainable': 208402, 'epoch': '1.781'}
-
 89%|████████▉ | 456/512 [51:27<06:18,  6.76s/it]
 89%|████████▉ | 457/512 [51:34<06:11,  6.75s/it]
                                                 
{'loss': '7.903e-05', 'grad_norm': '0.01138', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.61', 'tokens/total': 13830512, 'tokens/trainable': 208904, 'epoch': '1.785'}
-
 89%|████████▉ | 457/512 [51:34<06:11,  6.75s/it]
 89%|████████▉ | 458/512 [51:41<06:03,  6.74s/it]
                                                 
{'loss': '1.925e-05', 'grad_norm': '0.001357', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.64', 'tokens/total': 13860832, 'tokens/trainable': 209382, 'epoch': '1.789'}
-
 89%|████████▉ | 458/512 [51:41<06:03,  6.74s/it]
 90%|████████▉ | 459/512 [51:47<05:56,  6.72s/it]
                                                 
{'loss': '1.119e-05', 'grad_norm': '0.000474', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 13890992, 'tokens/trainable': 209833, 'epoch': '1.793'}
-
 90%|████████▉ | 459/512 [51:47<05:56,  6.72s/it]
 90%|████████▉ | 460/512 [51:54<05:48,  6.71s/it]
                                                 
{'loss': '0.000175', 'grad_norm': '0.04677', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.5', 'tokens/total': 13921424, 'tokens/trainable': 210284, 'epoch': '1.797'}
-
 90%|████████▉ | 460/512 [51:54<05:48,  6.71s/it]
 90%|█████████ | 461/512 [52:01<05:41,  6.69s/it]
                                                 
{'loss': '9.222e-05', 'grad_norm': '0.02466', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 13951504, 'tokens/trainable': 210769, 'epoch': '1.801'}
-
 90%|█████████ | 461/512 [52:01<05:41,  6.69s/it]
 90%|█████████ | 462/512 [52:07<05:26,  6.54s/it]
                                                 
{'loss': '1.07e-05', 'grad_norm': '0.000474', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 13979728, 'tokens/trainable': 211193, 'epoch': '1.805'}
-
 90%|█████████ | 462/512 [52:07<05:26,  6.54s/it]
 90%|█████████ | 463/512 [52:14<05:23,  6.60s/it]
                                                 
{'loss': '1.394e-05', 'grad_norm': '0.0007591', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 14009936, 'tokens/trainable': 211655, 'epoch': '1.809'}
-
 90%|█████████ | 463/512 [52:14<05:23,  6.60s/it]
 91%|█████████ | 464/512 [52:20<05:17,  6.61s/it]
                                                 
{'loss': '2.675e-05', 'grad_norm': '0.002109', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.28', 'tokens/total': 14039872, 'tokens/trainable': 212085, 'epoch': '1.812'}
-
 91%|█████████ | 464/512 [52:20<05:17,  6.61s/it]
 91%|█████████ | 465/512 [52:27<05:12,  6.65s/it]
                                                 
{'loss': '1.082e-05', 'grad_norm': '0.0009436', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.61', 'tokens/total': 14070096, 'tokens/trainable': 212570, 'epoch': '1.816'}
-
 91%|█████████ | 465/512 [52:27<05:12,  6.65s/it]
 91%|█████████ | 466/512 [52:34<05:06,  6.66s/it]
                                                 
{'loss': '5.798e-05', 'grad_norm': '0.01104', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 14100192, 'tokens/trainable': 213011, 'epoch': '1.82'}
-
 91%|█████████ | 466/512 [52:34<05:06,  6.66s/it]
 91%|█████████ | 467/512 [52:41<05:00,  6.69s/it]
                                                 
{'loss': '2.876e-05', 'grad_norm': '0.008272', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.44', 'tokens/total': 14130448, 'tokens/trainable': 213450, 'epoch': '1.824'}
-
 91%|█████████ | 467/512 [52:41<05:00,  6.69s/it]
 91%|█████████▏| 468/512 [52:47<04:53,  6.68s/it]
                                                 
{'loss': '0.0001435', 'grad_norm': '0.02348', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.62', 'tokens/total': 14160480, 'tokens/trainable': 213898, 'epoch': '1.828'}
-
 91%|█████████▏| 468/512 [52:47<04:53,  6.68s/it]
 92%|█████████▏| 469/512 [52:54<04:47,  6.69s/it]
                                                 
{'loss': '2.56e-05', 'grad_norm': '0.002308', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.78', 'tokens/total': 14191152, 'tokens/trainable': 214366, 'epoch': '1.832'}
-
 92%|█████████▏| 469/512 [52:54<04:47,  6.69s/it]
 92%|█████████▏| 470/512 [53:00<04:35,  6.55s/it]
                                                 
{'loss': '2.02e-05', 'grad_norm': '0.001436', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.64', 'tokens/total': 14219392, 'tokens/trainable': 214800, 'epoch': '1.836'}
-
 92%|█████████▏| 470/512 [53:00<04:35,  6.55s/it]
 92%|█████████▏| 471/512 [53:07<04:31,  6.63s/it]
                                                 
{'loss': '4.487e-05', 'grad_norm': '0.0116', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 14249920, 'tokens/trainable': 215292, 'epoch': '1.84'}
-
 92%|█████████▏| 471/512 [53:07<04:31,  6.63s/it]
 92%|█████████▏| 472/512 [53:14<04:26,  6.65s/it]
                                                 
{'loss': '2.048e-05', 'grad_norm': '0.001163', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 14280048, 'tokens/trainable': 215741, 'epoch': '1.844'}
-
 92%|█████████▏| 472/512 [53:14<04:26,  6.65s/it]
 92%|█████████▏| 473/512 [53:20<04:20,  6.68s/it]
                                                 
{'loss': '5.009e-05', 'grad_norm': '0.006094', 'learning_rate': '1.149e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.64', 'tokens/total': 14310432, 'tokens/trainable': 216194, 'epoch': '1.848'}
-
 92%|█████████▏| 473/512 [53:20<04:20,  6.68s/it]
 93%|█████████▎| 474/512 [53:27<04:13,  6.68s/it]
                                                 
{'loss': '3.059e-05', 'grad_norm': '0.003623', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.3', 'tokens/total': 14340496, 'tokens/trainable': 216634, 'epoch': '1.852'}
-
 93%|█████████▎| 474/512 [53:27<04:13,  6.68s/it]
 93%|█████████▎| 475/512 [53:34<04:07,  6.68s/it]
                                                 
{'loss': '3.954e-05', 'grad_norm': '0.006924', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '30.48', 'tokens/total': 14370720, 'tokens/trainable': 217040, 'epoch': '1.855'}
-
 93%|█████████▎| 475/512 [53:34<04:07,  6.68s/it]
 93%|█████████▎| 476/512 [53:41<04:01,  6.71s/it]
                                                 
{'loss': '3.342e-05', 'grad_norm': '0.00236', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.55', 'tokens/total': 14401136, 'tokens/trainable': 217530, 'epoch': '1.859'}
-
 93%|█████████▎| 476/512 [53:41<04:01,  6.71s/it]
 93%|█████████▎| 477/512 [53:47<03:54,  6.70s/it]
                                                 
{'loss': '0.0001011', 'grad_norm': '0.009585', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 14431360, 'tokens/trainable': 217988, 'epoch': '1.863'}
-
 93%|█████████▎| 477/512 [53:47<03:54,  6.70s/it]
 93%|█████████▎| 478/512 [53:54<03:49,  6.74s/it]
                                                 
{'loss': '0.0002209', 'grad_norm': '0.03044', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.47', 'tokens/total': 14461968, 'tokens/trainable': 218461, 'epoch': '1.867'}
-
 93%|█████████▎| 478/512 [53:54<03:49,  6.74s/it]
 94%|█████████▎| 479/512 [54:01<03:41,  6.71s/it]
                                                 
{'loss': '0.0002753', 'grad_norm': '0.03494', 'learning_rate': '1.108e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.77', 'tokens/total': 14491984, 'tokens/trainable': 218909, 'epoch': '1.871'}
-
 94%|█████████▎| 479/512 [54:01<03:41,  6.71s/it]
 94%|█████████▍| 480/512 [54:07<03:35,  6.73s/it]
                                                 
{'loss': '0.0002576', 'grad_norm': '0.301', 'learning_rate': '1.102e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 14522544, 'tokens/trainable': 219350, 'epoch': '1.875'}
-
 94%|█████████▍| 480/512 [54:07<03:35,  6.73s/it][2026-08-18 13:33:23,660] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
-
 94%|█████████▍| 481/512 [54:16<03:46,  7.30s/it]
                                                 
{'loss': '0.0002504', 'grad_norm': '0.07506', 'learning_rate': '1.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '29.47', 'tokens/total': 14552560, 'tokens/trainable': 219776, 'epoch': '1.879'}
-
 94%|█████████▍| 481/512 [54:16<03:46,  7.30s/it]
 94%|█████████▍| 482/512 [54:23<03:34,  7.14s/it]
                                                 
{'loss': '1.01e-05', 'grad_norm': '0.0008993', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 14583056, 'tokens/trainable': 220284, 'epoch': '1.883'}
-
 94%|█████████▍| 482/512 [54:23<03:34,  7.14s/it]
 94%|█████████▍| 483/512 [54:30<03:22,  7.00s/it]
                                                 
{'loss': '2.53e-05', 'grad_norm': '0.001677', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 14613152, 'tokens/trainable': 220714, 'epoch': '1.887'}
-
 94%|█████████▍| 483/512 [54:30<03:22,  7.00s/it]
 95%|█████████▍| 484/512 [54:36<03:13,  6.91s/it]
                                                 
{'loss': '0.0001353', 'grad_norm': '0.01496', 'learning_rate': '1.079e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 14643328, 'tokens/trainable': 221156, 'epoch': '1.891'}
-
 95%|█████████▍| 484/512 [54:36<03:13,  6.91s/it]
 95%|█████████▍| 485/512 [54:43<03:04,  6.84s/it]
                                                 
{'loss': '1.105e-05', 'grad_norm': '0.004265', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 14673488, 'tokens/trainable': 221605, 'epoch': '1.895'}
-
 95%|█████████▍| 485/512 [54:43<03:04,  6.84s/it]
 95%|█████████▍| 486/512 [54:50<02:56,  6.79s/it]
                                                 
{'loss': '2.202e-05', 'grad_norm': '0.002974', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 14704000, 'tokens/trainable': 222042, 'epoch': '1.898'}
-
 95%|█████████▍| 486/512 [54:50<02:56,  6.79s/it]
 95%|█████████▌| 487/512 [54:56<02:48,  6.76s/it]
                                                 
{'loss': '7.452e-05', 'grad_norm': '0.008692', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.95', 'tokens/total': 14733984, 'tokens/trainable': 222474, 'epoch': '1.902'}
-
 95%|█████████▌| 487/512 [54:56<02:48,  6.76s/it]
 95%|█████████▌| 488/512 [55:03<02:42,  6.75s/it]
                                                 
{'loss': '7.341e-06', 'grad_norm': '0.0005811', 'learning_rate': '1.058e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 14764400, 'tokens/trainable': 222967, 'epoch': '1.906'}
-
 95%|█████████▌| 488/512 [55:03<02:42,  6.75s/it]
 96%|█████████▌| 489/512 [55:10<02:34,  6.73s/it]
                                                 
{'loss': '3.556e-05', 'grad_norm': '0.004224', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.78', 'tokens/total': 14794672, 'tokens/trainable': 223420, 'epoch': '1.91'}
-
 96%|█████████▌| 489/512 [55:10<02:34,  6.73s/it]
 96%|█████████▌| 490/512 [55:16<02:27,  6.71s/it]
                                                 
{'loss': '2.799e-05', 'grad_norm': '0.01521', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '29.42', 'tokens/total': 14824784, 'tokens/trainable': 223864, 'epoch': '1.914'}
-
 96%|█████████▌| 490/512 [55:16<02:27,  6.71s/it]
 96%|█████████▌| 491/512 [55:23<02:21,  6.72s/it]
                                                 
{'loss': '4.807e-05', 'grad_norm': '0.02447', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 14855056, 'tokens/trainable': 224321, 'epoch': '1.918'}
-
 96%|█████████▌| 491/512 [55:23<02:21,  6.72s/it]
 96%|█████████▌| 492/512 [55:30<02:14,  6.73s/it]
                                                 
{'loss': '3.398e-05', 'grad_norm': '0.003633', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.29', 'tokens/total': 14885392, 'tokens/trainable': 224802, 'epoch': '1.922'}
-
 96%|█████████▌| 492/512 [55:30<02:14,  6.73s/it]
 96%|█████████▋| 493/512 [55:37<02:08,  6.74s/it]
                                                 
{'loss': '1.887e-05', 'grad_norm': '0.0009831', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.77', 'tokens/total': 14915760, 'tokens/trainable': 225284, 'epoch': '1.926'}
-
 96%|█████████▋| 493/512 [55:37<02:08,  6.74s/it]
 96%|█████████▋| 494/512 [55:43<02:01,  6.74s/it]
                                                 
{'loss': '3.033e-05', 'grad_norm': '0.002036', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 14946192, 'tokens/trainable': 225733, 'epoch': '1.93'}
-
 96%|█████████▋| 494/512 [55:43<02:01,  6.74s/it]
 97%|█████████▋| 495/512 [55:50<01:54,  6.72s/it]
                                                 
{'loss': '6.913e-06', 'grad_norm': '0.0002904', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 14976272, 'tokens/trainable': 226172, 'epoch': '1.934'}
-
 97%|█████████▋| 495/512 [55:50<01:54,  6.72s/it]
 97%|█████████▋| 496/512 [55:57<01:47,  6.73s/it]
                                                 
{'loss': '0.0007051', 'grad_norm': '0.1167', 'learning_rate': '1.027e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.77', 'tokens/total': 15006544, 'tokens/trainable': 226636, 'epoch': '1.938'}
-
 97%|█████████▋| 496/512 [55:57<01:47,  6.73s/it]
 97%|█████████▋| 497/512 [56:03<01:40,  6.72s/it]
                                                 
{'loss': '1.047e-05', 'grad_norm': '0.0008453', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 15037040, 'tokens/trainable': 227067, 'epoch': '1.941'}
-
 97%|█████████▋| 497/512 [56:03<01:40,  6.72s/it]
 97%|█████████▋| 498/512 [56:10<01:34,  6.72s/it]
                                                 
{'loss': '0.0001397', 'grad_norm': '0.02769', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.58', 'tokens/total': 15067344, 'tokens/trainable': 227532, 'epoch': '1.945'}
-
 97%|█████████▋| 498/512 [56:10<01:34,  6.72s/it]
 97%|█████████▋| 499/512 [56:17<01:27,  6.72s/it]
                                                 
{'loss': '0.001548', 'grad_norm': '0.2443', 'learning_rate': '1.018e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 15097648, 'tokens/trainable': 228007, 'epoch': '1.949'}
-
 97%|█████████▋| 499/512 [56:17<01:27,  6.72s/it]
 98%|█████████▊| 500/512 [56:24<01:20,  6.72s/it]
                                                 
{'loss': '1.88e-05', 'grad_norm': '0.001633', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.84', 'tokens/total': 15128000, 'tokens/trainable': 228457, 'epoch': '1.953'}
-
 98%|█████████▊| 500/512 [56:24<01:20,  6.72s/it]
 98%|█████████▊| 501/512 [56:30<01:13,  6.71s/it]
                                                 
{'loss': '3.361e-05', 'grad_norm': '0.004103', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 15158352, 'tokens/trainable': 228928, 'epoch': '1.957'}
-
 98%|█████████▊| 501/512 [56:30<01:13,  6.71s/it]
 98%|█████████▊| 502/512 [56:37<01:07,  6.74s/it]
                                                 
{'loss': '6.559e-06', 'grad_norm': '0.0003138', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 15189136, 'tokens/trainable': 229368, 'epoch': '1.961'}
-
 98%|█████████▊| 502/512 [56:37<01:07,  6.74s/it]
 98%|█████████▊| 503/512 [56:44<01:00,  6.72s/it]
                                                 
{'loss': '1.713e-05', 'grad_norm': '0.002323', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.75', 'tokens/total': 15219616, 'tokens/trainable': 229815, 'epoch': '1.965'}
-
 98%|█████████▊| 503/512 [56:44<01:00,  6.72s/it]
 98%|█████████▊| 504/512 [56:51<00:53,  6.73s/it]
                                                 
{'loss': '0.002064', 'grad_norm': '0.1925', 'learning_rate': '1.008e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '38.82', 'tokens/total': 15250032, 'tokens/trainable': 230319, 'epoch': '1.969'}
-
 98%|█████████▊| 504/512 [56:51<00:53,  6.73s/it]
 99%|█████████▊| 505/512 [56:57<00:47,  6.73s/it]
                                                 
{'loss': '0.0001955', 'grad_norm': '0.02836', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.42', 'tokens/total': 15280592, 'tokens/trainable': 230807, 'epoch': '1.973'}
-
 99%|█████████▊| 505/512 [56:57<00:47,  6.73s/it]
 99%|█████████▉| 506/512 [57:04<00:40,  6.73s/it]
                                                 
{'loss': '1.307e-05', 'grad_norm': '0.0009315', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 15310816, 'tokens/trainable': 231257, 'epoch': '1.977'}
-
 99%|█████████▉| 506/512 [57:04<00:40,  6.73s/it]
 99%|█████████▉| 507/512 [57:11<00:33,  6.75s/it]
                                                 
{'loss': '0.0003411', 'grad_norm': '0.04385', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.31', 'tokens/total': 15341424, 'tokens/trainable': 231746, 'epoch': '1.98'}
-
 99%|█████████▉| 507/512 [57:11<00:33,  6.75s/it]
 99%|█████████▉| 508/512 [57:17<00:26,  6.58s/it]
                                                 
{'loss': '7.53e-06', 'grad_norm': '0.0003456', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.33', 'memory/max_allocated (GiB)': '33.33', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 15369664, 'tokens/trainable': 232172, 'epoch': '1.984'}
-
 99%|█████████▉| 508/512 [57:17<00:26,  6.58s/it]
 99%|█████████▉| 509/512 [57:24<00:20,  6.71s/it]
                                                 
{'loss': '0.000155', 'grad_norm': '0.03426', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.04', 'tokens/total': 15400496, 'tokens/trainable': 232680, 'epoch': '1.988'}
-
 99%|█████████▉| 509/512 [57:24<00:20,  6.71s/it]
100%|█████████▉| 510/512 [57:31<00:13,  6.73s/it]
                                                 
{'loss': '1.387e-05', 'grad_norm': '0.005425', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.02', 'tokens/total': 15430864, 'tokens/trainable': 233169, 'epoch': '1.992'}
-
100%|█████████▉| 510/512 [57:31<00:13,  6.73s/it]
100%|█████████▉| 511/512 [57:37<00:06,  6.72s/it]
                                                 
{'loss': '0.0001456', 'grad_norm': '0.04471', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 15461232, 'tokens/trainable': 233609, 'epoch': '1.996'}
-
100%|█████████▉| 511/512 [57:37<00:06,  6.72s/it]
100%|██████████| 512/512 [57:44<00:00,  6.76s/it]
                                                 
{'loss': '2.066e-05', 'grad_norm': '0.001178', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.16', 'tokens/total': 15491760, 'tokens/trainable': 234078, 'epoch': '2'}
-
100%|██████████| 512/512 [57:44<00:00,  6.76s/it][2026-08-18 13:37:00,528] [INFO] [axolotl.core.trainers.base._save:828] [PID:2978] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
-
                                                 
{'train_runtime': '3467', 'train_samples_per_second': '4.726', 'train_steps_per_second': '0.148', 'train_loss': '0.006838', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '36.16', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
-
100%|██████████| 512/512 [57:46<00:00,  6.76s/it]
100%|██████████| 512/512 [57:46<00:00,  6.77s/it]
-[2026-08-18 13:37:02,416] [INFO] [axolotl.train.save_trained_model:267] [PID:2978] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
-[2026-08-18 13:37:02,873] [INFO] [axolotl.train.save_trained_model:388] [PID:2978] Model successfully saved to /workspace/wave/training/checkpoints
+[2026-08-18 14:04:52,127] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:10630] after adapters 0.000GB ()
+[2026-08-18 14:04:53,116] [INFO] [axolotl.monkeypatch.lora_kernels.apply_lora_kernel_patches:418] [PID:10630] LoRA kernels: dropout=0.05 enabled
+[2026-08-18 14:04:58,288] [INFO] [axolotl.train.save_initial_configs:450] [PID:10630] Pre-saving adapter config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,289] [INFO] [axolotl.train.save_initial_configs:454] [PID:10630] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,638] [INFO] [axolotl.train.save_initial_configs:459] [PID:10630] Pre-saving model config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,642] [INFO] [axolotl.train.save_initial_configs:463] [PID:10630] Pre-saving processor to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,967] [INFO] [axolotl.train.execute_training:226] [PID:10630] Starting trainer...
+
  0%|          | 0/512 [00:00<?, ?it/s]
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
                                                 
{'loss': '0.1392', 'grad_norm': '1.378', 'learning_rate': '0', 'ppl': '1.149', 'memory/max_active (GiB)': '32.79', 'memory/max_allocated (GiB)': '32.79', 'memory/device_reserved (GiB)': '33.9', 'tokens/train_per_sec_per_gpu': '27.09', 'tokens/total': 30176, 'tokens/trainable': 417, 'epoch': '0.003906'}
+
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
  0%|          | 2/512 [00:14<1:01:59,  7.29s/it]
                                                 
{'loss': '0.151', 'grad_norm': '1.233', 'learning_rate': '4e-06', 'ppl': '1.163', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '34.63', 'tokens/total': 60272, 'tokens/trainable': 873, 'epoch': '0.007812'}
+
  0%|          | 2/512 [00:14<1:01:59,  7.29s/it]
  1%|          | 3/512 [00:21<59:40,  7.03s/it]  
                                               
{'loss': '0.1442', 'grad_norm': '5.272', 'learning_rate': '8e-06', 'ppl': '1.155', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 90512, 'tokens/trainable': 1353, 'epoch': '0.01172'}
+
  1%|          | 3/512 [00:21<59:40,  7.03s/it]
  1%|          | 4/512 [00:28<58:33,  6.92s/it]
                                               
{'loss': '0.1343', 'grad_norm': '1.396', 'learning_rate': '1.2e-05', 'ppl': '1.144', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.73', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 120944, 'tokens/trainable': 1778, 'epoch': '0.01562'}
+
  1%|          | 4/512 [00:28<58:33,  6.92s/it]
  1%|          | 5/512 [00:35<57:44,  6.83s/it]
                                               
{'loss': '0.1441', 'grad_norm': '1.407', 'learning_rate': '1.6e-05', 'ppl': '1.155', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.12', 'tokens/train_per_sec_per_gpu': '36.8', 'tokens/total': 151440, 'tokens/trainable': 2261, 'epoch': '0.01953'}
+
  1%|          | 5/512 [00:35<57:44,  6.83s/it]
  1%|          | 6/512 [00:41<57:14,  6.79s/it]
                                               
{'loss': '0.111', 'grad_norm': '1.676', 'learning_rate': '2e-05', 'ppl': '1.117', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 181984, 'tokens/trainable': 2705, 'epoch': '0.02344'}
+
  1%|          | 6/512 [00:41<57:14,  6.79s/it]
  1%|▏         | 7/512 [00:48<56:53,  6.76s/it]
                                               
{'loss': '0.08474', 'grad_norm': '1.306', 'learning_rate': '2.4e-05', 'ppl': '1.088', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 212336, 'tokens/trainable': 3136, 'epoch': '0.02734'}
+
  1%|▏         | 7/512 [00:48<56:53,  6.76s/it]
  2%|▏         | 8/512 [00:55<56:36,  6.74s/it]
                                               
{'loss': '0.08459', 'grad_norm': '15.48', 'learning_rate': '2.8e-05', 'ppl': '1.088', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 242592, 'tokens/trainable': 3579, 'epoch': '0.03125'}
+
  2%|▏         | 8/512 [00:55<56:36,  6.74s/it]
  2%|▏         | 9/512 [01:01<56:18,  6.72s/it]
                                               
{'loss': '0.06938', 'grad_norm': '1.313', 'learning_rate': '3.2e-05', 'ppl': '1.072', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '29.12', 'tokens/total': 272784, 'tokens/trainable': 4000, 'epoch': '0.03516'}
+
  2%|▏         | 9/512 [01:01<56:18,  6.72s/it]
  2%|▏         | 10/512 [01:08<56:07,  6.71s/it]
                                                
{'loss': '0.04889', 'grad_norm': '1.293', 'learning_rate': '3.6e-05', 'ppl': '1.05', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '37.66', 'tokens/total': 303184, 'tokens/trainable': 4474, 'epoch': '0.03906'}
+
  2%|▏         | 10/512 [01:08<56:07,  6.71s/it]
  2%|▏         | 11/512 [01:15<55:59,  6.71s/it]
                                                
{'loss': '0.03002', 'grad_norm': '1.032', 'learning_rate': '4e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '33.15', 'tokens/total': 333296, 'tokens/trainable': 4942, 'epoch': '0.04297'}
+
  2%|▏         | 11/512 [01:15<55:59,  6.71s/it]
  2%|▏         | 12/512 [01:21<55:54,  6.71s/it]
                                                
{'loss': '0.05269', 'grad_norm': '1.494', 'learning_rate': '4.4e-05', 'ppl': '1.054', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 363840, 'tokens/trainable': 5424, 'epoch': '0.04688'}
+
  2%|▏         | 12/512 [01:21<55:54,  6.71s/it]
  3%|▎         | 13/512 [01:28<55:49,  6.71s/it]
                                                
{'loss': '0.06069', 'grad_norm': '2.024', 'learning_rate': '4.8e-05', 'ppl': '1.063', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 394112, 'tokens/trainable': 5855, 'epoch': '0.05078'}
+
  3%|▎         | 13/512 [01:28<55:49,  6.71s/it]
  3%|▎         | 14/512 [01:35<55:50,  6.73s/it]
                                                
{'loss': '0.03631', 'grad_norm': '1.517', 'learning_rate': '5.2e-05', 'ppl': '1.037', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.75', 'tokens/total': 424656, 'tokens/trainable': 6344, 'epoch': '0.05469'}
+
  3%|▎         | 14/512 [01:35<55:50,  6.73s/it]
  3%|▎         | 15/512 [01:42<55:44,  6.73s/it]
                                                
{'loss': '0.05663', 'grad_norm': '1.741', 'learning_rate': '5.6e-05', 'ppl': '1.058', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 455152, 'tokens/trainable': 6825, 'epoch': '0.05859'}
+
  3%|▎         | 15/512 [01:42<55:44,  6.73s/it]
  3%|▎         | 16/512 [01:48<55:23,  6.70s/it]
                                                
{'loss': '0.07075', 'grad_norm': '2.507', 'learning_rate': '6e-05', 'ppl': '1.073', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '30.03', 'tokens/total': 485328, 'tokens/trainable': 7230, 'epoch': '0.0625'}
+
  3%|▎         | 16/512 [01:48<55:23,  6.70s/it]
  3%|▎         | 17/512 [01:55<55:22,  6.71s/it]
                                                
{'loss': '0.041', 'grad_norm': '1.257', 'learning_rate': '6.4e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.91', 'tokens/total': 515744, 'tokens/trainable': 7699, 'epoch': '0.06641'}
+
  3%|▎         | 17/512 [01:55<55:22,  6.71s/it]
  4%|▎         | 18/512 [02:02<55:25,  6.73s/it]
                                                
{'loss': '0.02119', 'grad_norm': '36.94', 'learning_rate': '6.8e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 546144, 'tokens/trainable': 8184, 'epoch': '0.07031'}
+
  4%|▎         | 18/512 [02:02<55:25,  6.73s/it]
  4%|▎         | 19/512 [02:08<55:19,  6.73s/it]
                                                
{'loss': '0.01386', 'grad_norm': '0.3777', 'learning_rate': '7.2e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 576560, 'tokens/trainable': 8675, 'epoch': '0.07422'}
+
  4%|▎         | 19/512 [02:08<55:19,  6.73s/it]
  4%|▍         | 20/512 [02:15<55:15,  6.74s/it]
                                                
{'loss': '0.0455', 'grad_norm': '1.427', 'learning_rate': '7.6e-05', 'ppl': '1.047', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.72', 'tokens/total': 607056, 'tokens/trainable': 9189, 'epoch': '0.07812'}
+
  4%|▍         | 20/512 [02:15<55:15,  6.74s/it]
  4%|▍         | 21/512 [02:22<55:08,  6.74s/it]
                                                
{'loss': '0.0321', 'grad_norm': '0.5615', 'learning_rate': '8e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.61', 'tokens/total': 637488, 'tokens/trainable': 9614, 'epoch': '0.08203'}
+
  4%|▍         | 21/512 [02:22<55:08,  6.74s/it]
  4%|▍         | 22/512 [02:29<54:56,  6.73s/it]
                                                
{'loss': '0.04279', 'grad_norm': '1.153', 'learning_rate': '8.4e-05', 'ppl': '1.044', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '36.11', 'tokens/total': 667632, 'tokens/trainable': 10086, 'epoch': '0.08594'}
+
  4%|▍         | 22/512 [02:29<54:56,  6.73s/it]
  4%|▍         | 23/512 [02:35<54:49,  6.73s/it]
                                                
{'loss': '0.01939', 'grad_norm': '0.9407', 'learning_rate': '8.8e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 697872, 'tokens/trainable': 10535, 'epoch': '0.08984'}
+
  4%|▍         | 23/512 [02:35<54:49,  6.73s/it]
  5%|▍         | 24/512 [02:42<54:43,  6.73s/it]
                                                
{'loss': '0.01258', 'grad_norm': '0.35', 'learning_rate': '9.2e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 728432, 'tokens/trainable': 10993, 'epoch': '0.09375'}
+
  5%|▍         | 24/512 [02:42<54:43,  6.73s/it]
  5%|▍         | 25/512 [02:48<53:20,  6.57s/it]
                                                
{'loss': '0.04129', 'grad_norm': '1.678', 'learning_rate': '9.6e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.78', 'tokens/total': 756656, 'tokens/trainable': 11417, 'epoch': '0.09766'}
+
  5%|▍         | 25/512 [02:48<53:20,  6.57s/it]
  5%|▌         | 26/512 [02:55<53:29,  6.60s/it]
                                                
{'loss': '0.008186', 'grad_norm': '0.5761', 'learning_rate': '0.0001', 'ppl': '1.008', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 787120, 'tokens/trainable': 11874, 'epoch': '0.1016'}
+
  5%|▌         | 26/512 [02:55<53:29,  6.60s/it]
  5%|▌         | 27/512 [03:02<53:42,  6.64s/it]
                                                
{'loss': '0.01083', 'grad_norm': '5.236', 'learning_rate': '0.0001', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.62', 'tokens/total': 817504, 'tokens/trainable': 12316, 'epoch': '0.1055'}
+
  5%|▌         | 27/512 [03:02<53:42,  6.64s/it]
  5%|▌         | 28/512 [03:09<53:51,  6.68s/it]
                                                
{'loss': '0.05158', 'grad_norm': '1.404', 'learning_rate': '0.0001', 'ppl': '1.053', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 848128, 'tokens/trainable': 12800, 'epoch': '0.1094'}
+
  5%|▌         | 28/512 [03:09<53:51,  6.68s/it]
  6%|▌         | 29/512 [03:15<53:51,  6.69s/it]
                                                
{'loss': '0.05181', 'grad_norm': '3.018', 'learning_rate': '9.999e-05', 'ppl': '1.053', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.43', 'tokens/total': 878448, 'tokens/trainable': 13270, 'epoch': '0.1133'}
+
  6%|▌         | 29/512 [03:15<53:51,  6.69s/it]
  6%|▌         | 30/512 [03:22<53:53,  6.71s/it]
                                                
{'loss': '0.02502', 'grad_norm': '1.414', 'learning_rate': '9.999e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 909008, 'tokens/trainable': 13752, 'epoch': '0.1172'}
+
  6%|▌         | 30/512 [03:22<53:53,  6.71s/it]
  6%|▌         | 31/512 [03:29<53:50,  6.72s/it]
                                                
{'loss': '0.008201', 'grad_norm': '1.161', 'learning_rate': '9.998e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 939312, 'tokens/trainable': 14218, 'epoch': '0.1211'}
+
  6%|▌         | 31/512 [03:29<53:50,  6.72s/it]
  6%|▋         | 32/512 [03:35<53:32,  6.69s/it]
                                                
{'loss': '0.006646', 'grad_norm': '0.2631', 'learning_rate': '9.997e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 969264, 'tokens/trainable': 14655, 'epoch': '0.125'}
+
  6%|▋         | 32/512 [03:35<53:32,  6.69s/it][2026-08-18 14:08:35,358] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
+
  6%|▋         | 33/512 [03:44<58:20,  7.31s/it]
                                                
{'loss': '0.01719', 'grad_norm': '0.7988', 'learning_rate': '9.995e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '33.23', 'tokens/total': 999792, 'tokens/trainable': 15087, 'epoch': '0.1289'}
+
  6%|▋         | 33/512 [03:44<58:20,  7.31s/it]
  7%|▋         | 34/512 [03:51<56:40,  7.11s/it]
                                                
{'loss': '0.02046', 'grad_norm': '0.507', 'learning_rate': '9.994e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.81', 'tokens/total': 1029840, 'tokens/trainable': 15548, 'epoch': '0.1328'}
+
  7%|▋         | 34/512 [03:51<56:40,  7.11s/it]
  7%|▋         | 35/512 [03:58<55:43,  7.01s/it]
                                                
{'loss': '0.02475', 'grad_norm': '0.8837', 'learning_rate': '9.992e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.69', 'tokens/total': 1060416, 'tokens/trainable': 16037, 'epoch': '0.1367'}
+
  7%|▋         | 35/512 [03:58<55:43,  7.01s/it]
  7%|▋         | 36/512 [04:04<54:45,  6.90s/it]
                                                
{'loss': '0.00774', 'grad_norm': '0.7192', 'learning_rate': '9.991e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '31', 'tokens/total': 1090464, 'tokens/trainable': 16475, 'epoch': '0.1406'}
+
  7%|▋         | 36/512 [04:04<54:45,  6.90s/it]
  7%|▋         | 37/512 [04:11<54:05,  6.83s/it]
                                                
{'loss': '0.02497', 'grad_norm': '0.8888', 'learning_rate': '9.989e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 1120944, 'tokens/trainable': 16919, 'epoch': '0.1445'}
+
  7%|▋         | 37/512 [04:11<54:05,  6.83s/it]
  7%|▋         | 38/512 [04:18<53:37,  6.79s/it]
                                                
{'loss': '0.02262', 'grad_norm': '1.009', 'learning_rate': '9.987e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.9', 'tokens/total': 1151360, 'tokens/trainable': 17333, 'epoch': '0.1484'}
+
  7%|▋         | 38/512 [04:18<53:37,  6.79s/it]
  8%|▊         | 39/512 [04:24<53:26,  6.78s/it]
                                                
{'loss': '0.009409', 'grad_norm': '0.5544', 'learning_rate': '9.984e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 1181728, 'tokens/trainable': 17806, 'epoch': '0.1523'}
+
  8%|▊         | 39/512 [04:24<53:26,  6.78s/it]
  8%|▊         | 40/512 [04:31<53:11,  6.76s/it]
                                                
{'loss': '0.01569', 'grad_norm': '0.6828', 'learning_rate': '9.982e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 1212064, 'tokens/trainable': 18290, 'epoch': '0.1562'}
+
  8%|▊         | 40/512 [04:31<53:11,  6.76s/it]
  8%|▊         | 41/512 [04:38<52:56,  6.74s/it]
                                                
{'loss': '0.006071', 'grad_norm': '0.3203', 'learning_rate': '9.979e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.93', 'tokens/total': 1242592, 'tokens/trainable': 18747, 'epoch': '0.1602'}
+
  8%|▊         | 41/512 [04:38<52:56,  6.74s/it]
  8%|▊         | 42/512 [04:44<52:42,  6.73s/it]
                                                
{'loss': '0.0388', 'grad_norm': '1.351', 'learning_rate': '9.976e-05', 'ppl': '1.04', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 1272800, 'tokens/trainable': 19172, 'epoch': '0.1641'}
+
  8%|▊         | 42/512 [04:44<52:42,  6.73s/it]
  8%|▊         | 43/512 [04:51<52:45,  6.75s/it]
                                                
{'loss': '0.01924', 'grad_norm': '0.871', 'learning_rate': '9.973e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.21', 'tokens/total': 1303600, 'tokens/trainable': 19622, 'epoch': '0.168'}
+
  8%|▊         | 43/512 [04:51<52:45,  6.75s/it]
  9%|▊         | 44/512 [04:58<52:31,  6.73s/it]
                                                
{'loss': '0.01226', 'grad_norm': '0.7336', 'learning_rate': '9.97e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 1333936, 'tokens/trainable': 20073, 'epoch': '0.1719'}
+
  9%|▊         | 44/512 [04:58<52:31,  6.73s/it]
  9%|▉         | 45/512 [05:05<52:20,  6.73s/it]
                                                
{'loss': '0.01287', 'grad_norm': '1.955', 'learning_rate': '9.966e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 1364288, 'tokens/trainable': 20497, 'epoch': '0.1758'}
+
  9%|▉         | 45/512 [05:05<52:20,  6.73s/it]
  9%|▉         | 46/512 [05:11<52:10,  6.72s/it]
                                                
{'loss': '0.006207', 'grad_norm': '0.5353', 'learning_rate': '9.963e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '34.21', 'tokens/total': 1394368, 'tokens/trainable': 20940, 'epoch': '0.1797'}
+
  9%|▉         | 46/512 [05:11<52:10,  6.72s/it]
  9%|▉         | 47/512 [05:18<52:12,  6.74s/it]
                                                
{'loss': '0.006637', 'grad_norm': '0.6219', 'learning_rate': '9.959e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 1425024, 'tokens/trainable': 21405, 'epoch': '0.1836'}
+
  9%|▉         | 47/512 [05:18<52:12,  6.74s/it]
  9%|▉         | 48/512 [05:25<51:58,  6.72s/it]
                                                
{'loss': '0.008137', 'grad_norm': '0.7588', 'learning_rate': '9.955e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 1455392, 'tokens/trainable': 21828, 'epoch': '0.1875'}
+
  9%|▉         | 48/512 [05:25<51:58,  6.72s/it]
 10%|▉         | 49/512 [05:31<51:38,  6.69s/it]
                                                
{'loss': '0.0193', 'grad_norm': '0.6957', 'learning_rate': '9.951e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.43', 'tokens/total': 1485616, 'tokens/trainable': 22241, 'epoch': '0.1914'}
+
 10%|▉         | 49/512 [05:31<51:38,  6.69s/it]
 10%|▉         | 50/512 [05:38<51:34,  6.70s/it]
                                                
{'loss': '0.006366', 'grad_norm': '0.7796', 'learning_rate': '9.946e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 1515984, 'tokens/trainable': 22688, 'epoch': '0.1953'}
+
 10%|▉         | 50/512 [05:38<51:34,  6.70s/it]
 10%|▉         | 51/512 [05:45<51:21,  6.68s/it]
                                                
{'loss': '0.002192', 'grad_norm': '0.208', 'learning_rate': '9.942e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 1546224, 'tokens/trainable': 23136, 'epoch': '0.1992'}
+
 10%|▉         | 51/512 [05:45<51:21,  6.68s/it]
 10%|█         | 52/512 [05:51<51:19,  6.69s/it]
                                                
{'loss': '0.01217', 'grad_norm': '0.7914', 'learning_rate': '9.937e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.92', 'tokens/total': 1576528, 'tokens/trainable': 23631, 'epoch': '0.2031'}
+
 10%|█         | 52/512 [05:51<51:19,  6.69s/it]
 10%|█         | 53/512 [05:58<51:17,  6.70s/it]
                                                
{'loss': '0.00895', 'grad_norm': '0.7186', 'learning_rate': '9.932e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 1606768, 'tokens/trainable': 24096, 'epoch': '0.207'}
+
 10%|█         | 53/512 [05:58<51:17,  6.70s/it]
 11%|█         | 54/512 [06:05<51:02,  6.69s/it]
                                                
{'loss': '0.01141', 'grad_norm': '0.8439', 'learning_rate': '9.927e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.77', 'tokens/total': 1636640, 'tokens/trainable': 24569, 'epoch': '0.2109'}
+
 11%|█         | 54/512 [06:05<51:02,  6.69s/it]
 11%|█         | 55/512 [06:12<51:06,  6.71s/it]
                                                
{'loss': '0.004429', 'grad_norm': '0.4795', 'learning_rate': '9.921e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 1667136, 'tokens/trainable': 25067, 'epoch': '0.2148'}
+
 11%|█         | 55/512 [06:12<51:06,  6.71s/it]
 11%|█         | 56/512 [06:18<51:01,  6.71s/it]
                                                
{'loss': '0.003403', 'grad_norm': '0.4258', 'learning_rate': '9.916e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.47', 'tokens/total': 1697472, 'tokens/trainable': 25541, 'epoch': '0.2188'}
+
 11%|█         | 56/512 [06:18<51:01,  6.71s/it]
 11%|█         | 57/512 [06:25<50:54,  6.71s/it]
                                                
{'loss': '0.01049', 'grad_norm': '0.7323', 'learning_rate': '9.91e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 1727680, 'tokens/trainable': 25965, 'epoch': '0.2227'}
+
 11%|█         | 57/512 [06:25<50:54,  6.71s/it]
 11%|█▏        | 58/512 [06:32<50:44,  6.71s/it]
                                                
{'loss': '0.004758', 'grad_norm': '2.37', 'learning_rate': '9.904e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.79', 'tokens/total': 1758192, 'tokens/trainable': 26409, 'epoch': '0.2266'}
+
 11%|█▏        | 58/512 [06:32<50:44,  6.71s/it]
 12%|█▏        | 59/512 [06:38<50:42,  6.72s/it]
                                                
{'loss': '0.002968', 'grad_norm': '1.733', 'learning_rate': '9.898e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.99', 'tokens/total': 1788816, 'tokens/trainable': 26883, 'epoch': '0.2305'}
+
 12%|█▏        | 59/512 [06:38<50:42,  6.72s/it]
 12%|█▏        | 60/512 [06:45<50:35,  6.72s/it]
                                                
{'loss': '0.005325', 'grad_norm': '4.904', 'learning_rate': '9.892e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 1819056, 'tokens/trainable': 27323, 'epoch': '0.2344'}
+
 12%|█▏        | 60/512 [06:45<50:35,  6.72s/it]
 12%|█▏        | 61/512 [06:52<50:30,  6.72s/it]
                                                
{'loss': '0.0007892', 'grad_norm': '0.08419', 'learning_rate': '9.886e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '39.09', 'tokens/total': 1849664, 'tokens/trainable': 27830, 'epoch': '0.2383'}
+
 12%|█▏        | 61/512 [06:52<50:30,  6.72s/it]
 12%|█▏        | 62/512 [06:59<50:39,  6.75s/it]
                                                
{'loss': '0.01735', 'grad_norm': '0.7356', 'learning_rate': '9.879e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 1880272, 'tokens/trainable': 28281, 'epoch': '0.2422'}
+
 12%|█▏        | 62/512 [06:59<50:39,  6.75s/it]
 12%|█▏        | 63/512 [07:05<50:27,  6.74s/it]
                                                
{'loss': '0.01441', 'grad_norm': '0.9251', 'learning_rate': '9.872e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.16', 'tokens/total': 1910752, 'tokens/trainable': 28742, 'epoch': '0.2461'}
+
 12%|█▏        | 63/512 [07:05<50:27,  6.74s/it]
 12%|█▎        | 64/512 [07:12<50:08,  6.72s/it]
                                                
{'loss': '0.002559', 'grad_norm': '0.2563', 'learning_rate': '9.865e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.54', 'tokens/total': 1940848, 'tokens/trainable': 29183, 'epoch': '0.25'}
+
 12%|█▎        | 64/512 [07:12<50:08,  6.72s/it][2026-08-18 14:12:12,138] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
+
 13%|█▎        | 65/512 [07:21<54:03,  7.26s/it]
                                                
{'loss': '0.007829', 'grad_norm': '0.5092', 'learning_rate': '9.858e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 1971344, 'tokens/trainable': 29648, 'epoch': '0.2539'}
+
 13%|█▎        | 65/512 [07:21<54:03,  7.26s/it]
 13%|█▎        | 66/512 [07:27<52:37,  7.08s/it]
                                                
{'loss': '0.001763', 'grad_norm': '0.5972', 'learning_rate': '9.851e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.06', 'tokens/train_per_sec_per_gpu': '33.44', 'tokens/total': 2001712, 'tokens/trainable': 30075, 'epoch': '0.2578'}
+
 13%|█▎        | 66/512 [07:27<52:37,  7.08s/it]
 13%|█▎        | 67/512 [07:34<50:38,  6.83s/it]
                                                
{'loss': '0.005683', 'grad_norm': '0.6683', 'learning_rate': '9.844e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.53', 'tokens/total': 2029936, 'tokens/trainable': 30546, 'epoch': '0.2617'}
+
 13%|█▎        | 67/512 [07:34<50:38,  6.83s/it]
 13%|█▎        | 68/512 [07:40<50:18,  6.80s/it]
                                                
{'loss': '0.01054', 'grad_norm': '2.3', 'learning_rate': '9.836e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 2060384, 'tokens/trainable': 31025, 'epoch': '0.2656'}
+
 13%|█▎        | 68/512 [07:40<50:18,  6.80s/it]
 13%|█▎        | 69/512 [07:47<50:03,  6.78s/it]
                                                
{'loss': '0.0167', 'grad_norm': '2.232', 'learning_rate': '9.828e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 2090688, 'tokens/trainable': 31505, 'epoch': '0.2695'}
+
 13%|█▎        | 69/512 [07:47<50:03,  6.78s/it]
 14%|█▎        | 70/512 [07:54<49:42,  6.75s/it]
                                                
{'loss': '0.005516', 'grad_norm': '0.455', 'learning_rate': '9.82e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 2120848, 'tokens/trainable': 31956, 'epoch': '0.2734'}
+
 14%|█▎        | 70/512 [07:54<49:42,  6.75s/it]
 14%|█▍        | 71/512 [08:00<48:21,  6.58s/it]
                                                
{'loss': '0.008853', 'grad_norm': '0.7483', 'learning_rate': '9.812e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.35', 'memory/max_allocated (GiB)': '33.35', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 2149120, 'tokens/trainable': 32374, 'epoch': '0.2773'}
+
 14%|█▍        | 71/512 [08:00<48:21,  6.58s/it]
 14%|█▍        | 72/512 [08:07<48:31,  6.62s/it]
                                                
{'loss': '0.001421', 'grad_norm': '0.4359', 'learning_rate': '9.803e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.46', 'tokens/total': 2179472, 'tokens/trainable': 32833, 'epoch': '0.2812'}
+
 14%|█▍        | 72/512 [08:07<48:31,  6.62s/it]
 14%|█▍        | 73/512 [08:13<48:32,  6.63s/it]
                                                
{'loss': '0.005964', 'grad_norm': '0.3272', 'learning_rate': '9.795e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.67', 'tokens/total': 2209984, 'tokens/trainable': 33304, 'epoch': '0.2852'}
+
 14%|█▍        | 73/512 [08:13<48:32,  6.63s/it]
 14%|█▍        | 74/512 [08:20<48:48,  6.69s/it]
                                                
{'loss': '0.005361', 'grad_norm': '0.5767', 'learning_rate': '9.786e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.33', 'tokens/total': 2240512, 'tokens/trainable': 33771, 'epoch': '0.2891'}
+
 14%|█▍        | 74/512 [08:20<48:48,  6.69s/it]
 15%|█▍        | 75/512 [08:27<48:40,  6.68s/it]
                                                
{'loss': '0.006908', 'grad_norm': '0.9308', 'learning_rate': '9.777e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.46', 'tokens/total': 2270608, 'tokens/trainable': 34186, 'epoch': '0.293'}
+
 15%|█▍        | 75/512 [08:27<48:40,  6.68s/it]
 15%|█▍        | 76/512 [08:33<48:42,  6.70s/it]
                                                
{'loss': '0.001703', 'grad_norm': '0.5172', 'learning_rate': '9.768e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.83', 'tokens/total': 2300864, 'tokens/trainable': 34667, 'epoch': '0.2969'}
+
 15%|█▍        | 76/512 [08:33<48:42,  6.70s/it]
 15%|█▌        | 77/512 [08:40<48:29,  6.69s/it]
                                                
{'loss': '0.01024', 'grad_norm': '4.432', 'learning_rate': '9.759e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 2330960, 'tokens/trainable': 35078, 'epoch': '0.3008'}
+
 15%|█▌        | 77/512 [08:40<48:29,  6.69s/it]
 15%|█▌        | 78/512 [08:47<48:35,  6.72s/it]
                                                
{'loss': '0.01689', 'grad_norm': '1.699', 'learning_rate': '9.749e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.8', 'tokens/total': 2361584, 'tokens/trainable': 35550, 'epoch': '0.3047'}
+
 15%|█▌        | 78/512 [08:47<48:35,  6.72s/it]
 15%|█▌        | 79/512 [08:54<48:24,  6.71s/it]
                                                
{'loss': '0.005958', 'grad_norm': '1.176', 'learning_rate': '9.74e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.46', 'tokens/total': 2391936, 'tokens/trainable': 35977, 'epoch': '0.3086'}
+
 15%|█▌        | 79/512 [08:54<48:24,  6.71s/it]
 16%|█▌        | 80/512 [09:00<48:20,  6.72s/it]
                                                
{'loss': '0.03401', 'grad_norm': '2.761', 'learning_rate': '9.73e-05', 'ppl': '1.035', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.2', 'tokens/total': 2422432, 'tokens/trainable': 36462, 'epoch': '0.3125'}
+
 16%|█▌        | 80/512 [09:00<48:20,  6.72s/it]
 16%|█▌        | 81/512 [09:07<48:11,  6.71s/it]
                                                
{'loss': '0.0001726', 'grad_norm': '0.0173', 'learning_rate': '9.72e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 2452752, 'tokens/trainable': 36919, 'epoch': '0.3164'}
+
 16%|█▌        | 81/512 [09:07<48:11,  6.71s/it]
 16%|█▌        | 82/512 [09:13<47:02,  6.56s/it]
                                                
{'loss': '0.008328', 'grad_norm': '0.6305', 'learning_rate': '9.71e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.4', 'tokens/total': 2481216, 'tokens/trainable': 37379, 'epoch': '0.3203'}
+
 16%|█▌        | 82/512 [09:13<47:02,  6.56s/it]
 16%|█▌        | 83/512 [09:20<47:16,  6.61s/it]
                                                
{'loss': '0.0007961', 'grad_norm': '0.08372', 'learning_rate': '9.699e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.05', 'tokens/total': 2511488, 'tokens/trainable': 37832, 'epoch': '0.3242'}
+
 16%|█▌        | 83/512 [09:20<47:16,  6.61s/it]
 16%|█▋        | 84/512 [09:27<47:25,  6.65s/it]
                                                
{'loss': '0.001538', 'grad_norm': '0.4056', 'learning_rate': '9.689e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 2541984, 'tokens/trainable': 38303, 'epoch': '0.3281'}
+
 16%|█▋        | 84/512 [09:27<47:25,  6.65s/it]
 17%|█▋        | 85/512 [09:33<47:24,  6.66s/it]
                                                
{'loss': '0.0007728', 'grad_norm': '0.1657', 'learning_rate': '9.678e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.62', 'tokens/total': 2572224, 'tokens/trainable': 38758, 'epoch': '0.332'}
+
 17%|█▋        | 85/512 [09:33<47:24,  6.66s/it]
 17%|█▋        | 86/512 [09:40<47:22,  6.67s/it]
                                                
{'loss': '0.008873', 'grad_norm': '1.778', 'learning_rate': '9.667e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.62', 'tokens/total': 2602704, 'tokens/trainable': 39179, 'epoch': '0.3359'}
+
 17%|█▋        | 86/512 [09:40<47:22,  6.67s/it]
 17%|█▋        | 87/512 [09:47<47:21,  6.69s/it]
                                                
{'loss': '0.01371', 'grad_norm': '0.6024', 'learning_rate': '9.656e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 2633072, 'tokens/trainable': 39679, 'epoch': '0.3398'}
+
 17%|█▋        | 87/512 [09:47<47:21,  6.69s/it]
 17%|█▋        | 88/512 [09:53<47:06,  6.67s/it]
                                                
{'loss': '0.02105', 'grad_norm': '3.774', 'learning_rate': '9.645e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 2663232, 'tokens/trainable': 40118, 'epoch': '0.3438'}
+
 17%|█▋        | 88/512 [09:53<47:06,  6.67s/it]
 17%|█▋        | 89/512 [10:00<47:02,  6.67s/it]
                                                
{'loss': '0.002541', 'grad_norm': '0.209', 'learning_rate': '9.633e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 2693696, 'tokens/trainable': 40571, 'epoch': '0.3477'}
+
 17%|█▋        | 89/512 [10:00<47:02,  6.67s/it]
 18%|█▊        | 90/512 [10:07<46:59,  6.68s/it]
                                                
{'loss': '0.003635', 'grad_norm': '0.3318', 'learning_rate': '9.622e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.75', 'tokens/total': 2723984, 'tokens/trainable': 40997, 'epoch': '0.3516'}
+
 18%|█▊        | 90/512 [10:07<46:59,  6.68s/it]
 18%|█▊        | 91/512 [10:14<46:57,  6.69s/it]
                                                
{'loss': '0.007066', 'grad_norm': '0.4436', 'learning_rate': '9.61e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 2754240, 'tokens/trainable': 41462, 'epoch': '0.3555'}
+
 18%|█▊        | 91/512 [10:14<46:57,  6.69s/it]
 18%|█▊        | 92/512 [10:20<46:50,  6.69s/it]
                                                
{'loss': '0.01106', 'grad_norm': '0.8875', 'learning_rate': '9.598e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.79', 'tokens/total': 2784672, 'tokens/trainable': 41927, 'epoch': '0.3594'}
+
 18%|█▊        | 92/512 [10:20<46:50,  6.69s/it]
 18%|█▊        | 93/512 [10:27<46:47,  6.70s/it]
                                                
{'loss': '0.02101', 'grad_norm': '1.71', 'learning_rate': '9.586e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.84', 'tokens/total': 2815120, 'tokens/trainable': 42397, 'epoch': '0.3633'}
+
 18%|█▊        | 93/512 [10:27<46:47,  6.70s/it]
 18%|█▊        | 94/512 [10:34<46:44,  6.71s/it]
                                                
{'loss': '0.0007664', 'grad_norm': '0.05346', 'learning_rate': '9.574e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 2845600, 'tokens/trainable': 42883, 'epoch': '0.3672'}
+
 18%|█▊        | 94/512 [10:34<46:44,  6.71s/it]
 19%|█▊        | 95/512 [10:40<46:43,  6.72s/it]
                                                
{'loss': '0.001947', 'grad_norm': '0.3566', 'learning_rate': '9.562e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.67', 'tokens/total': 2875952, 'tokens/trainable': 43381, 'epoch': '0.3711'}
+
 19%|█▊        | 95/512 [10:40<46:43,  6.72s/it]
 19%|█▉        | 96/512 [10:47<46:37,  6.72s/it]
                                                
{'loss': '0.01163', 'grad_norm': '0.6131', 'learning_rate': '9.549e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 2906288, 'tokens/trainable': 43824, 'epoch': '0.375'}
+
 19%|█▉        | 96/512 [10:47<46:37,  6.72s/it][2026-08-18 14:15:47,190] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
+
 19%|█▉        | 97/512 [10:56<50:07,  7.25s/it]
                                                
{'loss': '0.004611', 'grad_norm': '0.5786', 'learning_rate': '9.536e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 2936752, 'tokens/trainable': 44255, 'epoch': '0.3789'}
+
 19%|█▉        | 97/512 [10:56<50:07,  7.25s/it]
 19%|█▉        | 98/512 [11:02<48:52,  7.08s/it]
                                                
{'loss': '0.003711', 'grad_norm': '0.4322', 'learning_rate': '9.523e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.19', 'tokens/train_per_sec_per_gpu': '37.74', 'tokens/total': 2967072, 'tokens/trainable': 44739, 'epoch': '0.3828'}
+
 19%|█▉        | 98/512 [11:02<48:52,  7.08s/it]
 19%|█▉        | 99/512 [11:09<48:10,  7.00s/it]
                                                
{'loss': '0.002609', 'grad_norm': '0.1657', 'learning_rate': '9.51e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '35.76', 'tokens/train_per_sec_per_gpu': '36.2', 'tokens/total': 2997664, 'tokens/trainable': 45202, 'epoch': '0.3867'}
+
 19%|█▉        | 99/512 [11:09<48:10,  7.00s/it]
 20%|█▉        | 100/512 [11:16<47:24,  6.91s/it]
                                                 
{'loss': '0.005652', 'grad_norm': '0.4386', 'learning_rate': '9.497e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.05', 'tokens/total': 3028032, 'tokens/trainable': 45646, 'epoch': '0.3906'}
+
 20%|█▉        | 100/512 [11:16<47:24,  6.91s/it]
 20%|█▉        | 101/512 [11:23<46:54,  6.85s/it]
                                                 
{'loss': '0.001526', 'grad_norm': '0.1223', 'learning_rate': '9.484e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 3058272, 'tokens/trainable': 46089, 'epoch': '0.3945'}
+
 20%|█▉        | 101/512 [11:23<46:54,  6.85s/it]
 20%|█▉        | 102/512 [11:29<46:33,  6.81s/it]
                                                 
{'loss': '0.007584', 'grad_norm': '0.7791', 'learning_rate': '9.47e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 3088848, 'tokens/trainable': 46547, 'epoch': '0.3984'}
+
 20%|█▉        | 102/512 [11:29<46:33,  6.81s/it]
 20%|██        | 103/512 [11:36<46:12,  6.78s/it]
                                                 
{'loss': '0.002193', 'grad_norm': '0.1951', 'learning_rate': '9.456e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '31.96', 'tokens/total': 3119424, 'tokens/trainable': 47014, 'epoch': '0.4023'}
+
 20%|██        | 103/512 [11:36<46:12,  6.78s/it]
 20%|██        | 104/512 [11:43<45:57,  6.76s/it]
                                                 
{'loss': '0.007656', 'grad_norm': '0.3494', 'learning_rate': '9.442e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 3149696, 'tokens/trainable': 47503, 'epoch': '0.4062'}
+
 20%|██        | 104/512 [11:43<45:57,  6.76s/it]
 21%|██        | 105/512 [11:49<45:46,  6.75s/it]
                                                 
{'loss': '0.001468', 'grad_norm': '0.3206', 'learning_rate': '9.428e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.27', 'tokens/total': 3180064, 'tokens/trainable': 47996, 'epoch': '0.4102'}
+
 21%|██        | 105/512 [11:49<45:46,  6.75s/it]
 21%|██        | 106/512 [11:56<44:32,  6.58s/it]
                                                 
{'loss': '0.00197', 'grad_norm': '0.232', 'learning_rate': '9.414e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 3208320, 'tokens/trainable': 48447, 'epoch': '0.4141'}
+
 21%|██        | 106/512 [11:56<44:32,  6.58s/it]
 21%|██        | 107/512 [12:02<44:38,  6.61s/it]
                                                 
{'loss': '0.000285', 'grad_norm': '0.05074', 'learning_rate': '9.4e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.69', 'tokens/total': 3238416, 'tokens/trainable': 48889, 'epoch': '0.418'}
+
 21%|██        | 107/512 [12:02<44:38,  6.61s/it]
 21%|██        | 108/512 [12:09<44:42,  6.64s/it]
                                                 
{'loss': '0.002674', 'grad_norm': '0.6825', 'learning_rate': '9.385e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.89', 'tokens/total': 3268784, 'tokens/trainable': 49360, 'epoch': '0.4219'}
+
 21%|██        | 108/512 [12:09<44:42,  6.64s/it]
 21%|██▏       | 109/512 [12:16<44:45,  6.66s/it]
                                                 
{'loss': '0.001428', 'grad_norm': '0.2664', 'learning_rate': '9.37e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.78', 'tokens/total': 3298912, 'tokens/trainable': 49827, 'epoch': '0.4258'}
+
 21%|██▏       | 109/512 [12:16<44:45,  6.66s/it]
 21%|██▏       | 110/512 [12:22<44:43,  6.68s/it]
                                                 
{'loss': '0.0007402', 'grad_norm': '0.1222', 'learning_rate': '9.355e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 3329232, 'tokens/trainable': 50267, 'epoch': '0.4297'}
+
 21%|██▏       | 110/512 [12:22<44:43,  6.68s/it]
 22%|██▏       | 111/512 [12:29<44:40,  6.69s/it]
                                                 
{'loss': '0.02026', 'grad_norm': '1.222', 'learning_rate': '9.34e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 3359440, 'tokens/trainable': 50711, 'epoch': '0.4336'}
+
 22%|██▏       | 111/512 [12:29<44:40,  6.69s/it]
 22%|██▏       | 112/512 [12:36<44:33,  6.68s/it]
                                                 
{'loss': '0.01277', 'grad_norm': '1.257', 'learning_rate': '9.325e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.59', 'tokens/total': 3389472, 'tokens/trainable': 51157, 'epoch': '0.4375'}
+
 22%|██▏       | 112/512 [12:36<44:33,  6.68s/it]
 22%|██▏       | 113/512 [12:43<44:32,  6.70s/it]
                                                 
{'loss': '0.00542', 'grad_norm': '0.8144', 'learning_rate': '9.31e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 3419920, 'tokens/trainable': 51613, 'epoch': '0.4414'}
+
 22%|██▏       | 113/512 [12:43<44:32,  6.70s/it]
 22%|██▏       | 114/512 [12:49<44:22,  6.69s/it]
                                                 
{'loss': '0.004596', 'grad_norm': '1.351', 'learning_rate': '9.294e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '29.1', 'tokens/total': 3450176, 'tokens/trainable': 52017, 'epoch': '0.4453'}
+
 22%|██▏       | 114/512 [12:49<44:22,  6.69s/it]
 22%|██▏       | 115/512 [12:56<44:18,  6.70s/it]
                                                 
{'loss': '0.005163', 'grad_norm': '2.766', 'learning_rate': '9.278e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 3480544, 'tokens/trainable': 52506, 'epoch': '0.4492'}
+
 22%|██▏       | 115/512 [12:56<44:18,  6.70s/it]
 23%|██▎       | 116/512 [13:03<44:15,  6.71s/it]
                                                 
{'loss': '0.02482', 'grad_norm': '1.783', 'learning_rate': '9.263e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '38.63', 'tokens/total': 3510848, 'tokens/trainable': 53008, 'epoch': '0.4531'}
+
 23%|██▎       | 116/512 [13:03<44:15,  6.71s/it]
 23%|██▎       | 117/512 [13:09<44:08,  6.71s/it]
                                                 
{'loss': '0.01225', 'grad_norm': '0.868', 'learning_rate': '9.247e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.36', 'tokens/total': 3541296, 'tokens/trainable': 53467, 'epoch': '0.457'}
+
 23%|██▎       | 117/512 [13:09<44:08,  6.71s/it]
 23%|██▎       | 118/512 [13:16<44:06,  6.72s/it]
                                                 
{'loss': '0.01154', 'grad_norm': '0.8031', 'learning_rate': '9.23e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 3571936, 'tokens/trainable': 53923, 'epoch': '0.4609'}
+
 23%|██▎       | 118/512 [13:16<44:06,  6.72s/it]
 23%|██▎       | 119/512 [13:23<43:53,  6.70s/it]
                                                 
{'loss': '0.004753', 'grad_norm': '1.132', 'learning_rate': '9.214e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 3602000, 'tokens/trainable': 54363, 'epoch': '0.4648'}
+
 23%|██▎       | 119/512 [13:23<43:53,  6.70s/it]
 23%|██▎       | 120/512 [13:29<43:47,  6.70s/it]
                                                 
{'loss': '0.006918', 'grad_norm': '0.2578', 'learning_rate': '9.198e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 3632288, 'tokens/trainable': 54839, 'epoch': '0.4688'}
+
 23%|██▎       | 120/512 [13:29<43:47,  6.70s/it]
 24%|██▎       | 121/512 [13:36<43:39,  6.70s/it]
                                                 
{'loss': '0.003779', 'grad_norm': '0.3194', 'learning_rate': '9.181e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.59', 'tokens/total': 3662608, 'tokens/trainable': 55267, 'epoch': '0.4727'}
+
 24%|██▎       | 121/512 [13:36<43:39,  6.70s/it]
 24%|██▍       | 122/512 [13:43<43:31,  6.70s/it]
                                                 
{'loss': '0.004217', 'grad_norm': '0.3439', 'learning_rate': '9.164e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.52', 'tokens/total': 3692864, 'tokens/trainable': 55700, 'epoch': '0.4766'}
+
 24%|██▍       | 122/512 [13:43<43:31,  6.70s/it]
 24%|██▍       | 123/512 [13:49<43:19,  6.68s/it]
                                                 
{'loss': '0.003874', 'grad_norm': '0.4916', 'learning_rate': '9.147e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.39', 'tokens/total': 3722864, 'tokens/trainable': 56144, 'epoch': '0.4805'}
+
 24%|██▍       | 123/512 [13:49<43:19,  6.68s/it]
 24%|██▍       | 124/512 [13:56<43:11,  6.68s/it]
                                                 
{'loss': '0.00401', 'grad_norm': '0.3992', 'learning_rate': '9.13e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 3753056, 'tokens/trainable': 56589, 'epoch': '0.4844'}
+
 24%|██▍       | 124/512 [13:56<43:11,  6.68s/it]
 24%|██▍       | 125/512 [14:03<43:11,  6.70s/it]
                                                 
{'loss': '0.005606', 'grad_norm': '0.4556', 'learning_rate': '9.113e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 3783312, 'tokens/trainable': 57081, 'epoch': '0.4883'}
+
 24%|██▍       | 125/512 [14:03<43:11,  6.70s/it]
 25%|██▍       | 126/512 [14:10<43:01,  6.69s/it]
                                                 
{'loss': '0.0005784', 'grad_norm': '0.04188', 'learning_rate': '9.096e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 3813712, 'tokens/trainable': 57527, 'epoch': '0.4922'}
+
 25%|██▍       | 126/512 [14:10<43:01,  6.69s/it]
 25%|██▍       | 127/512 [14:16<43:01,  6.71s/it]
                                                 
{'loss': '0.002429', 'grad_norm': '0.2182', 'learning_rate': '9.078e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.84', 'tokens/total': 3844336, 'tokens/trainable': 58005, 'epoch': '0.4961'}
+
 25%|██▍       | 127/512 [14:16<43:01,  6.71s/it]
 25%|██▌       | 128/512 [14:23<41:55,  6.55s/it]
                                                 
{'loss': '0.001604', 'grad_norm': '0.4637', 'learning_rate': '9.06e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 3872688, 'tokens/trainable': 58454, 'epoch': '0.5'}
+
 25%|██▌       | 128/512 [14:23<41:55,  6.55s/it][2026-08-18 14:19:22,518] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
+
 25%|██▌       | 129/512 [14:31<45:18,  7.10s/it]
                                                 
{'loss': '0.0002776', 'grad_norm': '0.03312', 'learning_rate': '9.043e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.97', 'tokens/total': 3902640, 'tokens/trainable': 58896, 'epoch': '0.5039'}
+
 25%|██▌       | 129/512 [14:31<45:18,  7.10s/it]
 25%|██▌       | 130/512 [14:38<44:26,  6.98s/it]
                                                 
{'loss': '8.183e-05', 'grad_norm': '0.0194', 'learning_rate': '9.025e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '35.8', 'tokens/total': 3933104, 'tokens/trainable': 59356, 'epoch': '0.5078'}
+
 25%|██▌       | 130/512 [14:38<44:26,  6.98s/it]
 26%|██▌       | 131/512 [14:44<43:48,  6.90s/it]
                                                 
{'loss': '0.005079', 'grad_norm': '0.3952', 'learning_rate': '9.007e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 3963184, 'tokens/trainable': 59792, 'epoch': '0.5117'}
+
 26%|██▌       | 131/512 [14:44<43:48,  6.90s/it]
 26%|██▌       | 132/512 [14:51<43:18,  6.84s/it]
                                                 
{'loss': '0.0001526', 'grad_norm': '0.03258', 'learning_rate': '8.988e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.68', 'tokens/total': 3993648, 'tokens/trainable': 60266, 'epoch': '0.5156'}
+
 26%|██▌       | 132/512 [14:51<43:18,  6.84s/it]
 26%|██▌       | 133/512 [14:58<42:58,  6.80s/it]
                                                 
{'loss': '0.0003361', 'grad_norm': '0.1087', 'learning_rate': '8.97e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.44', 'tokens/total': 4024000, 'tokens/trainable': 60759, 'epoch': '0.5195'}
+
 26%|██▌       | 133/512 [14:58<42:58,  6.80s/it]
 26%|██▌       | 134/512 [15:04<41:43,  6.62s/it]
                                                 
{'loss': '0.0001435', 'grad_norm': '0.01607', 'learning_rate': '8.951e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '31.76', 'tokens/total': 4052480, 'tokens/trainable': 61204, 'epoch': '0.5234'}
+
 26%|██▌       | 134/512 [15:04<41:43,  6.62s/it]
 26%|██▋       | 135/512 [15:11<41:49,  6.66s/it]
                                                 
{'loss': '0.003308', 'grad_norm': '0.5635', 'learning_rate': '8.933e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.93', 'tokens/total': 4082864, 'tokens/trainable': 61684, 'epoch': '0.5273'}
+
 26%|██▋       | 135/512 [15:11<41:49,  6.66s/it]
 27%|██▋       | 136/512 [15:17<41:41,  6.65s/it]
                                                 
{'loss': '0.0003551', 'grad_norm': '0.1293', 'learning_rate': '8.914e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '38.22', 'tokens/total': 4113040, 'tokens/trainable': 62159, 'epoch': '0.5312'}
+
 27%|██▋       | 136/512 [15:17<41:41,  6.65s/it]
 27%|██▋       | 137/512 [15:24<41:42,  6.67s/it]
                                                 
{'loss': '0.0002933', 'grad_norm': '0.03812', 'learning_rate': '8.895e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.28', 'tokens/total': 4143344, 'tokens/trainable': 62649, 'epoch': '0.5352'}
+
 27%|██▋       | 137/512 [15:24<41:42,  6.67s/it]
 27%|██▋       | 138/512 [15:31<41:37,  6.68s/it]
                                                 
{'loss': '0.009558', 'grad_norm': '0.4708', 'learning_rate': '8.876e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.65', 'tokens/total': 4173648, 'tokens/trainable': 63100, 'epoch': '0.5391'}
+
 27%|██▋       | 138/512 [15:31<41:37,  6.68s/it]
 27%|██▋       | 139/512 [15:37<41:37,  6.70s/it]
                                                 
{'loss': '0.006467', 'grad_norm': '0.6402', 'learning_rate': '8.856e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.56', 'tokens/total': 4204080, 'tokens/trainable': 63588, 'epoch': '0.543'}
+
 27%|██▋       | 139/512 [15:37<41:37,  6.70s/it]
 27%|██▋       | 140/512 [15:44<40:29,  6.53s/it]
                                                 
{'loss': '0.01086', 'grad_norm': '0.9977', 'learning_rate': '8.837e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.42', 'memory/max_allocated (GiB)': '33.42', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 4232400, 'tokens/trainable': 64005, 'epoch': '0.5469'}
+
 27%|██▋       | 140/512 [15:44<40:29,  6.53s/it]
 28%|██▊       | 141/512 [15:50<40:43,  6.59s/it]
                                                 
{'loss': '0.002833', 'grad_norm': '0.931', 'learning_rate': '8.817e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.12', 'tokens/total': 4262624, 'tokens/trainable': 64462, 'epoch': '0.5508'}
+
 28%|██▊       | 141/512 [15:50<40:43,  6.59s/it]
 28%|██▊       | 142/512 [15:57<40:52,  6.63s/it]
                                                 
{'loss': '0.00226', 'grad_norm': '0.8449', 'learning_rate': '8.798e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 4293168, 'tokens/trainable': 64900, 'epoch': '0.5547'}
+
 28%|██▊       | 142/512 [15:57<40:52,  6.63s/it]
 28%|██▊       | 143/512 [16:04<40:52,  6.65s/it]
                                                 
{'loss': '0.001097', 'grad_norm': '0.1501', 'learning_rate': '8.778e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.52', 'tokens/total': 4323248, 'tokens/trainable': 65379, 'epoch': '0.5586'}
+
 28%|██▊       | 143/512 [16:04<40:52,  6.65s/it]
 28%|██▊       | 144/512 [16:10<40:49,  6.66s/it]
                                                 
{'loss': '0.02481', 'grad_norm': '0.7081', 'learning_rate': '8.758e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 4353344, 'tokens/trainable': 65845, 'epoch': '0.5625'}
+
 28%|██▊       | 144/512 [16:10<40:49,  6.66s/it]
 28%|██▊       | 145/512 [16:17<40:46,  6.67s/it]
                                                 
{'loss': '0.01163', 'grad_norm': '0.8303', 'learning_rate': '8.738e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 4383584, 'tokens/trainable': 66285, 'epoch': '0.5664'}
+
 28%|██▊       | 145/512 [16:17<40:46,  6.67s/it]
 29%|██▊       | 146/512 [16:24<40:46,  6.68s/it]
                                                 
{'loss': '0.006236', 'grad_norm': '0.4801', 'learning_rate': '8.718e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 4413856, 'tokens/trainable': 66736, 'epoch': '0.5703'}
+
 29%|██▊       | 146/512 [16:24<40:46,  6.68s/it]
 29%|██▊       | 147/512 [16:31<40:40,  6.69s/it]
                                                 
{'loss': '0.004551', 'grad_norm': '0.2672', 'learning_rate': '8.697e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 4444112, 'tokens/trainable': 67150, 'epoch': '0.5742'}
+
 29%|██▊       | 147/512 [16:31<40:40,  6.69s/it]
 29%|██▉       | 148/512 [16:37<40:41,  6.71s/it]
                                                 
{'loss': '0.005152', 'grad_norm': '0.2334', 'learning_rate': '8.677e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 4474464, 'tokens/trainable': 67633, 'epoch': '0.5781'}
+
 29%|██▉       | 148/512 [16:37<40:41,  6.71s/it]
 29%|██▉       | 149/512 [16:44<40:35,  6.71s/it]
                                                 
{'loss': '0.01038', 'grad_norm': '1.36', 'learning_rate': '8.656e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 4504880, 'tokens/trainable': 68077, 'epoch': '0.582'}
+
 29%|██▉       | 149/512 [16:44<40:35,  6.71s/it]
 29%|██▉       | 150/512 [16:51<40:27,  6.70s/it]
                                                 
{'loss': '0.001625', 'grad_norm': '0.2275', 'learning_rate': '8.635e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 4534976, 'tokens/trainable': 68545, 'epoch': '0.5859'}
+
 29%|██▉       | 150/512 [16:51<40:27,  6.70s/it]
 29%|██▉       | 151/512 [16:57<40:21,  6.71s/it]
                                                 
{'loss': '0.02714', 'grad_norm': '0.6304', 'learning_rate': '8.615e-05', 'ppl': '1.028', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.26', 'tokens/total': 4565088, 'tokens/trainable': 69011, 'epoch': '0.5898'}
+
 29%|██▉       | 151/512 [16:57<40:21,  6.71s/it]
 30%|██▉       | 152/512 [17:04<40:17,  6.71s/it]
                                                 
{'loss': '0.006482', 'grad_norm': '0.3306', 'learning_rate': '8.594e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 4595600, 'tokens/trainable': 69465, 'epoch': '0.5938'}
+
 30%|██▉       | 152/512 [17:04<40:17,  6.71s/it]
 30%|██▉       | 153/512 [17:11<40:15,  6.73s/it]
                                                 
{'loss': '0.01488', 'grad_norm': '2.509', 'learning_rate': '8.572e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 4626032, 'tokens/trainable': 69928, 'epoch': '0.5977'}
+
 30%|██▉       | 153/512 [17:11<40:15,  6.73s/it]
 30%|███       | 154/512 [17:18<40:06,  6.72s/it]
                                                 
{'loss': '0.001413', 'grad_norm': '0.08187', 'learning_rate': '8.551e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.67', 'tokens/total': 4656352, 'tokens/trainable': 70395, 'epoch': '0.6016'}
+
 30%|███       | 154/512 [17:18<40:06,  6.72s/it]
 30%|███       | 155/512 [17:24<39:02,  6.56s/it]
                                                 
{'loss': '0.001624', 'grad_norm': '0.203', 'learning_rate': '8.53e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 4684672, 'tokens/trainable': 70829, 'epoch': '0.6055'}
+
 30%|███       | 155/512 [17:24<39:02,  6.56s/it]
 30%|███       | 156/512 [17:31<39:13,  6.61s/it]
                                                 
{'loss': '0.001468', 'grad_norm': '0.1048', 'learning_rate': '8.508e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 4715216, 'tokens/trainable': 71287, 'epoch': '0.6094'}
+
 30%|███       | 156/512 [17:31<39:13,  6.61s/it]
 31%|███       | 157/512 [17:37<39:19,  6.65s/it]
                                                 
{'loss': '0.004101', 'grad_norm': '0.2741', 'learning_rate': '8.487e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 4745376, 'tokens/trainable': 71749, 'epoch': '0.6133'}
+
 31%|███       | 157/512 [17:37<39:19,  6.65s/it]
 31%|███       | 158/512 [17:44<39:23,  6.68s/it]
                                                 
{'loss': '0.004408', 'grad_norm': '0.3079', 'learning_rate': '8.465e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.18', 'tokens/total': 4775776, 'tokens/trainable': 72185, 'epoch': '0.6172'}
+
 31%|███       | 158/512 [17:44<39:23,  6.68s/it]
 31%|███       | 159/512 [17:51<39:13,  6.67s/it]
                                                 
{'loss': '0.003525', 'grad_norm': '0.3344', 'learning_rate': '8.443e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 4805856, 'tokens/trainable': 72604, 'epoch': '0.6211'}
+
 31%|███       | 159/512 [17:51<39:13,  6.67s/it]
 31%|███▏      | 160/512 [17:57<39:12,  6.68s/it]
                                                 
{'loss': '0.01557', 'grad_norm': '0.6212', 'learning_rate': '8.421e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.56', 'tokens/total': 4836304, 'tokens/trainable': 73061, 'epoch': '0.625'}
+
 31%|███▏      | 160/512 [17:57<39:12,  6.68s/it][2026-08-18 14:22:57,346] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
+
 31%|███▏      | 161/512 [18:06<41:56,  7.17s/it]
                                                 
{'loss': '0.005395', 'grad_norm': '0.5695', 'learning_rate': '8.399e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.98', 'tokens/total': 4866544, 'tokens/trainable': 73474, 'epoch': '0.6289'}
+
 31%|███▏      | 161/512 [18:06<41:56,  7.17s/it]
 32%|███▏      | 162/512 [18:12<41:03,  7.04s/it]
                                                 
{'loss': '0.004907', 'grad_norm': '0.3221', 'learning_rate': '8.376e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 4897120, 'tokens/trainable': 73923, 'epoch': '0.6328'}
+
 32%|███▏      | 162/512 [18:12<41:03,  7.04s/it]
 32%|███▏      | 163/512 [18:19<40:21,  6.94s/it]
                                                 
{'loss': '0.01177', 'grad_norm': '0.2563', 'learning_rate': '8.354e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 4927440, 'tokens/trainable': 74379, 'epoch': '0.6367'}
+
 32%|███▏      | 163/512 [18:19<40:21,  6.94s/it]
 32%|███▏      | 164/512 [18:26<40:05,  6.91s/it]
                                                 
{'loss': '0.01483', 'grad_norm': '0.3738', 'learning_rate': '8.332e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.45', 'tokens/total': 4958032, 'tokens/trainable': 74830, 'epoch': '0.6406'}
+
 32%|███▏      | 164/512 [18:26<40:05,  6.91s/it]
 32%|███▏      | 165/512 [18:33<39:34,  6.84s/it]
                                                 
{'loss': '0.00068', 'grad_norm': '0.0768', 'learning_rate': '8.309e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.73', 'tokens/total': 4988368, 'tokens/trainable': 75262, 'epoch': '0.6445'}
+
 32%|███▏      | 165/512 [18:33<39:34,  6.84s/it]
 32%|███▏      | 166/512 [18:39<39:08,  6.79s/it]
                                                 
{'loss': '0.0005167', 'grad_norm': '0.1594', 'learning_rate': '8.286e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.84', 'tokens/total': 5018672, 'tokens/trainable': 75703, 'epoch': '0.6484'}
+
 32%|███▏      | 166/512 [18:39<39:08,  6.79s/it]
 33%|███▎      | 167/512 [18:46<39:03,  6.79s/it]
                                                 
{'loss': '0.003161', 'grad_norm': '0.1887', 'learning_rate': '8.263e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.3', 'tokens/total': 5049344, 'tokens/trainable': 76164, 'epoch': '0.6523'}
+
 33%|███▎      | 167/512 [18:46<39:03,  6.79s/it]
 33%|███▎      | 168/512 [18:53<38:51,  6.78s/it]
                                                 
{'loss': '0.01094', 'grad_norm': '0.3047', 'learning_rate': '8.24e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 5079856, 'tokens/trainable': 76623, 'epoch': '0.6562'}
+
 33%|███▎      | 168/512 [18:53<38:51,  6.78s/it]
 33%|███▎      | 169/512 [18:59<38:32,  6.74s/it]
                                                 
{'loss': '0.009002', 'grad_norm': '0.3929', 'learning_rate': '8.217e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.39', 'tokens/total': 5110192, 'tokens/trainable': 77047, 'epoch': '0.6602'}
+
 33%|███▎      | 169/512 [18:59<38:32,  6.74s/it]
 33%|███▎      | 170/512 [19:06<38:25,  6.74s/it]
                                                 
{'loss': '0.001463', 'grad_norm': '0.09876', 'learning_rate': '8.194e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 5140592, 'tokens/trainable': 77496, 'epoch': '0.6641'}
+
 33%|███▎      | 170/512 [19:06<38:25,  6.74s/it]
 33%|███▎      | 171/512 [19:13<38:15,  6.73s/it]
                                                 
{'loss': '0.0004142', 'grad_norm': '0.02294', 'learning_rate': '8.171e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 5170848, 'tokens/trainable': 77956, 'epoch': '0.668'}
+
 33%|███▎      | 171/512 [19:13<38:15,  6.73s/it]
 34%|███▎      | 172/512 [19:20<38:10,  6.74s/it]
                                                 
{'loss': '0.006397', 'grad_norm': '0.2763', 'learning_rate': '8.147e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.78', 'tokens/total': 5201280, 'tokens/trainable': 78410, 'epoch': '0.6719'}
+
 34%|███▎      | 172/512 [19:20<38:10,  6.74s/it]
 34%|███▍      | 173/512 [19:26<37:56,  6.72s/it]
                                                 
{'loss': '0.001727', 'grad_norm': '0.07086', 'learning_rate': '8.124e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 5231712, 'tokens/trainable': 78849, 'epoch': '0.6758'}
+
 34%|███▍      | 173/512 [19:26<37:56,  6.72s/it]
 34%|███▍      | 174/512 [19:33<37:33,  6.67s/it]
                                                 
{'loss': '0.01194', 'grad_norm': '0.4524', 'learning_rate': '8.1e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.62', 'memory/max_allocated (GiB)': '33.62', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.07', 'tokens/total': 5261440, 'tokens/trainable': 79266, 'epoch': '0.6797'}
+
 34%|███▍      | 174/512 [19:33<37:33,  6.67s/it]
 34%|███▍      | 175/512 [19:40<37:30,  6.68s/it]
                                                 
{'loss': '0.002157', 'grad_norm': '0.1293', 'learning_rate': '8.076e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.53', 'tokens/total': 5291728, 'tokens/trainable': 79736, 'epoch': '0.6836'}
+
 34%|███▍      | 175/512 [19:40<37:30,  6.68s/it]
 34%|███▍      | 176/512 [19:46<37:30,  6.70s/it]
                                                 
{'loss': '0.001708', 'grad_norm': '0.1321', 'learning_rate': '8.053e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 5322240, 'tokens/trainable': 80214, 'epoch': '0.6875'}
+
 34%|███▍      | 176/512 [19:46<37:30,  6.70s/it]
 35%|███▍      | 177/512 [19:53<37:23,  6.70s/it]
                                                 
{'loss': '0.002082', 'grad_norm': '0.09734', 'learning_rate': '8.029e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.56', 'tokens/total': 5352576, 'tokens/trainable': 80661, 'epoch': '0.6914'}
+
 35%|███▍      | 177/512 [19:53<37:23,  6.70s/it]
 35%|███▍      | 178/512 [20:00<37:19,  6.71s/it]
                                                 
{'loss': '0.006122', 'grad_norm': '0.8079', 'learning_rate': '8.005e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.82', 'tokens/total': 5382784, 'tokens/trainable': 81129, 'epoch': '0.6953'}
+
 35%|███▍      | 178/512 [20:00<37:19,  6.71s/it]
 35%|███▍      | 179/512 [20:06<37:05,  6.68s/it]
                                                 
{'loss': '0.003698', 'grad_norm': '0.8061', 'learning_rate': '7.98e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 5412784, 'tokens/trainable': 81576, 'epoch': '0.6992'}
+
 35%|███▍      | 179/512 [20:06<37:05,  6.68s/it]
 35%|███▌      | 180/512 [20:13<37:02,  6.69s/it]
                                                 
{'loss': '0.0007224', 'grad_norm': '0.05354', 'learning_rate': '7.956e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 5442960, 'tokens/trainable': 82049, 'epoch': '0.7031'}
+
 35%|███▌      | 180/512 [20:13<37:02,  6.69s/it]
 35%|███▌      | 181/512 [20:20<37:01,  6.71s/it]
                                                 
{'loss': '0.01404', 'grad_norm': '0.8788', 'learning_rate': '7.932e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '38.79', 'tokens/total': 5473216, 'tokens/trainable': 82566, 'epoch': '0.707'}
+
 35%|███▌      | 181/512 [20:20<37:01,  6.71s/it]
 36%|███▌      | 182/512 [20:27<36:56,  6.72s/it]
                                                 
{'loss': '0.001835', 'grad_norm': '0.2329', 'learning_rate': '7.907e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 5503568, 'tokens/trainable': 83054, 'epoch': '0.7109'}
+
 36%|███▌      | 182/512 [20:27<36:56,  6.72s/it]
 36%|███▌      | 183/512 [20:33<36:48,  6.71s/it]
                                                 
{'loss': '0.001123', 'grad_norm': '0.08039', 'learning_rate': '7.883e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 5533760, 'tokens/trainable': 83501, 'epoch': '0.7148'}
+
 36%|███▌      | 183/512 [20:33<36:48,  6.71s/it]
 36%|███▌      | 184/512 [20:40<36:41,  6.71s/it]
                                                 
{'loss': '0.0004567', 'grad_norm': '0.02801', 'learning_rate': '7.858e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.48', 'tokens/total': 5564240, 'tokens/trainable': 84007, 'epoch': '0.7188'}
+
 36%|███▌      | 184/512 [20:40<36:41,  6.71s/it]
 36%|███▌      | 185/512 [20:47<36:36,  6.72s/it]
                                                 
{'loss': '0.0006747', 'grad_norm': '0.04384', 'learning_rate': '7.833e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.54', 'tokens/total': 5594448, 'tokens/trainable': 84491, 'epoch': '0.7227'}
+
 36%|███▌      | 185/512 [20:47<36:36,  6.72s/it]
 36%|███▋      | 186/512 [20:53<36:33,  6.73s/it]
                                                 
{'loss': '0.001428', 'grad_norm': '0.117', 'learning_rate': '7.808e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 5624816, 'tokens/trainable': 84952, 'epoch': '0.7266'}
+
 36%|███▋      | 186/512 [20:53<36:33,  6.73s/it]
 37%|███▋      | 187/512 [21:00<36:25,  6.72s/it]
                                                 
{'loss': '0.0002471', 'grad_norm': '0.05614', 'learning_rate': '7.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 5655104, 'tokens/trainable': 85400, 'epoch': '0.7305'}
+
 37%|███▋      | 187/512 [21:00<36:25,  6.72s/it]
 37%|███▋      | 188/512 [21:07<36:20,  6.73s/it]
                                                 
{'loss': '0.004403', 'grad_norm': '0.3061', 'learning_rate': '7.758e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 5685360, 'tokens/trainable': 85889, 'epoch': '0.7344'}
+
 37%|███▋      | 188/512 [21:07<36:20,  6.73s/it]
 37%|███▋      | 189/512 [21:14<36:14,  6.73s/it]
                                                 
{'loss': '0.00549', 'grad_norm': '0.4333', 'learning_rate': '7.733e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 5715648, 'tokens/trainable': 86396, 'epoch': '0.7383'}
+
 37%|███▋      | 189/512 [21:14<36:14,  6.73s/it]
 37%|███▋      | 190/512 [21:20<36:05,  6.73s/it]
                                                 
{'loss': '0.006627', 'grad_norm': '0.4977', 'learning_rate': '7.708e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 5746240, 'tokens/trainable': 86867, 'epoch': '0.7422'}
+
 37%|███▋      | 190/512 [21:20<36:05,  6.73s/it]
 37%|███▋      | 191/512 [21:27<36:02,  6.74s/it]
                                                 
{'loss': '0.009029', 'grad_norm': '1', 'learning_rate': '7.683e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.07', 'tokens/total': 5776816, 'tokens/trainable': 87314, 'epoch': '0.7461'}
+
 37%|███▋      | 191/512 [21:27<36:02,  6.74s/it]
 38%|███▊      | 192/512 [21:34<35:50,  6.72s/it]
                                                 
{'loss': '5.811e-05', 'grad_norm': '0.006457', 'learning_rate': '7.657e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 5807248, 'tokens/trainable': 87736, 'epoch': '0.75'}
+
 38%|███▊      | 192/512 [21:34<35:50,  6.72s/it][2026-08-18 14:26:33,856] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
+
 38%|███▊      | 193/512 [21:42<38:34,  7.26s/it]
                                                 
{'loss': '0.001513', 'grad_norm': '0.6313', 'learning_rate': '7.632e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 5837680, 'tokens/trainable': 88219, 'epoch': '0.7539'}
+
 38%|███▊      | 193/512 [21:42<38:34,  7.26s/it]
 38%|███▊      | 194/512 [21:49<37:39,  7.10s/it]
                                                 
{'loss': '3.505e-05', 'grad_norm': '0.003887', 'learning_rate': '7.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 5868000, 'tokens/trainable': 88699, 'epoch': '0.7578'}
+
 38%|███▊      | 194/512 [21:49<37:39,  7.10s/it]
 38%|███▊      | 195/512 [21:55<36:01,  6.82s/it]
                                                 
{'loss': '0.0001714', 'grad_norm': '0.05365', 'learning_rate': '7.58e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.39', 'tokens/total': 5896224, 'tokens/trainable': 89132, 'epoch': '0.7617'}
+
 38%|███▊      | 195/512 [21:55<36:01,  6.82s/it]
 38%|███▊      | 196/512 [22:02<35:45,  6.79s/it]
                                                 
{'loss': '0.007072', 'grad_norm': '1.521', 'learning_rate': '7.555e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '37.24', 'tokens/total': 5926624, 'tokens/trainable': 89594, 'epoch': '0.7656'}
+
 38%|███▊      | 196/512 [22:02<35:45,  6.79s/it]
 38%|███▊      | 197/512 [22:08<34:42,  6.61s/it]
                                                 
{'loss': '0.0001249', 'grad_norm': '0.01594', 'learning_rate': '7.529e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.34', 'memory/max_allocated (GiB)': '33.34', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '38.08', 'tokens/total': 5954944, 'tokens/trainable': 90042, 'epoch': '0.7695'}
+
 38%|███▊      | 197/512 [22:08<34:42,  6.61s/it]
 39%|███▊      | 198/512 [22:15<34:46,  6.65s/it]
                                                 
{'loss': '0.004654', 'grad_norm': '0.4873', 'learning_rate': '7.503e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 5985200, 'tokens/trainable': 90503, 'epoch': '0.7734'}
+
 39%|███▊      | 198/512 [22:15<34:46,  6.65s/it]
 39%|███▉      | 199/512 [22:22<34:59,  6.71s/it]
                                                 
{'loss': '0.0007038', 'grad_norm': '0.06984', 'learning_rate': '7.477e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 6015952, 'tokens/trainable': 90982, 'epoch': '0.7773'}
+
 39%|███▉      | 199/512 [22:22<34:59,  6.71s/it]
 39%|███▉      | 200/512 [22:28<34:54,  6.71s/it]
                                                 
{'loss': '0.001063', 'grad_norm': '0.09597', 'learning_rate': '7.451e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '39.3', 'tokens/total': 6046208, 'tokens/trainable': 91466, 'epoch': '0.7812'}
+
 39%|███▉      | 200/512 [22:28<34:54,  6.71s/it]
 39%|███▉      | 201/512 [22:35<34:42,  6.70s/it]
                                                 
{'loss': '0.0004073', 'grad_norm': '0.03423', 'learning_rate': '7.424e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 6076208, 'tokens/trainable': 91912, 'epoch': '0.7852'}
+
 39%|███▉      | 201/512 [22:35<34:42,  6.70s/it]
 39%|███▉      | 202/512 [22:42<34:34,  6.69s/it]
                                                 
{'loss': '0.003431', 'grad_norm': '0.2397', 'learning_rate': '7.398e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.86', 'tokens/total': 6106560, 'tokens/trainable': 92385, 'epoch': '0.7891'}
+
 39%|███▉      | 202/512 [22:42<34:34,  6.69s/it]
 40%|███▉      | 203/512 [22:48<34:26,  6.69s/it]
                                                 
{'loss': '0.0216', 'grad_norm': '0.88', 'learning_rate': '7.372e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 6136640, 'tokens/trainable': 92842, 'epoch': '0.793'}
+
 40%|███▉      | 203/512 [22:48<34:26,  6.69s/it]
 40%|███▉      | 204/512 [22:55<34:22,  6.70s/it]
                                                 
{'loss': '0.0006239', 'grad_norm': '0.1207', 'learning_rate': '7.345e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 6167184, 'tokens/trainable': 93323, 'epoch': '0.7969'}
+
 40%|███▉      | 204/512 [22:55<34:22,  6.70s/it]
 40%|████      | 205/512 [23:02<34:10,  6.68s/it]
                                                 
{'loss': '0.002666', 'grad_norm': '0.3556', 'learning_rate': '7.319e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 6197376, 'tokens/trainable': 93756, 'epoch': '0.8008'}
+
 40%|████      | 205/512 [23:02<34:10,  6.68s/it]
 40%|████      | 206/512 [23:09<34:09,  6.70s/it]
                                                 
{'loss': '0.001572', 'grad_norm': '0.2445', 'learning_rate': '7.292e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.54', 'tokens/total': 6227600, 'tokens/trainable': 94235, 'epoch': '0.8047'}
+
 40%|████      | 206/512 [23:09<34:09,  6.70s/it]
 40%|████      | 207/512 [23:15<34:03,  6.70s/it]
                                                 
{'loss': '0.003192', 'grad_norm': '0.2322', 'learning_rate': '7.266e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 6257760, 'tokens/trainable': 94733, 'epoch': '0.8086'}
+
 40%|████      | 207/512 [23:15<34:03,  6.70s/it]
 41%|████      | 208/512 [23:22<33:57,  6.70s/it]
                                                 
{'loss': '0.001887', 'grad_norm': '0.1369', 'learning_rate': '7.239e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.74', 'tokens/total': 6288048, 'tokens/trainable': 95218, 'epoch': '0.8125'}
+
 41%|████      | 208/512 [23:22<33:57,  6.70s/it]
 41%|████      | 209/512 [23:29<33:44,  6.68s/it]
                                                 
{'loss': '0.001802', 'grad_norm': '0.2183', 'learning_rate': '7.212e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '29.23', 'tokens/total': 6318688, 'tokens/trainable': 95619, 'epoch': '0.8164'}
+
 41%|████      | 209/512 [23:29<33:44,  6.68s/it]
 41%|████      | 210/512 [23:35<33:39,  6.69s/it]
                                                 
{'loss': '0.0125', 'grad_norm': '0.8775', 'learning_rate': '7.185e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 6348944, 'tokens/trainable': 96076, 'epoch': '0.8203'}
+
 41%|████      | 210/512 [23:35<33:39,  6.69s/it]
 41%|████      | 211/512 [23:42<33:29,  6.68s/it]
                                                 
{'loss': '0.006944', 'grad_norm': '0.4041', 'learning_rate': '7.158e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.34', 'tokens/total': 6378960, 'tokens/trainable': 96505, 'epoch': '0.8242'}
+
 41%|████      | 211/512 [23:42<33:29,  6.68s/it]
 41%|████▏     | 212/512 [23:49<33:25,  6.68s/it]
                                                 
{'loss': '0.001738', 'grad_norm': '0.5415', 'learning_rate': '7.131e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 6409472, 'tokens/trainable': 96930, 'epoch': '0.8281'}
+
 41%|████▏     | 212/512 [23:49<33:25,  6.68s/it]
 42%|████▏     | 213/512 [23:55<33:19,  6.69s/it]
                                                 
{'loss': '0.0002014', 'grad_norm': '0.01034', 'learning_rate': '7.104e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.72', 'tokens/total': 6439904, 'tokens/trainable': 97361, 'epoch': '0.832'}
+
 42%|████▏     | 213/512 [23:55<33:19,  6.69s/it]
 42%|████▏     | 214/512 [24:02<33:17,  6.70s/it]
                                                 
{'loss': '0.009724', 'grad_norm': '0.5834', 'learning_rate': '7.077e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 6470176, 'tokens/trainable': 97858, 'epoch': '0.8359'}
+
 42%|████▏     | 214/512 [24:02<33:17,  6.70s/it]
 42%|████▏     | 215/512 [24:09<33:06,  6.69s/it]
                                                 
{'loss': '0.004063', 'grad_norm': '0.3147', 'learning_rate': '7.05e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.47', 'tokens/total': 6500336, 'tokens/trainable': 98308, 'epoch': '0.8398'}
+
 42%|████▏     | 215/512 [24:09<33:06,  6.69s/it]
 42%|████▏     | 216/512 [24:15<32:58,  6.68s/it]
                                                 
{'loss': '0.0002783', 'grad_norm': '0.01362', 'learning_rate': '7.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.41', 'tokens/total': 6530688, 'tokens/trainable': 98781, 'epoch': '0.8438'}
+
 42%|████▏     | 216/512 [24:15<32:58,  6.68s/it]
 42%|████▏     | 217/512 [24:22<32:54,  6.69s/it]
                                                 
{'loss': '0.0002952', 'grad_norm': '0.02845', 'learning_rate': '6.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.76', 'tokens/total': 6561040, 'tokens/trainable': 99266, 'epoch': '0.8477'}
+
 42%|████▏     | 217/512 [24:22<32:54,  6.69s/it]
 43%|████▎     | 218/512 [24:28<32:02,  6.54s/it]
                                                 
{'loss': '0.002927', 'grad_norm': '0.1757', 'learning_rate': '6.968e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.65', 'tokens/total': 6589392, 'tokens/trainable': 99731, 'epoch': '0.8516'}
+
 43%|████▎     | 218/512 [24:28<32:02,  6.54s/it]
 43%|████▎     | 219/512 [24:35<32:09,  6.58s/it]
                                                 
{'loss': '0.005347', 'grad_norm': '0.2501', 'learning_rate': '6.94e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 6619824, 'tokens/trainable': 100164, 'epoch': '0.8555'}
+
 43%|████▎     | 219/512 [24:35<32:09,  6.58s/it]
 43%|████▎     | 220/512 [24:42<32:16,  6.63s/it]
                                                 
{'loss': '0.00676', 'grad_norm': '0.366', 'learning_rate': '6.913e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 6650224, 'tokens/trainable': 100620, 'epoch': '0.8594'}
+
 43%|████▎     | 220/512 [24:42<32:16,  6.63s/it]
 43%|████▎     | 221/512 [24:49<32:17,  6.66s/it]
                                                 
{'loss': '0.004021', 'grad_norm': '0.2871', 'learning_rate': '6.885e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 6680384, 'tokens/trainable': 101080, 'epoch': '0.8633'}
+
 43%|████▎     | 221/512 [24:49<32:17,  6.66s/it]
 43%|████▎     | 222/512 [24:55<32:13,  6.67s/it]
                                                 
{'loss': '0.0006698', 'grad_norm': '0.05318', 'learning_rate': '6.857e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.8', 'tokens/total': 6710896, 'tokens/trainable': 101546, 'epoch': '0.8672'}
+
 43%|████▎     | 222/512 [24:55<32:13,  6.67s/it]
 44%|████▎     | 223/512 [25:02<32:06,  6.67s/it]
                                                 
{'loss': '0.0002042', 'grad_norm': '0.01632', 'learning_rate': '6.83e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 6741280, 'tokens/trainable': 101987, 'epoch': '0.8711'}
+
 44%|████▎     | 223/512 [25:02<32:06,  6.67s/it]
 44%|████▍     | 224/512 [25:09<32:05,  6.69s/it]
                                                 
{'loss': '0.001851', 'grad_norm': '0.09936', 'learning_rate': '6.802e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 6771488, 'tokens/trainable': 102435, 'epoch': '0.875'}
+
 44%|████▍     | 224/512 [25:09<32:05,  6.69s/it][2026-08-18 14:30:08,595] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
+
 44%|████▍     | 225/512 [25:17<34:28,  7.21s/it]
                                                 
{'loss': '0.0007717', 'grad_norm': '0.04545', 'learning_rate': '6.774e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '28.86', 'tokens/total': 6801744, 'tokens/trainable': 102843, 'epoch': '0.8789'}
+
 44%|████▍     | 225/512 [25:17<34:28,  7.21s/it]
 44%|████▍     | 226/512 [25:24<33:50,  7.10s/it]
                                                 
{'loss': '0.001548', 'grad_norm': '0.1694', 'learning_rate': '6.746e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.39', 'tokens/total': 6832432, 'tokens/trainable': 103342, 'epoch': '0.8828'}
+
 44%|████▍     | 226/512 [25:24<33:50,  7.10s/it]
 44%|████▍     | 227/512 [25:31<33:06,  6.97s/it]
                                                 
{'loss': '0.0002161', 'grad_norm': '0.01104', 'learning_rate': '6.718e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.71', 'tokens/train_per_sec_per_gpu': '40.81', 'tokens/total': 6862448, 'tokens/trainable': 103844, 'epoch': '0.8867'}
+
 44%|████▍     | 227/512 [25:31<33:06,  6.97s/it]
 45%|████▍     | 228/512 [25:37<32:37,  6.89s/it]
                                                 
{'loss': '0.007129', 'grad_norm': '0.3398', 'learning_rate': '6.69e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 6892832, 'tokens/trainable': 104307, 'epoch': '0.8906'}
+
 45%|████▍     | 228/512 [25:37<32:37,  6.89s/it]
 45%|████▍     | 229/512 [25:44<32:10,  6.82s/it]
                                                 
{'loss': '0.0003861', 'grad_norm': '0.02249', 'learning_rate': '6.662e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.49', 'tokens/total': 6922832, 'tokens/trainable': 104772, 'epoch': '0.8945'}
+
 45%|████▍     | 229/512 [25:44<32:10,  6.82s/it]
 45%|████▍     | 230/512 [25:51<31:56,  6.80s/it]
                                                 
{'loss': '0.0002201', 'grad_norm': '0.03081', 'learning_rate': '6.634e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 6953104, 'tokens/trainable': 105240, 'epoch': '0.8984'}
+
 45%|████▍     | 230/512 [25:51<31:56,  6.80s/it]
 45%|████▌     | 231/512 [25:57<31:41,  6.77s/it]
                                                 
{'loss': '0.001256', 'grad_norm': '0.1375', 'learning_rate': '6.606e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 6983440, 'tokens/trainable': 105703, 'epoch': '0.9023'}
+
 45%|████▌     | 231/512 [25:57<31:41,  6.77s/it]
 45%|████▌     | 232/512 [26:04<31:28,  6.74s/it]
                                                 
{'loss': '6.874e-05', 'grad_norm': '0.003156', 'learning_rate': '6.578e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 7013840, 'tokens/trainable': 106179, 'epoch': '0.9062'}
+
 45%|████▌     | 232/512 [26:04<31:28,  6.74s/it]
 46%|████▌     | 233/512 [26:11<31:21,  6.74s/it]
                                                 
{'loss': '0.000162', 'grad_norm': '0.02122', 'learning_rate': '6.55e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38', 'tokens/total': 7044240, 'tokens/trainable': 106683, 'epoch': '0.9102'}
+
 46%|████▌     | 233/512 [26:11<31:21,  6.74s/it]
 46%|████▌     | 234/512 [26:17<31:06,  6.71s/it]
                                                 
{'loss': '0.0003435', 'grad_norm': '0.03298', 'learning_rate': '6.522e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 7074224, 'tokens/trainable': 107121, 'epoch': '0.9141'}
+
 46%|████▌     | 234/512 [26:17<31:06,  6.71s/it]
 46%|████▌     | 235/512 [26:24<30:58,  6.71s/it]
                                                 
{'loss': '0.0006108', 'grad_norm': '0.08474', 'learning_rate': '6.493e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 7104672, 'tokens/trainable': 107561, 'epoch': '0.918'}
+
 46%|████▌     | 235/512 [26:24<30:58,  6.71s/it]
 46%|████▌     | 236/512 [26:31<30:49,  6.70s/it]
                                                 
{'loss': '0.0001098', 'grad_norm': '0.012', 'learning_rate': '6.465e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 7134784, 'tokens/trainable': 108018, 'epoch': '0.9219'}
+
 46%|████▌     | 236/512 [26:31<30:49,  6.70s/it]
 46%|████▋     | 237/512 [26:37<30:39,  6.69s/it]
                                                 
{'loss': '6.363e-05', 'grad_norm': '0.01181', 'learning_rate': '6.437e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.32', 'tokens/total': 7164960, 'tokens/trainable': 108446, 'epoch': '0.9258'}
+
 46%|████▋     | 237/512 [26:37<30:39,  6.69s/it]
 46%|████▋     | 238/512 [26:44<30:30,  6.68s/it]
                                                 
{'loss': '0.0002591', 'grad_norm': '0.05564', 'learning_rate': '6.408e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.15', 'tokens/total': 7195344, 'tokens/trainable': 108866, 'epoch': '0.9297'}
+
 46%|████▋     | 238/512 [26:44<30:30,  6.68s/it]
 47%|████▋     | 239/512 [26:51<30:28,  6.70s/it]
                                                 
{'loss': '0.004054', 'grad_norm': '0.7226', 'learning_rate': '6.38e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 7225680, 'tokens/trainable': 109316, 'epoch': '0.9336'}
+
 47%|████▋     | 239/512 [26:51<30:28,  6.70s/it]
 47%|████▋     | 240/512 [26:58<30:24,  6.71s/it]
                                                 
{'loss': '0.0004681', 'grad_norm': '0.0682', 'learning_rate': '6.351e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.27', 'tokens/total': 7256112, 'tokens/trainable': 109787, 'epoch': '0.9375'}
+
 47%|████▋     | 240/512 [26:58<30:24,  6.71s/it]
 47%|████▋     | 241/512 [27:04<30:13,  6.69s/it]
                                                 
{'loss': '0.001304', 'grad_norm': '0.174', 'learning_rate': '6.323e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.35', 'tokens/total': 7286384, 'tokens/trainable': 110246, 'epoch': '0.9414'}
+
 47%|████▋     | 241/512 [27:04<30:13,  6.69s/it]
 47%|████▋     | 242/512 [27:11<30:07,  6.70s/it]
                                                 
{'loss': '0.0004871', 'grad_norm': '0.05733', 'learning_rate': '6.294e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.22', 'tokens/total': 7316688, 'tokens/trainable': 110667, 'epoch': '0.9453'}
+
 47%|████▋     | 242/512 [27:11<30:07,  6.70s/it]
 47%|████▋     | 243/512 [27:18<30:02,  6.70s/it]
                                                 
{'loss': '0.01397', 'grad_norm': '0.804', 'learning_rate': '6.266e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.13', 'tokens/total': 7346992, 'tokens/trainable': 111154, 'epoch': '0.9492'}
+
 47%|████▋     | 243/512 [27:18<30:02,  6.70s/it]
 48%|████▊     | 244/512 [27:24<29:59,  6.72s/it]
                                                 
{'loss': '0.0001144', 'grad_norm': '0.0327', 'learning_rate': '6.237e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 7377632, 'tokens/trainable': 111611, 'epoch': '0.9531'}
+
 48%|████▊     | 244/512 [27:24<29:59,  6.72s/it]
 48%|████▊     | 245/512 [27:31<29:48,  6.70s/it]
                                                 
{'loss': '4.604e-05', 'grad_norm': '0.006486', 'learning_rate': '6.208e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 7407872, 'tokens/trainable': 112102, 'epoch': '0.957'}
+
 48%|████▊     | 245/512 [27:31<29:48,  6.70s/it]
 48%|████▊     | 246/512 [27:38<29:40,  6.69s/it]
                                                 
{'loss': '0.0002825', 'grad_norm': '0.04384', 'learning_rate': '6.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.24', 'tokens/total': 7438448, 'tokens/trainable': 112569, 'epoch': '0.9609'}
+
 48%|████▊     | 246/512 [27:38<29:40,  6.69s/it]
 48%|████▊     | 247/512 [27:44<28:57,  6.56s/it]
                                                 
{'loss': '0.002453', 'grad_norm': '0.17', 'learning_rate': '6.151e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 7466768, 'tokens/trainable': 113023, 'epoch': '0.9648'}
+
 48%|████▊     | 247/512 [27:44<28:57,  6.56s/it]
 48%|████▊     | 248/512 [27:51<29:08,  6.62s/it]
                                                 
{'loss': '0.0001876', 'grad_norm': '0.01652', 'learning_rate': '6.122e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.12', 'tokens/total': 7497360, 'tokens/trainable': 113464, 'epoch': '0.9688'}
+
 48%|████▊     | 248/512 [27:51<29:08,  6.62s/it]
 49%|████▊     | 249/512 [27:57<29:05,  6.64s/it]
                                                 
{'loss': '0.002968', 'grad_norm': '0.2792', 'learning_rate': '6.093e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 7527504, 'tokens/trainable': 113908, 'epoch': '0.9727'}
+
 49%|████▊     | 249/512 [27:57<29:05,  6.64s/it]
 49%|████▉     | 250/512 [28:04<28:24,  6.51s/it]
                                                 
{'loss': '0.005398', 'grad_norm': '0.4556', 'learning_rate': '6.065e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 7555808, 'tokens/trainable': 114348, 'epoch': '0.9766'}
+
 49%|████▉     | 250/512 [28:04<28:24,  6.51s/it]
 49%|████▉     | 251/512 [28:10<28:33,  6.57s/it]
                                                 
{'loss': '0.0003356', 'grad_norm': '0.03177', 'learning_rate': '6.036e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.23', 'tokens/total': 7586320, 'tokens/trainable': 114830, 'epoch': '0.9805'}
+
 49%|████▉     | 251/512 [28:10<28:33,  6.57s/it]
 49%|████▉     | 252/512 [28:17<28:38,  6.61s/it]
                                                 
{'loss': '0.002921', 'grad_norm': '0.297', 'learning_rate': '6.007e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 7616576, 'tokens/trainable': 115278, 'epoch': '0.9844'}
+
 49%|████▉     | 252/512 [28:17<28:38,  6.61s/it]
 49%|████▉     | 253/512 [28:24<28:42,  6.65s/it]
                                                 
{'loss': '0.0004065', 'grad_norm': '0.09317', 'learning_rate': '5.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.29', 'tokens/total': 7647152, 'tokens/trainable': 115699, 'epoch': '0.9883'}
+
 49%|████▉     | 253/512 [28:24<28:42,  6.65s/it]
 50%|████▉     | 254/512 [28:30<28:39,  6.66s/it]
                                                 
{'loss': '0.001994', 'grad_norm': '0.2161', 'learning_rate': '5.949e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '30.81', 'tokens/total': 7677584, 'tokens/trainable': 116133, 'epoch': '0.9922'}
+
 50%|████▉     | 254/512 [28:30<28:39,  6.66s/it]
 50%|████▉     | 255/512 [28:37<28:33,  6.67s/it]
                                                 
{'loss': '0.0008475', 'grad_norm': '0.08144', 'learning_rate': '5.92e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 7708016, 'tokens/trainable': 116591, 'epoch': '0.9961'}
+
 50%|████▉     | 255/512 [28:37<28:33,  6.67s/it]
 50%|█████     | 256/512 [28:44<28:41,  6.72s/it]
                                                 
{'loss': '0.00071', 'grad_norm': '0.06009', 'learning_rate': '5.891e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.11', 'tokens/total': 7738608, 'tokens/trainable': 117039, 'epoch': '1'}
+
 50%|█████     | 256/512 [28:44<28:41,  6.72s/it][2026-08-18 14:33:44,024] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
+
 50%|█████     | 257/512 [28:53<31:19,  7.37s/it]
                                                 
{'loss': '0.0001019', 'grad_norm': '0.0194', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 7768944, 'tokens/trainable': 117495, 'epoch': '1.004'}
+
 50%|█████     | 257/512 [28:53<31:19,  7.37s/it]
 50%|█████     | 258/512 [29:00<30:19,  7.16s/it]
                                                 
{'loss': '0.0001532', 'grad_norm': '0.02072', 'learning_rate': '5.834e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 7799232, 'tokens/trainable': 117925, 'epoch': '1.008'}
+
 50%|█████     | 258/512 [29:00<30:19,  7.16s/it]
 51%|█████     | 259/512 [29:06<29:37,  7.02s/it]
                                                 
{'loss': '4.119e-05', 'grad_norm': '0.002171', 'learning_rate': '5.805e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.66', 'tokens/total': 7829568, 'tokens/trainable': 118373, 'epoch': '1.012'}
+
 51%|█████     | 259/512 [29:06<29:37,  7.02s/it]
 51%|█████     | 260/512 [29:13<29:05,  6.93s/it]
                                                 
{'loss': '2.263e-05', 'grad_norm': '0.001111', 'learning_rate': '5.776e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 7860000, 'tokens/trainable': 118816, 'epoch': '1.016'}
+
 51%|█████     | 260/512 [29:13<29:05,  6.93s/it]
 51%|█████     | 261/512 [29:20<28:39,  6.85s/it]
                                                 
{'loss': '7.258e-05', 'grad_norm': '0.0133', 'learning_rate': '5.747e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 7890464, 'tokens/trainable': 119263, 'epoch': '1.02'}
+
 51%|█████     | 261/512 [29:20<28:39,  6.85s/it]
 51%|█████     | 262/512 [29:26<28:25,  6.82s/it]
                                                 
{'loss': '5.408e-05', 'grad_norm': '0.06038', 'learning_rate': '5.718e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 7920944, 'tokens/trainable': 119772, 'epoch': '1.023'}
+
 51%|█████     | 262/512 [29:26<28:25,  6.82s/it]
 51%|█████▏    | 263/512 [29:33<28:12,  6.80s/it]
                                                 
{'loss': '5.063e-05', 'grad_norm': '0.003908', 'learning_rate': '5.689e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 7951296, 'tokens/trainable': 120269, 'epoch': '1.027'}
+
 51%|█████▏    | 263/512 [29:33<28:12,  6.80s/it]
 52%|█████▏    | 264/512 [29:40<28:01,  6.78s/it]
                                                 
{'loss': '3.142e-05', 'grad_norm': '0.008469', 'learning_rate': '5.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.05', 'tokens/total': 7981856, 'tokens/trainable': 120746, 'epoch': '1.031'}
+
 52%|█████▏    | 264/512 [29:40<28:01,  6.78s/it]
 52%|█████▏    | 265/512 [29:47<27:46,  6.74s/it]
                                                 
{'loss': '0.001199', 'grad_norm': '0.2763', 'learning_rate': '5.631e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.3', 'tokens/total': 8012192, 'tokens/trainable': 121204, 'epoch': '1.035'}
+
 52%|█████▏    | 265/512 [29:47<27:46,  6.74s/it]
 52%|█████▏    | 266/512 [29:53<27:40,  6.75s/it]
                                                 
{'loss': '0.01105', 'grad_norm': '0.3999', 'learning_rate': '5.602e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.23', 'tokens/total': 8042736, 'tokens/trainable': 121670, 'epoch': '1.039'}
+
 52%|█████▏    | 266/512 [29:53<27:40,  6.75s/it]
 52%|█████▏    | 267/512 [30:00<27:32,  6.74s/it]
                                                 
{'loss': '0.002067', 'grad_norm': '0.1673', 'learning_rate': '5.573e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '40.11', 'tokens/total': 8073024, 'tokens/trainable': 122183, 'epoch': '1.043'}
+
 52%|█████▏    | 267/512 [30:00<27:32,  6.74s/it]
 52%|█████▏    | 268/512 [30:07<27:22,  6.73s/it]
                                                 
{'loss': '0.0002631', 'grad_norm': '0.03506', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.41', 'tokens/total': 8103280, 'tokens/trainable': 122645, 'epoch': '1.047'}
+
 52%|█████▏    | 268/512 [30:07<27:22,  6.73s/it]
 53%|█████▎    | 269/512 [30:13<27:11,  6.71s/it]
                                                 
{'loss': '7.13e-05', 'grad_norm': '0.006357', 'learning_rate': '5.515e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.09', 'tokens/total': 8133440, 'tokens/trainable': 123105, 'epoch': '1.051'}
+
 53%|█████▎    | 269/512 [30:13<27:11,  6.71s/it]
 53%|█████▎    | 270/512 [30:20<27:06,  6.72s/it]
                                                 
{'loss': '0.000502', 'grad_norm': '0.06589', 'learning_rate': '5.485e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 8163920, 'tokens/trainable': 123595, 'epoch': '1.055'}
+
 53%|█████▎    | 270/512 [30:20<27:06,  6.72s/it]
 53%|█████▎    | 271/512 [30:27<26:57,  6.71s/it]
                                                 
{'loss': '0.0007096', 'grad_norm': '0.05696', 'learning_rate': '5.456e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.65', 'tokens/total': 8194032, 'tokens/trainable': 124077, 'epoch': '1.059'}
+
 53%|█████▎    | 271/512 [30:27<26:57,  6.71s/it]
 53%|█████▎    | 272/512 [30:34<26:52,  6.72s/it]
                                                 
{'loss': '0.0002944', 'grad_norm': '0.09019', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.77', 'tokens/total': 8224320, 'tokens/trainable': 124547, 'epoch': '1.062'}
+
 53%|█████▎    | 272/512 [30:34<26:52,  6.72s/it]
 53%|█████▎    | 273/512 [30:40<26:42,  6.71s/it]
                                                 
{'loss': '0.001365', 'grad_norm': '0.1311', 'learning_rate': '5.398e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 8254368, 'tokens/trainable': 124977, 'epoch': '1.066'}
+
 53%|█████▎    | 273/512 [30:40<26:42,  6.71s/it]
 54%|█████▎    | 274/512 [30:47<26:34,  6.70s/it]
                                                 
{'loss': '0.003348', 'grad_norm': '0.3515', 'learning_rate': '5.369e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 8284944, 'tokens/trainable': 125438, 'epoch': '1.07'}
+
 54%|█████▎    | 274/512 [30:47<26:34,  6.70s/it]
 54%|█████▎    | 275/512 [30:54<26:30,  6.71s/it]
                                                 
{'loss': '0.001243', 'grad_norm': '0.3394', 'learning_rate': '5.34e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.49', 'tokens/total': 8315584, 'tokens/trainable': 125889, 'epoch': '1.074'}
+
 54%|█████▎    | 275/512 [30:54<26:30,  6.71s/it]
 54%|█████▍    | 276/512 [31:00<26:24,  6.71s/it]
                                                 
{'loss': '0.004629', 'grad_norm': '0.3979', 'learning_rate': '5.311e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 8345952, 'tokens/trainable': 126363, 'epoch': '1.078'}
+
 54%|█████▍    | 276/512 [31:00<26:24,  6.71s/it]
 54%|█████▍    | 277/512 [31:07<26:20,  6.73s/it]
                                                 
{'loss': '0.004942', 'grad_norm': '0.2943', 'learning_rate': '5.282e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 8376416, 'tokens/trainable': 126828, 'epoch': '1.082'}
+
 54%|█████▍    | 277/512 [31:07<26:20,  6.73s/it]
 54%|█████▍    | 278/512 [31:14<26:12,  6.72s/it]
                                                 
{'loss': '0.0001571', 'grad_norm': '0.04226', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.21', 'tokens/total': 8406640, 'tokens/trainable': 127331, 'epoch': '1.086'}
+
 54%|█████▍    | 278/512 [31:14<26:12,  6.72s/it]
 54%|█████▍    | 279/512 [31:21<26:03,  6.71s/it]
                                                 
{'loss': '9.962e-05', 'grad_norm': '0.008294', 'learning_rate': '5.224e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.39', 'tokens/total': 8436784, 'tokens/trainable': 127798, 'epoch': '1.09'}
+
 54%|█████▍    | 279/512 [31:21<26:03,  6.71s/it]
 55%|█████▍    | 280/512 [31:27<25:56,  6.71s/it]
                                                 
{'loss': '0.0002119', 'grad_norm': '0.01799', 'learning_rate': '5.195e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.84', 'tokens/total': 8467152, 'tokens/trainable': 128247, 'epoch': '1.094'}
+
 55%|█████▍    | 280/512 [31:27<25:56,  6.71s/it]
 55%|█████▍    | 281/512 [31:34<25:51,  6.72s/it]
                                                 
{'loss': '0.0003', 'grad_norm': '0.05587', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 8497424, 'tokens/trainable': 128729, 'epoch': '1.098'}
+
 55%|█████▍    | 281/512 [31:34<25:51,  6.72s/it]
 55%|█████▌    | 282/512 [31:41<25:45,  6.72s/it]
                                                 
{'loss': '0.0003475', 'grad_norm': '0.0335', 'learning_rate': '5.138e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 8527616, 'tokens/trainable': 129203, 'epoch': '1.102'}
+
 55%|█████▌    | 282/512 [31:41<25:45,  6.72s/it]
 55%|█████▌    | 283/512 [31:47<25:39,  6.72s/it]
                                                 
{'loss': '2.523e-05', 'grad_norm': '0.00135', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 8557984, 'tokens/trainable': 129653, 'epoch': '1.105'}
+
 55%|█████▌    | 283/512 [31:47<25:39,  6.72s/it]
 55%|█████▌    | 284/512 [31:54<25:34,  6.73s/it]
                                                 
{'loss': '0.0002408', 'grad_norm': '0.03435', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 8588384, 'tokens/trainable': 130100, 'epoch': '1.109'}
+
 55%|█████▌    | 284/512 [31:54<25:34,  6.73s/it]
 56%|█████▌    | 285/512 [32:01<25:24,  6.72s/it]
                                                 
{'loss': '0.0001933', 'grad_norm': '0.06312', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.49', 'tokens/total': 8618768, 'tokens/trainable': 130543, 'epoch': '1.113'}
+
 56%|█████▌    | 285/512 [32:01<25:24,  6.72s/it]
 56%|█████▌    | 286/512 [32:08<25:17,  6.72s/it]
                                                 
{'loss': '0.0001494', 'grad_norm': '0.01357', 'learning_rate': '5.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 8648976, 'tokens/trainable': 131000, 'epoch': '1.117'}
+
 56%|█████▌    | 286/512 [32:08<25:17,  6.72s/it]
 56%|█████▌    | 287/512 [32:14<25:09,  6.71s/it]
                                                 
{'loss': '9.589e-05', 'grad_norm': '0.008714', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 8679152, 'tokens/trainable': 131449, 'epoch': '1.121'}
+
 56%|█████▌    | 287/512 [32:14<25:09,  6.71s/it]
 56%|█████▋    | 288/512 [32:21<25:00,  6.70s/it]
                                                 
{'loss': '3.912e-05', 'grad_norm': '0.00438', 'learning_rate': '4.964e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '35.97', 'tokens/total': 8709504, 'tokens/trainable': 131923, 'epoch': '1.125'}
+
 56%|█████▋    | 288/512 [32:21<25:00,  6.70s/it][2026-08-18 14:37:20,961] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
+
 56%|█████▋    | 289/512 [32:30<27:12,  7.32s/it]
                                                 
{'loss': '0.003404', 'grad_norm': '0.348', 'learning_rate': '4.935e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '33.15', 'tokens/total': 8739680, 'tokens/trainable': 132397, 'epoch': '1.129'}
+
 56%|█████▋    | 289/512 [32:30<27:12,  7.32s/it]
 57%|█████▋    | 290/512 [32:36<26:24,  7.14s/it]
                                                 
{'loss': '2.929e-05', 'grad_norm': '0.002436', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '36.4', 'tokens/total': 8770112, 'tokens/trainable': 132846, 'epoch': '1.133'}
+
 57%|█████▋    | 290/512 [32:36<26:24,  7.14s/it]
 57%|█████▋    | 291/512 [32:43<25:49,  7.01s/it]
                                                 
{'loss': '0.001151', 'grad_norm': '0.5378', 'learning_rate': '4.878e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '35.74', 'tokens/total': 8800368, 'tokens/trainable': 133347, 'epoch': '1.137'}
+
 57%|█████▋    | 291/512 [32:43<25:49,  7.01s/it]
 57%|█████▋    | 292/512 [32:50<25:20,  6.91s/it]
                                                 
{'loss': '4.125e-05', 'grad_norm': '0.005875', 'learning_rate': '4.849e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 8830688, 'tokens/trainable': 133779, 'epoch': '1.141'}
+
 57%|█████▋    | 292/512 [32:50<25:20,  6.91s/it]
 57%|█████▋    | 293/512 [32:57<25:02,  6.86s/it]
                                                 
{'loss': '0.000114', 'grad_norm': '0.04662', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 8861008, 'tokens/trainable': 134251, 'epoch': '1.145'}
+
 57%|█████▋    | 293/512 [32:57<25:02,  6.86s/it]
 57%|█████▋    | 294/512 [33:03<24:49,  6.83s/it]
                                                 
{'loss': '3.974e-05', 'grad_norm': '0.00466', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '33.55', 'tokens/total': 8891760, 'tokens/trainable': 134683, 'epoch': '1.148'}
+
 57%|█████▋    | 294/512 [33:03<24:49,  6.83s/it]
 58%|█████▊    | 295/512 [33:10<24:35,  6.80s/it]
                                                 
{'loss': '1.725e-05', 'grad_norm': '0.003478', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 8922192, 'tokens/trainable': 135165, 'epoch': '1.152'}
+
 58%|█████▊    | 295/512 [33:10<24:35,  6.80s/it]
 58%|█████▊    | 296/512 [33:17<24:19,  6.75s/it]
                                                 
{'loss': '6.337e-05', 'grad_norm': '0.00705', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '33.24', 'tokens/total': 8952496, 'tokens/trainable': 135599, 'epoch': '1.156'}
+
 58%|█████▊    | 296/512 [33:17<24:19,  6.75s/it]
 58%|█████▊    | 297/512 [33:23<24:07,  6.73s/it]
                                                 
{'loss': '0.0001625', 'grad_norm': '0.03387', 'learning_rate': '4.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.67', 'tokens/total': 8982768, 'tokens/trainable': 136038, 'epoch': '1.16'}
+
 58%|█████▊    | 297/512 [33:23<24:07,  6.73s/it]
 58%|█████▊    | 298/512 [33:30<23:59,  6.73s/it]
                                                 
{'loss': '0.0001614', 'grad_norm': '0.01463', 'learning_rate': '4.677e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 9013248, 'tokens/trainable': 136495, 'epoch': '1.164'}
+
 58%|█████▊    | 298/512 [33:30<23:59,  6.73s/it]
 58%|█████▊    | 299/512 [33:37<23:53,  6.73s/it]
                                                 
{'loss': '2.856e-05', 'grad_norm': '0.002429', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.34', 'tokens/total': 9043632, 'tokens/trainable': 136953, 'epoch': '1.168'}
+
 58%|█████▊    | 299/512 [33:37<23:53,  6.73s/it]
 59%|█████▊    | 300/512 [33:44<23:42,  6.71s/it]
                                                 
{'loss': '0.000198', 'grad_norm': '0.02774', 'learning_rate': '4.62e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 9073936, 'tokens/trainable': 137412, 'epoch': '1.172'}
+
 59%|█████▊    | 300/512 [33:44<23:42,  6.71s/it]
 59%|█████▉    | 301/512 [33:50<23:35,  6.71s/it]
                                                 
{'loss': '0.001319', 'grad_norm': '0.2262', 'learning_rate': '4.592e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.68', 'tokens/total': 9104416, 'tokens/trainable': 137907, 'epoch': '1.176'}
+
 59%|█████▉    | 301/512 [33:50<23:35,  6.71s/it]
 59%|█████▉    | 302/512 [33:57<23:30,  6.71s/it]
                                                 
{'loss': '0.0001245', 'grad_norm': '0.03463', 'learning_rate': '4.563e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.33', 'tokens/total': 9134688, 'tokens/trainable': 138383, 'epoch': '1.18'}
+
 59%|█████▉    | 302/512 [33:57<23:30,  6.71s/it]
 59%|█████▉    | 303/512 [34:04<23:26,  6.73s/it]
                                                 
{'loss': '5.559e-05', 'grad_norm': '0.005657', 'learning_rate': '4.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '36.4', 'tokens/total': 9165088, 'tokens/trainable': 138842, 'epoch': '1.184'}
+
 59%|█████▉    | 303/512 [34:04<23:26,  6.73s/it]
 59%|█████▉    | 304/512 [34:10<23:14,  6.71s/it]
                                                 
{'loss': '1.902e-05', 'grad_norm': '0.007515', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '30.67', 'tokens/total': 9195584, 'tokens/trainable': 139269, 'epoch': '1.188'}
+
 59%|█████▉    | 304/512 [34:10<23:14,  6.71s/it]
 60%|█████▉    | 305/512 [34:17<23:08,  6.71s/it]
                                                 
{'loss': '2.345e-05', 'grad_norm': '0.002446', 'learning_rate': '4.478e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 9225760, 'tokens/trainable': 139736, 'epoch': '1.191'}
+
 60%|█████▉    | 305/512 [34:17<23:08,  6.71s/it]
 60%|█████▉    | 306/512 [34:24<23:01,  6.71s/it]
                                                 
{'loss': '0.0005937', 'grad_norm': '0.1801', 'learning_rate': '4.45e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '37.01', 'tokens/total': 9256352, 'tokens/trainable': 140226, 'epoch': '1.195'}
+
 60%|█████▉    | 306/512 [34:24<23:01,  6.71s/it]
 60%|█████▉    | 307/512 [34:30<22:17,  6.52s/it]
                                                 
{'loss': '2.11e-05', 'grad_norm': '0.001028', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.19', 'memory/max_allocated (GiB)': '33.19', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 9284256, 'tokens/trainable': 140664, 'epoch': '1.199'}
+
 60%|█████▉    | 307/512 [34:30<22:17,  6.52s/it]
 60%|██████    | 308/512 [34:37<22:27,  6.61s/it]
                                                 
{'loss': '5.623e-05', 'grad_norm': '0.005494', 'learning_rate': '4.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '29.84', 'tokens/total': 9314768, 'tokens/trainable': 141115, 'epoch': '1.203'}
+
 60%|██████    | 308/512 [34:37<22:27,  6.61s/it]
 60%|██████    | 309/512 [34:43<22:28,  6.64s/it]
                                                 
{'loss': '1.449e-05', 'grad_norm': '0.0006828', 'learning_rate': '4.366e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '28.87', 'tokens/total': 9344976, 'tokens/trainable': 141530, 'epoch': '1.207'}
+
 60%|██████    | 309/512 [34:43<22:28,  6.64s/it]
 61%|██████    | 310/512 [34:50<22:23,  6.65s/it]
                                                 
{'loss': '0.002186', 'grad_norm': '0.7971', 'learning_rate': '4.338e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 9375280, 'tokens/trainable': 141964, 'epoch': '1.211'}
+
 61%|██████    | 310/512 [34:50<22:23,  6.65s/it]
 61%|██████    | 311/512 [34:57<22:18,  6.66s/it]
                                                 
{'loss': '1.984e-05', 'grad_norm': '0.003865', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 9405536, 'tokens/trainable': 142428, 'epoch': '1.215'}
+
 61%|██████    | 311/512 [34:57<22:18,  6.66s/it]
 61%|██████    | 312/512 [35:03<22:10,  6.65s/it]
                                                 
{'loss': '0.000209', 'grad_norm': '0.05813', 'learning_rate': '4.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '38.16', 'tokens/total': 9435696, 'tokens/trainable': 142894, 'epoch': '1.219'}
+
 61%|██████    | 312/512 [35:03<22:10,  6.65s/it]
 61%|██████    | 313/512 [35:10<22:07,  6.67s/it]
                                                 
{'loss': '0.000148', 'grad_norm': '0.01986', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.37', 'tokens/total': 9466080, 'tokens/trainable': 143368, 'epoch': '1.223'}
+
 61%|██████    | 313/512 [35:10<22:07,  6.67s/it]
 61%|██████▏   | 314/512 [35:17<22:01,  6.68s/it]
                                                 
{'loss': '0.0001475', 'grad_norm': '0.05822', 'learning_rate': '4.226e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 9496400, 'tokens/trainable': 143824, 'epoch': '1.227'}
+
 61%|██████▏   | 314/512 [35:17<22:01,  6.68s/it]
 62%|██████▏   | 315/512 [35:24<22:02,  6.71s/it]
                                                 
{'loss': '0.0001151', 'grad_norm': '0.009975', 'learning_rate': '4.198e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.58', 'tokens/total': 9526880, 'tokens/trainable': 144306, 'epoch': '1.23'}
+
 62%|██████▏   | 315/512 [35:24<22:02,  6.71s/it]
 62%|██████▏   | 316/512 [35:30<21:54,  6.71s/it]
                                                 
{'loss': '6.155e-05', 'grad_norm': '0.00919', 'learning_rate': '4.17e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 9557280, 'tokens/trainable': 144763, 'epoch': '1.234'}
+
 62%|██████▏   | 316/512 [35:30<21:54,  6.71s/it]
 62%|██████▏   | 317/512 [35:37<21:49,  6.71s/it]
                                                 
{'loss': '2.067e-05', 'grad_norm': '0.001632', 'learning_rate': '4.143e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.45', 'tokens/total': 9587712, 'tokens/trainable': 145255, 'epoch': '1.238'}
+
 62%|██████▏   | 317/512 [35:37<21:49,  6.71s/it]
 62%|██████▏   | 318/512 [35:44<21:41,  6.71s/it]
                                                 
{'loss': '5.585e-05', 'grad_norm': '0.007459', 'learning_rate': '4.115e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 9618064, 'tokens/trainable': 145714, 'epoch': '1.242'}
+
 62%|██████▏   | 318/512 [35:44<21:41,  6.71s/it]
 62%|██████▏   | 319/512 [35:50<21:35,  6.71s/it]
                                                 
{'loss': '0.0004338', 'grad_norm': '0.07343', 'learning_rate': '4.087e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.04', 'tokens/total': 9648336, 'tokens/trainable': 146200, 'epoch': '1.246'}
+
 62%|██████▏   | 319/512 [35:50<21:35,  6.71s/it]
 62%|██████▎   | 320/512 [35:57<21:28,  6.71s/it]
                                                 
{'loss': '4.525e-05', 'grad_norm': '0.007983', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 9678688, 'tokens/trainable': 146668, 'epoch': '1.25'}
+
 62%|██████▎   | 320/512 [35:57<21:28,  6.71s/it][2026-08-18 14:40:57,132] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
+
 63%|██████▎   | 321/512 [36:05<22:54,  7.20s/it]
                                                 
{'loss': '6.388e-05', 'grad_norm': '0.016', 'learning_rate': '4.032e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 9708848, 'tokens/trainable': 147114, 'epoch': '1.254'}
+
 63%|██████▎   | 321/512 [36:05<22:54,  7.20s/it]
 63%|██████▎   | 322/512 [36:12<22:18,  7.04s/it]
                                                 
{'loss': '2.179e-05', 'grad_norm': '0.004208', 'learning_rate': '4.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 9739360, 'tokens/trainable': 147564, 'epoch': '1.258'}
+
 63%|██████▎   | 322/512 [36:12<22:18,  7.04s/it]
 63%|██████▎   | 323/512 [36:19<21:51,  6.94s/it]
                                                 
{'loss': '3.835e-05', 'grad_norm': '0.00943', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.24', 'tokens/total': 9769776, 'tokens/trainable': 148013, 'epoch': '1.262'}
+
 63%|██████▎   | 323/512 [36:19<21:51,  6.94s/it]
 63%|██████▎   | 324/512 [36:26<21:35,  6.89s/it]
                                                 
{'loss': '0.0001877', 'grad_norm': '0.05088', 'learning_rate': '3.95e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '38.06', 'tokens/total': 9800096, 'tokens/trainable': 148470, 'epoch': '1.266'}
+
 63%|██████▎   | 324/512 [36:26<21:35,  6.89s/it]
 63%|██████▎   | 325/512 [36:32<21:15,  6.82s/it]
                                                 
{'loss': '0.01083', 'grad_norm': '0.9416', 'learning_rate': '3.923e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 9830304, 'tokens/trainable': 148926, 'epoch': '1.27'}
+
 63%|██████▎   | 325/512 [36:32<21:15,  6.82s/it]
 64%|██████▎   | 326/512 [36:39<21:04,  6.80s/it]
                                                 
{'loss': '8.612e-06', 'grad_norm': '0.0003319', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.82', 'tokens/total': 9860608, 'tokens/trainable': 149420, 'epoch': '1.273'}
+
 64%|██████▎   | 326/512 [36:39<21:04,  6.80s/it]
 64%|██████▍   | 327/512 [36:46<20:52,  6.77s/it]
                                                 
{'loss': '1.583e-05', 'grad_norm': '0.0005477', 'learning_rate': '3.869e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '39.41', 'tokens/total': 9890560, 'tokens/trainable': 149873, 'epoch': '1.277'}
+
 64%|██████▍   | 327/512 [36:46<20:52,  6.77s/it]
 64%|██████▍   | 328/512 [36:52<20:43,  6.76s/it]
                                                 
{'loss': '1.643e-05', 'grad_norm': '0.000826', 'learning_rate': '3.842e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 9920880, 'tokens/trainable': 150347, 'epoch': '1.281'}
+
 64%|██████▍   | 328/512 [36:52<20:43,  6.76s/it]
 64%|██████▍   | 329/512 [36:59<20:33,  6.74s/it]
                                                 
{'loss': '1.752e-05', 'grad_norm': '0.001146', 'learning_rate': '3.815e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 9951280, 'tokens/trainable': 150822, 'epoch': '1.285'}
+
 64%|██████▍   | 329/512 [36:59<20:33,  6.74s/it]
 64%|██████▍   | 330/512 [37:06<20:26,  6.74s/it]
                                                 
{'loss': '2.001e-05', 'grad_norm': '0.002252', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 9981600, 'tokens/trainable': 151266, 'epoch': '1.289'}
+
 64%|██████▍   | 330/512 [37:06<20:26,  6.74s/it]
 65%|██████▍   | 331/512 [37:13<20:17,  6.73s/it]
                                                 
{'loss': '0.0003773', 'grad_norm': '0.08719', 'learning_rate': '3.761e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.15', 'tokens/total': 10012016, 'tokens/trainable': 151734, 'epoch': '1.293'}
+
 65%|██████▍   | 331/512 [37:13<20:17,  6.73s/it]
 65%|██████▍   | 332/512 [37:19<20:08,  6.71s/it]
                                                 
{'loss': '1.175e-05', 'grad_norm': '0.0008834', 'learning_rate': '3.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.01', 'tokens/total': 10042448, 'tokens/trainable': 152184, 'epoch': '1.297'}
+
 65%|██████▍   | 332/512 [37:19<20:08,  6.71s/it]
 65%|██████▌   | 333/512 [37:26<20:01,  6.71s/it]
                                                 
{'loss': '5.219e-05', 'grad_norm': '0.01143', 'learning_rate': '3.708e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.85', 'tokens/total': 10073024, 'tokens/trainable': 152623, 'epoch': '1.301'}
+
 65%|██████▌   | 333/512 [37:26<20:01,  6.71s/it]
 65%|██████▌   | 334/512 [37:33<19:51,  6.69s/it]
                                                 
{'loss': '0.003487', 'grad_norm': '0.5781', 'learning_rate': '3.681e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.44', 'tokens/total': 10103504, 'tokens/trainable': 153040, 'epoch': '1.305'}
+
 65%|██████▌   | 334/512 [37:33<19:51,  6.69s/it]
 65%|██████▌   | 335/512 [37:39<19:18,  6.54s/it]
                                                 
{'loss': '0.004716', 'grad_norm': '0.7214', 'learning_rate': '3.655e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.78', 'tokens/total': 10131632, 'tokens/trainable': 153463, 'epoch': '1.309'}
+
 65%|██████▌   | 335/512 [37:39<19:18,  6.54s/it]
 66%|██████▌   | 336/512 [37:45<19:18,  6.58s/it]
                                                 
{'loss': '0.0002346', 'grad_norm': '0.03132', 'learning_rate': '3.628e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 10161824, 'tokens/trainable': 153908, 'epoch': '1.312'}
+
 66%|██████▌   | 336/512 [37:45<19:18,  6.58s/it]
 66%|██████▌   | 337/512 [37:52<19:22,  6.64s/it]
                                                 
{'loss': '0.0001198', 'grad_norm': '0.03333', 'learning_rate': '3.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 10192448, 'tokens/trainable': 154387, 'epoch': '1.316'}
+
 66%|██████▌   | 337/512 [37:52<19:22,  6.64s/it]
 66%|██████▌   | 338/512 [37:59<19:20,  6.67s/it]
                                                 
{'loss': '3.549e-05', 'grad_norm': '0.004504', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.94', 'tokens/total': 10222704, 'tokens/trainable': 154836, 'epoch': '1.32'}
+
 66%|██████▌   | 338/512 [37:59<19:20,  6.67s/it]
 66%|██████▌   | 339/512 [38:06<19:19,  6.70s/it]
                                                 
{'loss': '1.789e-05', 'grad_norm': '0.001243', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.81', 'tokens/total': 10253168, 'tokens/trainable': 155261, 'epoch': '1.324'}
+
 66%|██████▌   | 339/512 [38:06<19:19,  6.70s/it]
 66%|██████▋   | 340/512 [38:12<19:08,  6.68s/it]
                                                 
{'loss': '0.00279', 'grad_norm': '0.3251', 'learning_rate': '3.523e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.24', 'tokens/total': 10283552, 'tokens/trainable': 155674, 'epoch': '1.328'}
+
 66%|██████▋   | 340/512 [38:12<19:08,  6.68s/it]
 67%|██████▋   | 341/512 [38:19<19:05,  6.70s/it]
                                                 
{'loss': '0.0001325', 'grad_norm': '0.03649', 'learning_rate': '3.497e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.78', 'tokens/total': 10313872, 'tokens/trainable': 156164, 'epoch': '1.332'}
+
 67%|██████▋   | 341/512 [38:19<19:05,  6.70s/it]
 67%|██████▋   | 342/512 [38:26<18:57,  6.69s/it]
                                                 
{'loss': '0.002143', 'grad_norm': '0.2706', 'learning_rate': '3.471e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.65', 'tokens/total': 10344256, 'tokens/trainable': 156611, 'epoch': '1.336'}
+
 67%|██████▋   | 342/512 [38:26<18:57,  6.69s/it]
 67%|██████▋   | 343/512 [38:33<18:50,  6.69s/it]
                                                 
{'loss': '0.0001378', 'grad_norm': '0.01915', 'learning_rate': '3.445e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.68', 'tokens/total': 10374544, 'tokens/trainable': 157073, 'epoch': '1.34'}
+
 67%|██████▋   | 343/512 [38:33<18:50,  6.69s/it]
 67%|██████▋   | 344/512 [38:39<18:45,  6.70s/it]
                                                 
{'loss': '5.135e-05', 'grad_norm': '0.0254', 'learning_rate': '3.42e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 10405040, 'tokens/trainable': 157523, 'epoch': '1.344'}
+
 67%|██████▋   | 344/512 [38:39<18:45,  6.70s/it]
 67%|██████▋   | 345/512 [38:46<18:37,  6.69s/it]
                                                 
{'loss': '0.009803', 'grad_norm': '0.3926', 'learning_rate': '3.394e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.16', 'tokens/total': 10435424, 'tokens/trainable': 157964, 'epoch': '1.348'}
+
 67%|██████▋   | 345/512 [38:46<18:37,  6.69s/it]
 68%|██████▊   | 346/512 [38:53<18:30,  6.69s/it]
                                                 
{'loss': '2.997e-05', 'grad_norm': '0.002907', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '30.44', 'tokens/total': 10465760, 'tokens/trainable': 158390, 'epoch': '1.352'}
+
 68%|██████▊   | 346/512 [38:53<18:30,  6.69s/it]
 68%|██████▊   | 347/512 [38:59<18:20,  6.67s/it]
                                                 
{'loss': '7.613e-05', 'grad_norm': '0.02028', 'learning_rate': '3.343e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 10495904, 'tokens/trainable': 158848, 'epoch': '1.355'}
+
 68%|██████▊   | 347/512 [38:59<18:20,  6.67s/it]
 68%|██████▊   | 348/512 [39:06<18:18,  6.70s/it]
                                                 
{'loss': '0.001124', 'grad_norm': '0.1487', 'learning_rate': '3.317e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.71', 'tokens/total': 10526624, 'tokens/trainable': 159270, 'epoch': '1.359'}
+
 68%|██████▊   | 348/512 [39:06<18:18,  6.70s/it]
 68%|██████▊   | 349/512 [39:13<18:10,  6.69s/it]
                                                 
{'loss': '7.921e-05', 'grad_norm': '0.01538', 'learning_rate': '3.292e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.51', 'tokens/total': 10556752, 'tokens/trainable': 159720, 'epoch': '1.363'}
+
 68%|██████▊   | 349/512 [39:13<18:10,  6.69s/it]
 68%|██████▊   | 350/512 [39:19<18:09,  6.73s/it]
                                                 
{'loss': '1.457e-05', 'grad_norm': '0.00237', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.14', 'tokens/total': 10587264, 'tokens/trainable': 160187, 'epoch': '1.367'}
+
 68%|██████▊   | 350/512 [39:19<18:09,  6.73s/it]
 69%|██████▊   | 351/512 [39:26<18:02,  6.72s/it]
                                                 
{'loss': '6.53e-05', 'grad_norm': '0.009544', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 10617504, 'tokens/trainable': 160645, 'epoch': '1.371'}
+
 69%|██████▊   | 351/512 [39:26<18:02,  6.72s/it]
 69%|██████▉   | 352/512 [39:33<17:54,  6.72s/it]
                                                 
{'loss': '0.01059', 'grad_norm': '0.6505', 'learning_rate': '3.217e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.49', 'tokens/total': 10647760, 'tokens/trainable': 161102, 'epoch': '1.375'}
+
 69%|██████▉   | 352/512 [39:33<17:54,  6.72s/it][2026-08-18 14:44:32,873] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
+
 69%|██████▉   | 353/512 [39:41<19:05,  7.21s/it]
                                                 
{'loss': '0.000143', 'grad_norm': '0.02473', 'learning_rate': '3.192e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.93', 'tokens/total': 10678080, 'tokens/trainable': 161555, 'epoch': '1.379'}
+
 69%|██████▉   | 353/512 [39:41<19:05,  7.21s/it]
 69%|██████▉   | 354/512 [39:48<18:34,  7.05s/it]
                                                 
{'loss': '7.214e-05', 'grad_norm': '0.03731', 'learning_rate': '3.167e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.41', 'tokens/train_per_sec_per_gpu': '36.75', 'tokens/total': 10708544, 'tokens/trainable': 162020, 'epoch': '1.383'}
+
 69%|██████▉   | 354/512 [39:48<18:34,  7.05s/it]
 69%|██████▉   | 355/512 [39:55<18:10,  6.95s/it]
                                                 
{'loss': '0.0001183', 'grad_norm': '0.00885', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 10739024, 'tokens/trainable': 162448, 'epoch': '1.387'}
+
 69%|██████▉   | 355/512 [39:55<18:10,  6.95s/it]
 70%|██████▉   | 356/512 [40:01<17:48,  6.85s/it]
                                                 
{'loss': '0.000361', 'grad_norm': '0.06792', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.61', 'memory/max_allocated (GiB)': '33.61', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '28.26', 'tokens/total': 10769024, 'tokens/trainable': 162863, 'epoch': '1.391'}
+
 70%|██████▉   | 356/512 [40:01<17:48,  6.85s/it]
 70%|██████▉   | 357/512 [40:08<17:35,  6.81s/it]
                                                 
{'loss': '3.174e-05', 'grad_norm': '0.001951', 'learning_rate': '3.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 10799328, 'tokens/trainable': 163318, 'epoch': '1.395'}
+
 70%|██████▉   | 357/512 [40:08<17:35,  6.81s/it]
 70%|██████▉   | 358/512 [40:15<17:23,  6.78s/it]
                                                 
{'loss': '0.01027', 'grad_norm': '0.714', 'learning_rate': '3.068e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 10829488, 'tokens/trainable': 163767, 'epoch': '1.398'}
+
 70%|██████▉   | 358/512 [40:15<17:23,  6.78s/it]
 70%|███████   | 359/512 [40:21<17:12,  6.75s/it]
                                                 
{'loss': '4.873e-05', 'grad_norm': '0.006852', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 10859936, 'tokens/trainable': 164208, 'epoch': '1.402'}
+
 70%|███████   | 359/512 [40:21<17:12,  6.75s/it]
 70%|███████   | 360/512 [40:28<17:03,  6.73s/it]
                                                 
{'loss': '3.888e-05', 'grad_norm': '0.001825', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 10890400, 'tokens/trainable': 164655, 'epoch': '1.406'}
+
 70%|███████   | 360/512 [40:28<17:03,  6.73s/it]
 71%|███████   | 361/512 [40:35<16:52,  6.71s/it]
                                                 
{'loss': '0.0003688', 'grad_norm': '0.04369', 'learning_rate': '2.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 10920624, 'tokens/trainable': 165100, 'epoch': '1.41'}
+
 71%|███████   | 361/512 [40:35<16:52,  6.71s/it]
 71%|███████   | 362/512 [40:41<16:48,  6.72s/it]
                                                 
{'loss': '0.0003961', 'grad_norm': '0.2674', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 10951136, 'tokens/trainable': 165587, 'epoch': '1.414'}
+
 71%|███████   | 362/512 [40:41<16:48,  6.72s/it]
 71%|███████   | 363/512 [40:48<16:38,  6.70s/it]
                                                 
{'loss': '0.0007331', 'grad_norm': '0.09645', 'learning_rate': '2.947e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 10981344, 'tokens/trainable': 166034, 'epoch': '1.418'}
+
 71%|███████   | 363/512 [40:48<16:38,  6.70s/it]
 71%|███████   | 364/512 [40:55<16:32,  6.71s/it]
                                                 
{'loss': '0.0005385', 'grad_norm': '0.05368', 'learning_rate': '2.924e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 11011552, 'tokens/trainable': 166485, 'epoch': '1.422'}
+
 71%|███████   | 364/512 [40:55<16:32,  6.71s/it]
 71%|███████▏  | 365/512 [41:02<16:26,  6.71s/it]
                                                 
{'loss': '0.0006869', 'grad_norm': '0.1007', 'learning_rate': '2.9e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '36.36', 'tokens/total': 11041872, 'tokens/trainable': 166973, 'epoch': '1.426'}
+
 71%|███████▏  | 365/512 [41:02<16:26,  6.71s/it]
 71%|███████▏  | 366/512 [41:08<16:19,  6.71s/it]
                                                 
{'loss': '0.0002465', 'grad_norm': '0.02698', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '29.12', 'tokens/total': 11072192, 'tokens/trainable': 167414, 'epoch': '1.43'}
+
 71%|███████▏  | 366/512 [41:08<16:19,  6.71s/it]
 72%|███████▏  | 367/512 [41:15<16:12,  6.71s/it]
                                                 
{'loss': '5.263e-05', 'grad_norm': '0.002621', 'learning_rate': '2.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.23', 'tokens/total': 11102736, 'tokens/trainable': 167851, 'epoch': '1.434'}
+
 72%|███████▏  | 367/512 [41:15<16:12,  6.71s/it]
 72%|███████▏  | 368/512 [41:22<16:05,  6.70s/it]
                                                 
{'loss': '2.951e-05', 'grad_norm': '0.001814', 'learning_rate': '2.829e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.56', 'tokens/total': 11133056, 'tokens/trainable': 168294, 'epoch': '1.438'}
+
 72%|███████▏  | 368/512 [41:22<16:05,  6.70s/it]
 72%|███████▏  | 369/512 [41:28<16:01,  6.72s/it]
                                                 
{'loss': '7.607e-05', 'grad_norm': '0.01205', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.23', 'tokens/total': 11163568, 'tokens/trainable': 168769, 'epoch': '1.441'}
+
 72%|███████▏  | 369/512 [41:28<16:01,  6.72s/it]
 72%|███████▏  | 370/512 [41:35<15:55,  6.73s/it]
                                                 
{'loss': '0.0001533', 'grad_norm': '0.01345', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.83', 'tokens/total': 11193936, 'tokens/trainable': 169270, 'epoch': '1.445'}
+
 72%|███████▏  | 370/512 [41:35<15:55,  6.73s/it]
 72%|███████▏  | 371/512 [41:42<15:51,  6.74s/it]
                                                 
{'loss': '5.633e-05', 'grad_norm': '0.003458', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 11224480, 'tokens/trainable': 169711, 'epoch': '1.449'}
+
 72%|███████▏  | 371/512 [41:42<15:51,  6.74s/it]
 73%|███████▎  | 372/512 [41:49<15:44,  6.75s/it]
                                                 
{'loss': '9.535e-05', 'grad_norm': '0.006362', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 11254912, 'tokens/trainable': 170185, 'epoch': '1.453'}
+
 73%|███████▎  | 372/512 [41:49<15:44,  6.75s/it]
 73%|███████▎  | 373/512 [41:55<15:37,  6.74s/it]
                                                 
{'loss': '0.0006136', 'grad_norm': '0.04559', 'learning_rate': '2.714e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 11285104, 'tokens/trainable': 170650, 'epoch': '1.457'}
+
 73%|███████▎  | 373/512 [41:55<15:37,  6.74s/it]
 73%|███████▎  | 374/512 [42:02<15:29,  6.74s/it]
                                                 
{'loss': '0.0002039', 'grad_norm': '0.01487', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '31.99', 'tokens/total': 11315600, 'tokens/trainable': 171126, 'epoch': '1.461'}
+
 73%|███████▎  | 374/512 [42:02<15:29,  6.74s/it]
 73%|███████▎  | 375/512 [42:09<15:19,  6.71s/it]
                                                 
{'loss': '0.001035', 'grad_norm': '0.05874', 'learning_rate': '2.668e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 11345776, 'tokens/trainable': 171600, 'epoch': '1.465'}
+
 73%|███████▎  | 375/512 [42:09<15:19,  6.71s/it]
 73%|███████▎  | 376/512 [42:15<15:11,  6.70s/it]
                                                 
{'loss': '0.003572', 'grad_norm': '0.2411', 'learning_rate': '2.646e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '28.14', 'tokens/total': 11376208, 'tokens/trainable': 172017, 'epoch': '1.469'}
+
 73%|███████▎  | 376/512 [42:15<15:11,  6.70s/it]
 74%|███████▎  | 377/512 [42:22<15:05,  6.71s/it]
                                                 
{'loss': '0.0007833', 'grad_norm': '0.08794', 'learning_rate': '2.624e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '36.07', 'tokens/total': 11406512, 'tokens/trainable': 172493, 'epoch': '1.473'}
+
 74%|███████▎  | 377/512 [42:22<15:05,  6.71s/it]
 74%|███████▍  | 378/512 [42:29<14:57,  6.70s/it]
                                                 
{'loss': '0.0001175', 'grad_norm': '0.009155', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.4', 'tokens/total': 11436544, 'tokens/trainable': 172951, 'epoch': '1.477'}
+
 74%|███████▍  | 378/512 [42:29<14:57,  6.70s/it]
 74%|███████▍  | 379/512 [42:36<14:50,  6.70s/it]
                                                 
{'loss': '0.0001157', 'grad_norm': '0.01292', 'learning_rate': '2.579e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.72', 'tokens/total': 11467008, 'tokens/trainable': 173435, 'epoch': '1.48'}
+
 74%|███████▍  | 379/512 [42:36<14:50,  6.70s/it]
 74%|███████▍  | 380/512 [42:42<14:44,  6.70s/it]
                                                 
{'loss': '8.939e-05', 'grad_norm': '0.008967', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.47', 'tokens/total': 11497184, 'tokens/trainable': 173900, 'epoch': '1.484'}
+
 74%|███████▍  | 380/512 [42:42<14:44,  6.70s/it]
 74%|███████▍  | 381/512 [42:49<14:39,  6.71s/it]
                                                 
{'loss': '0.0001308', 'grad_norm': '0.03182', 'learning_rate': '2.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 11527600, 'tokens/trainable': 174371, 'epoch': '1.488'}
+
 74%|███████▍  | 381/512 [42:49<14:39,  6.71s/it]
 75%|███████▍  | 382/512 [42:56<14:32,  6.71s/it]
                                                 
{'loss': '2.56e-05', 'grad_norm': '0.0009073', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 11557808, 'tokens/trainable': 174826, 'epoch': '1.492'}
+
 75%|███████▍  | 382/512 [42:56<14:32,  6.71s/it]
 75%|███████▍  | 383/512 [43:02<14:26,  6.71s/it]
                                                 
{'loss': '5.779e-05', 'grad_norm': '0.00506', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 11588352, 'tokens/trainable': 175276, 'epoch': '1.496'}
+
 75%|███████▍  | 383/512 [43:02<14:26,  6.71s/it]
 75%|███████▌  | 384/512 [43:09<14:20,  6.72s/it]
                                                 
{'loss': '0.0001251', 'grad_norm': '0.008305', 'learning_rate': '2.47e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.71', 'tokens/total': 11618640, 'tokens/trainable': 175720, 'epoch': '1.5'}
+
 75%|███████▌  | 384/512 [43:09<14:20,  6.72s/it][2026-08-18 14:48:09,177] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
+
 75%|███████▌  | 385/512 [43:18<15:20,  7.25s/it]
                                                 
{'loss': '4.852e-05', 'grad_norm': '0.003528', 'learning_rate': '2.449e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 11649072, 'tokens/trainable': 176203, 'epoch': '1.504'}
+
 75%|███████▌  | 385/512 [43:18<15:20,  7.25s/it]
 75%|███████▌  | 386/512 [43:24<14:53,  7.09s/it]
                                                 
{'loss': '3.331e-05', 'grad_norm': '0.001593', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 11679552, 'tokens/trainable': 176651, 'epoch': '1.508'}
+
 75%|███████▌  | 386/512 [43:24<14:53,  7.09s/it]
 76%|███████▌  | 387/512 [43:31<14:32,  6.98s/it]
                                                 
{'loss': '3.345e-05', 'grad_norm': '0.002167', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 11709760, 'tokens/trainable': 177115, 'epoch': '1.512'}
+
 76%|███████▌  | 387/512 [43:31<14:32,  6.98s/it]
 76%|███████▌  | 388/512 [43:38<14:15,  6.90s/it]
                                                 
{'loss': '5.528e-05', 'grad_norm': '0.003448', 'learning_rate': '2.385e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.13', 'tokens/total': 11740288, 'tokens/trainable': 177601, 'epoch': '1.516'}
+
 76%|███████▌  | 388/512 [43:38<14:15,  6.90s/it]
 76%|███████▌  | 389/512 [43:45<14:01,  6.84s/it]
                                                 
{'loss': '4.562e-05', 'grad_norm': '0.005298', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 11770592, 'tokens/trainable': 178050, 'epoch': '1.52'}
+
 76%|███████▌  | 389/512 [43:45<14:01,  6.84s/it]
 76%|███████▌  | 390/512 [43:51<13:50,  6.81s/it]
                                                 
{'loss': '3.829e-05', 'grad_norm': '0.003032', 'learning_rate': '2.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.42', 'tokens/total': 11800752, 'tokens/trainable': 178531, 'epoch': '1.523'}
+
 76%|███████▌  | 390/512 [43:51<13:50,  6.81s/it]
 76%|███████▋  | 391/512 [43:58<13:41,  6.79s/it]
                                                 
{'loss': '0.0003605', 'grad_norm': '0.03543', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 11831360, 'tokens/trainable': 179000, 'epoch': '1.527'}
+
 76%|███████▋  | 391/512 [43:58<13:41,  6.79s/it]
 77%|███████▋  | 392/512 [44:05<13:29,  6.74s/it]
                                                 
{'loss': '7.091e-05', 'grad_norm': '0.004966', 'learning_rate': '2.303e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 11861552, 'tokens/trainable': 179457, 'epoch': '1.531'}
+
 77%|███████▋  | 392/512 [44:05<13:29,  6.74s/it]
 77%|███████▋  | 393/512 [44:11<13:19,  6.71s/it]
                                                 
{'loss': '9.158e-05', 'grad_norm': '0.006339', 'learning_rate': '2.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 11891904, 'tokens/trainable': 179890, 'epoch': '1.535'}
+
 77%|███████▋  | 393/512 [44:11<13:19,  6.71s/it]
 77%|███████▋  | 394/512 [44:18<13:11,  6.71s/it]
                                                 
{'loss': '0.0003821', 'grad_norm': '0.03921', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 11922208, 'tokens/trainable': 180341, 'epoch': '1.539'}
+
 77%|███████▋  | 394/512 [44:18<13:11,  6.71s/it]
 77%|███████▋  | 395/512 [44:25<13:02,  6.69s/it]
                                                 
{'loss': '0.0006777', 'grad_norm': '0.08641', 'learning_rate': '2.242e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 11952672, 'tokens/trainable': 180762, 'epoch': '1.543'}
+
 77%|███████▋  | 395/512 [44:25<13:02,  6.69s/it]
 77%|███████▋  | 396/512 [44:31<12:56,  6.69s/it]
                                                 
{'loss': '0.0007017', 'grad_norm': '0.08662', 'learning_rate': '2.222e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 11983088, 'tokens/trainable': 181226, 'epoch': '1.547'}
+
 77%|███████▋  | 396/512 [44:31<12:56,  6.69s/it]
 78%|███████▊  | 397/512 [44:38<12:50,  6.70s/it]
                                                 
{'loss': '0.0001112', 'grad_norm': '0.01007', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 12013488, 'tokens/trainable': 181714, 'epoch': '1.551'}
+
 78%|███████▊  | 397/512 [44:38<12:50,  6.70s/it]
 78%|███████▊  | 398/512 [44:45<12:42,  6.69s/it]
                                                 
{'loss': '0.0002486', 'grad_norm': '0.03159', 'learning_rate': '2.183e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 12043968, 'tokens/trainable': 182142, 'epoch': '1.555'}
+
 78%|███████▊  | 398/512 [44:45<12:42,  6.69s/it]
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
                                                 
{'loss': '2.122e-05', 'grad_norm': '0.0008732', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.86', 'tokens/total': 12074160, 'tokens/trainable': 182604, 'epoch': '1.559'}
+
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
 78%|███████▊  | 400/512 [44:58<12:29,  6.69s/it]
                                                 
{'loss': '0.0001349', 'grad_norm': '0.009256', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '25.31', 'tokens/total': 12104352, 'tokens/trainable': 183011, 'epoch': '1.562'}
+
 78%|███████▊  | 400/512 [44:58<12:29,  6.69s/it]
 78%|███████▊  | 401/512 [45:05<12:22,  6.68s/it]
                                                 
{'loss': '7.793e-05', 'grad_norm': '0.01052', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.81', 'tokens/total': 12134240, 'tokens/trainable': 183459, 'epoch': '1.566'}
+
 78%|███████▊  | 401/512 [45:05<12:22,  6.68s/it]
 79%|███████▊  | 402/512 [45:12<12:17,  6.70s/it]
                                                 
{'loss': '2.772e-05', 'grad_norm': '0.001341', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.7', 'tokens/total': 12164480, 'tokens/trainable': 183959, 'epoch': '1.57'}
+
 79%|███████▊  | 402/512 [45:12<12:17,  6.70s/it]
 79%|███████▊  | 403/512 [45:18<12:11,  6.71s/it]
                                                 
{'loss': '0.001414', 'grad_norm': '0.1036', 'learning_rate': '2.086e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.5', 'tokens/total': 12194640, 'tokens/trainable': 184431, 'epoch': '1.574'}
+
 79%|███████▊  | 403/512 [45:18<12:11,  6.71s/it]
 79%|███████▉  | 404/512 [45:25<12:05,  6.72s/it]
                                                 
{'loss': '4.145e-05', 'grad_norm': '0.002605', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.37', 'tokens/total': 12224960, 'tokens/trainable': 184867, 'epoch': '1.578'}
+
 79%|███████▉  | 404/512 [45:25<12:05,  6.72s/it]
 79%|███████▉  | 405/512 [45:32<11:56,  6.70s/it]
                                                 
{'loss': '7.322e-05', 'grad_norm': '0.009412', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.37', 'tokens/total': 12255008, 'tokens/trainable': 185300, 'epoch': '1.582'}
+
 79%|███████▉  | 405/512 [45:32<11:56,  6.70s/it]
 79%|███████▉  | 406/512 [45:38<11:49,  6.70s/it]
                                                 
{'loss': '7.983e-05', 'grad_norm': '0.006792', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 12285248, 'tokens/trainable': 185754, 'epoch': '1.586'}
+
 79%|███████▉  | 406/512 [45:38<11:49,  6.70s/it]
 79%|███████▉  | 407/512 [45:45<11:41,  6.69s/it]
                                                 
{'loss': '3.743e-05', 'grad_norm': '0.01302', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 12315552, 'tokens/trainable': 186213, 'epoch': '1.59'}
+
 79%|███████▉  | 407/512 [45:45<11:41,  6.69s/it]
 80%|███████▉  | 408/512 [45:52<11:36,  6.69s/it]
                                                 
{'loss': '2.307e-05', 'grad_norm': '0.001309', 'learning_rate': '1.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 12345904, 'tokens/trainable': 186659, 'epoch': '1.594'}
+
 80%|███████▉  | 408/512 [45:52<11:36,  6.69s/it]
 80%|███████▉  | 409/512 [45:58<11:29,  6.70s/it]
                                                 
{'loss': '2.135e-05', 'grad_norm': '0.001227', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 12376336, 'tokens/trainable': 187083, 'epoch': '1.598'}
+
 80%|███████▉  | 409/512 [45:58<11:29,  6.70s/it]
 80%|████████  | 410/512 [46:05<11:23,  6.70s/it]
                                                 
{'loss': '3.752e-05', 'grad_norm': '0.01123', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 12406880, 'tokens/trainable': 187530, 'epoch': '1.602'}
+
 80%|████████  | 410/512 [46:05<11:23,  6.70s/it]
 80%|████████  | 411/512 [46:12<11:16,  6.70s/it]
                                                 
{'loss': '5.565e-05', 'grad_norm': '0.004873', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 12437088, 'tokens/trainable': 187972, 'epoch': '1.605'}
+
 80%|████████  | 411/512 [46:12<11:16,  6.70s/it]
 80%|████████  | 412/512 [46:18<10:53,  6.54s/it]
                                                 
{'loss': '0.0008922', 'grad_norm': '0.1939', 'learning_rate': '1.922e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '39.95', 'tokens/total': 12465440, 'tokens/trainable': 188417, 'epoch': '1.609'}
+
 80%|████████  | 412/512 [46:18<10:53,  6.54s/it]
 81%|████████  | 413/512 [46:25<10:53,  6.60s/it]
                                                 
{'loss': '0.0006304', 'grad_norm': '0.09062', 'learning_rate': '1.904e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 12495968, 'tokens/trainable': 188868, 'epoch': '1.613'}
+
 81%|████████  | 413/512 [46:25<10:53,  6.60s/it]
 81%|████████  | 414/512 [46:31<10:50,  6.64s/it]
                                                 
{'loss': '6.724e-05', 'grad_norm': '0.009725', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.95', 'tokens/total': 12526128, 'tokens/trainable': 189337, 'epoch': '1.617'}
+
 81%|████████  | 414/512 [46:31<10:50,  6.64s/it]
 81%|████████  | 415/512 [46:38<10:43,  6.64s/it]
                                                 
{'loss': '0.001907', 'grad_norm': '0.1954', 'learning_rate': '1.87e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 12556256, 'tokens/trainable': 189750, 'epoch': '1.621'}
+
 81%|████████  | 415/512 [46:38<10:43,  6.64s/it]
 81%|████████▏ | 416/512 [46:45<10:39,  6.66s/it]
                                                 
{'loss': '0.0005661', 'grad_norm': '0.04653', 'learning_rate': '1.853e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 12586736, 'tokens/trainable': 190216, 'epoch': '1.625'}
+
 81%|████████▏ | 416/512 [46:45<10:39,  6.66s/it][2026-08-18 14:51:44,795] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
+
 81%|████████▏ | 417/512 [46:53<11:22,  7.19s/it]
                                                 
{'loss': '1.887e-05', 'grad_norm': '0.001295', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 12617168, 'tokens/trainable': 190708, 'epoch': '1.629'}
+
 81%|████████▏ | 417/512 [46:53<11:22,  7.19s/it]
 82%|████████▏ | 418/512 [47:00<10:59,  7.02s/it]
                                                 
{'loss': '3.221e-05', 'grad_norm': '0.001875', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.9', 'tokens/total': 12647680, 'tokens/trainable': 191126, 'epoch': '1.633'}
+
 82%|████████▏ | 418/512 [47:00<10:59,  7.02s/it]
 82%|████████▏ | 419/512 [47:06<10:42,  6.91s/it]
                                                 
{'loss': '2.656e-05', 'grad_norm': '0.002248', 'learning_rate': '1.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.34', 'tokens/total': 12677952, 'tokens/trainable': 191540, 'epoch': '1.637'}
+
 82%|████████▏ | 419/512 [47:06<10:42,  6.91s/it]
 82%|████████▏ | 420/512 [47:13<10:29,  6.84s/it]
                                                 
{'loss': '0.0009822', 'grad_norm': '0.1634', 'learning_rate': '1.786e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 12708416, 'tokens/trainable': 192014, 'epoch': '1.641'}
+
 82%|████████▏ | 420/512 [47:13<10:29,  6.84s/it]
 82%|████████▏ | 421/512 [47:20<10:19,  6.80s/it]
                                                 
{'loss': '2.167e-05', 'grad_norm': '0.00194', 'learning_rate': '1.77e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '31.31', 'tokens/total': 12738640, 'tokens/trainable': 192463, 'epoch': '1.645'}
+
 82%|████████▏ | 421/512 [47:20<10:19,  6.80s/it]
 82%|████████▏ | 422/512 [47:27<10:09,  6.77s/it]
                                                 
{'loss': '1.654e-05', 'grad_norm': '0.00106', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 12769232, 'tokens/trainable': 192918, 'epoch': '1.648'}
+
 82%|████████▏ | 422/512 [47:27<10:09,  6.77s/it]
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
                                                 
{'loss': '0.0001707', 'grad_norm': '0.0176', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.46', 'tokens/total': 12799648, 'tokens/trainable': 193425, 'epoch': '1.652'}
+
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
 83%|████████▎ | 424/512 [47:40<09:53,  6.75s/it]
                                                 
{'loss': '2.332e-05', 'grad_norm': '0.002366', 'learning_rate': '1.722e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 12829920, 'tokens/trainable': 193898, 'epoch': '1.656'}
+
 83%|████████▎ | 424/512 [47:40<09:53,  6.75s/it]
 83%|████████▎ | 425/512 [47:47<09:45,  6.73s/it]
                                                 
{'loss': '0.0001577', 'grad_norm': '0.02454', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.29', 'tokens/total': 12860112, 'tokens/trainable': 194315, 'epoch': '1.66'}
+
 83%|████████▎ | 425/512 [47:47<09:45,  6.73s/it]
 83%|████████▎ | 426/512 [47:53<09:38,  6.73s/it]
                                                 
{'loss': '1.428e-05', 'grad_norm': '0.0009002', 'learning_rate': '1.69e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 12890320, 'tokens/trainable': 194796, 'epoch': '1.664'}
+
 83%|████████▎ | 426/512 [47:53<09:38,  6.73s/it]
 83%|████████▎ | 427/512 [48:00<09:32,  6.73s/it]
                                                 
{'loss': '7.186e-06', 'grad_norm': '0.00034', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 12920656, 'tokens/trainable': 195262, 'epoch': '1.668'}
+
 83%|████████▎ | 427/512 [48:00<09:32,  6.73s/it]
 84%|████████▎ | 428/512 [48:07<09:25,  6.73s/it]
                                                 
{'loss': '9.948e-06', 'grad_norm': '0.000536', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.91', 'tokens/total': 12951040, 'tokens/trainable': 195680, 'epoch': '1.672'}
+
 84%|████████▎ | 428/512 [48:07<09:25,  6.73s/it]
 84%|████████▍ | 429/512 [48:14<09:16,  6.71s/it]
                                                 
{'loss': '2.029e-05', 'grad_norm': '0.001489', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 12981232, 'tokens/trainable': 196145, 'epoch': '1.676'}
+
 84%|████████▍ | 429/512 [48:14<09:16,  6.71s/it]
 84%|████████▍ | 430/512 [48:20<09:10,  6.71s/it]
                                                 
{'loss': '0.0003293', 'grad_norm': '0.05131', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.42', 'tokens/total': 13011616, 'tokens/trainable': 196626, 'epoch': '1.68'}
+
 84%|████████▍ | 430/512 [48:20<09:10,  6.71s/it]
 84%|████████▍ | 431/512 [48:27<09:06,  6.75s/it]
                                                 
{'loss': '1.015e-05', 'grad_norm': '0.001407', 'learning_rate': '1.615e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '31.17', 'tokens/total': 13042208, 'tokens/trainable': 197067, 'epoch': '1.684'}
+
 84%|████████▍ | 431/512 [48:27<09:06,  6.75s/it]
 84%|████████▍ | 432/512 [48:34<08:59,  6.74s/it]
                                                 
{'loss': '3.515e-05', 'grad_norm': '0.004116', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.68', 'tokens/total': 13072800, 'tokens/trainable': 197565, 'epoch': '1.688'}
+
 84%|████████▍ | 432/512 [48:34<08:59,  6.74s/it]
 85%|████████▍ | 433/512 [48:41<08:51,  6.72s/it]
                                                 
{'loss': '7.314e-05', 'grad_norm': '0.01148', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 13103328, 'tokens/trainable': 198015, 'epoch': '1.691'}
+
 85%|████████▍ | 433/512 [48:41<08:51,  6.72s/it]
 85%|████████▍ | 434/512 [48:47<08:43,  6.71s/it]
                                                 
{'loss': '0.0002449', 'grad_norm': '0.03078', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 13133472, 'tokens/trainable': 198442, 'epoch': '1.695'}
+
 85%|████████▍ | 434/512 [48:47<08:43,  6.71s/it]
 85%|████████▍ | 435/512 [48:54<08:37,  6.72s/it]
                                                 
{'loss': '1.445e-05', 'grad_norm': '0.002246', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 13164064, 'tokens/trainable': 198914, 'epoch': '1.699'}
+
 85%|████████▍ | 435/512 [48:54<08:37,  6.72s/it]
 85%|████████▌ | 436/512 [49:01<08:30,  6.72s/it]
                                                 
{'loss': '2.742e-05', 'grad_norm': '0.002974', 'learning_rate': '1.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.45', 'tokens/total': 13194464, 'tokens/trainable': 199355, 'epoch': '1.703'}
+
 85%|████████▌ | 436/512 [49:01<08:30,  6.72s/it]
 85%|████████▌ | 437/512 [49:07<08:23,  6.71s/it]
                                                 
{'loss': '2.773e-05', 'grad_norm': '0.002261', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 13224992, 'tokens/trainable': 199830, 'epoch': '1.707'}
+
 85%|████████▌ | 437/512 [49:07<08:23,  6.71s/it]
 86%|████████▌ | 438/512 [49:14<08:17,  6.72s/it]
                                                 
{'loss': '9.154e-05', 'grad_norm': '0.01199', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.97', 'tokens/total': 13255296, 'tokens/trainable': 200338, 'epoch': '1.711'}
+
 86%|████████▌ | 438/512 [49:14<08:17,  6.72s/it]
 86%|████████▌ | 439/512 [49:21<08:08,  6.70s/it]
                                                 
{'loss': '3.515e-05', 'grad_norm': '0.003149', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 13285344, 'tokens/trainable': 200785, 'epoch': '1.715'}
+
 86%|████████▌ | 439/512 [49:21<08:08,  6.70s/it]
 86%|████████▌ | 440/512 [49:27<08:03,  6.71s/it]
                                                 
{'loss': '7.169e-06', 'grad_norm': '0.0003243', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 13315632, 'tokens/trainable': 201267, 'epoch': '1.719'}
+
 86%|████████▌ | 440/512 [49:27<08:03,  6.71s/it]
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
                                                 
{'loss': '0.0007142', 'grad_norm': '0.1277', 'learning_rate': '1.477e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 13346048, 'tokens/trainable': 201734, 'epoch': '1.723'}
+
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
 86%|████████▋ | 442/512 [49:41<07:49,  6.71s/it]
                                                 
{'loss': '0.0006704', 'grad_norm': '0.1148', 'learning_rate': '1.464e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.4', 'tokens/total': 13376304, 'tokens/trainable': 202174, 'epoch': '1.727'}
+
 86%|████████▋ | 442/512 [49:41<07:49,  6.71s/it]
 87%|████████▋ | 443/512 [49:48<07:42,  6.71s/it]
                                                 
{'loss': '9.368e-06', 'grad_norm': '0.0003201', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 13406720, 'tokens/trainable': 202612, 'epoch': '1.73'}
+
 87%|████████▋ | 443/512 [49:48<07:42,  6.71s/it]
 87%|████████▋ | 444/512 [49:54<07:33,  6.66s/it]
                                                 
{'loss': '5.53e-06', 'grad_norm': '0.000266', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.99', 'tokens/total': 13436464, 'tokens/trainable': 203020, 'epoch': '1.734'}
+
 87%|████████▋ | 444/512 [49:54<07:33,  6.66s/it]
 87%|████████▋ | 445/512 [50:01<07:26,  6.66s/it]
                                                 
{'loss': '6.467e-05', 'grad_norm': '0.04957', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.37', 'tokens/total': 13466672, 'tokens/trainable': 203479, 'epoch': '1.738'}
+
 87%|████████▋ | 445/512 [50:01<07:26,  6.66s/it]
 87%|████████▋ | 446/512 [50:08<07:20,  6.67s/it]
                                                 
{'loss': '1.261e-05', 'grad_norm': '0.001684', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 13496736, 'tokens/trainable': 203946, 'epoch': '1.742'}
+
 87%|████████▋ | 446/512 [50:08<07:20,  6.67s/it]
 87%|████████▋ | 447/512 [50:14<07:13,  6.68s/it]
                                                 
{'loss': '7.416e-06', 'grad_norm': '0.0004292', 'learning_rate': '1.402e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '37.45', 'tokens/total': 13526928, 'tokens/trainable': 204382, 'epoch': '1.746'}
+
 87%|████████▋ | 447/512 [50:14<07:13,  6.68s/it]
 88%|████████▊ | 448/512 [50:21<07:07,  6.68s/it]
                                                 
{'loss': '0.0006914', 'grad_norm': '0.2166', 'learning_rate': '1.39e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.35', 'tokens/total': 13557392, 'tokens/trainable': 204830, 'epoch': '1.75'}
+
 88%|████████▊ | 448/512 [50:21<07:07,  6.68s/it][2026-08-18 14:55:20,925] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
+
 88%|████████▊ | 449/512 [50:29<07:33,  7.20s/it]
                                                 
{'loss': '1.45e-05', 'grad_norm': '0.0009793', 'learning_rate': '1.378e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.47', 'tokens/total': 13587776, 'tokens/trainable': 205300, 'epoch': '1.754'}
+
 88%|████████▊ | 449/512 [50:29<07:33,  7.20s/it]
 88%|████████▊ | 450/512 [50:36<07:17,  7.06s/it]
                                                 
{'loss': '0.0002816', 'grad_norm': '0.0285', 'learning_rate': '1.367e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '30.6', 'tokens/total': 13618112, 'tokens/trainable': 205726, 'epoch': '1.758'}
+
 88%|████████▊ | 450/512 [50:36<07:17,  7.06s/it]
 88%|████████▊ | 451/512 [50:43<07:04,  6.96s/it]
                                                 
{'loss': '2.595e-05', 'grad_norm': '0.003279', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '31.63', 'tokens/total': 13648512, 'tokens/trainable': 206163, 'epoch': '1.762'}
+
 88%|████████▊ | 451/512 [50:43<07:04,  6.96s/it]
 88%|████████▊ | 452/512 [50:49<06:52,  6.88s/it]
                                                 
{'loss': '0.004595', 'grad_norm': '0.4962', 'learning_rate': '1.344e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 13678704, 'tokens/trainable': 206619, 'epoch': '1.766'}
+
 88%|████████▊ | 452/512 [50:49<06:52,  6.88s/it]
 88%|████████▊ | 453/512 [50:56<06:43,  6.85s/it]
                                                 
{'loss': '6.578e-05', 'grad_norm': '0.009086', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.24', 'tokens/total': 13709232, 'tokens/trainable': 207101, 'epoch': '1.77'}
+
 88%|████████▊ | 453/512 [50:56<06:43,  6.85s/it]
 89%|████████▊ | 454/512 [51:03<06:34,  6.80s/it]
                                                 
{'loss': '6.735e-05', 'grad_norm': '0.009036', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.77', 'tokens/total': 13739600, 'tokens/trainable': 207546, 'epoch': '1.773'}
+
 89%|████████▊ | 454/512 [51:03<06:34,  6.80s/it]
 89%|████████▉ | 455/512 [51:10<06:26,  6.77s/it]
                                                 
{'loss': '7.266e-05', 'grad_norm': '0.009665', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 13769936, 'tokens/trainable': 207989, 'epoch': '1.777'}
+
 89%|████████▉ | 455/512 [51:10<06:26,  6.77s/it]
 89%|████████▉ | 456/512 [51:16<06:17,  6.74s/it]
                                                 
{'loss': '6.039e-06', 'grad_norm': '0.0004666', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 13800240, 'tokens/trainable': 208402, 'epoch': '1.781'}
+
 89%|████████▉ | 456/512 [51:16<06:17,  6.74s/it]
 89%|████████▉ | 457/512 [51:23<06:10,  6.74s/it]
                                                 
{'loss': '8.293e-06', 'grad_norm': '0.0004485', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 13830512, 'tokens/trainable': 208904, 'epoch': '1.785'}
+
 89%|████████▉ | 457/512 [51:23<06:10,  6.74s/it]
 89%|████████▉ | 458/512 [51:30<06:03,  6.73s/it]
                                                 
{'loss': '2.316e-05', 'grad_norm': '0.003369', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 13860832, 'tokens/trainable': 209382, 'epoch': '1.789'}
+
 89%|████████▉ | 458/512 [51:30<06:03,  6.73s/it]
 90%|████████▉ | 459/512 [51:36<05:55,  6.71s/it]
                                                 
{'loss': '1.119e-05', 'grad_norm': '0.00108', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 13890992, 'tokens/trainable': 209833, 'epoch': '1.793'}
+
 90%|████████▉ | 459/512 [51:36<05:55,  6.71s/it]
 90%|████████▉ | 460/512 [51:43<05:48,  6.70s/it]
                                                 
{'loss': '0.0003375', 'grad_norm': '0.04659', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.53', 'tokens/total': 13921424, 'tokens/trainable': 210284, 'epoch': '1.797'}
+
 90%|████████▉ | 460/512 [51:43<05:48,  6.70s/it]
 90%|█████████ | 461/512 [51:50<05:40,  6.68s/it]
                                                 
{'loss': '0.0001611', 'grad_norm': '0.01848', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.07', 'tokens/total': 13951504, 'tokens/trainable': 210769, 'epoch': '1.801'}
+
 90%|█████████ | 461/512 [51:50<05:40,  6.68s/it]
 90%|█████████ | 462/512 [51:56<05:26,  6.53s/it]
                                                 
{'loss': '0.0001783', 'grad_norm': '0.05658', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.41', 'tokens/total': 13979728, 'tokens/trainable': 211193, 'epoch': '1.805'}
+
 90%|█████████ | 462/512 [51:56<05:26,  6.53s/it]
 90%|█████████ | 463/512 [52:03<05:22,  6.59s/it]
                                                 
{'loss': '9.709e-06', 'grad_norm': '0.0006595', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 14009936, 'tokens/trainable': 211655, 'epoch': '1.809'}
+
 90%|█████████ | 463/512 [52:03<05:22,  6.59s/it]
 91%|█████████ | 464/512 [52:09<05:17,  6.61s/it]
                                                 
{'loss': '0.0001757', 'grad_norm': '0.02738', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.2', 'tokens/total': 14039872, 'tokens/trainable': 212085, 'epoch': '1.812'}
+
 91%|█████████ | 464/512 [52:09<05:17,  6.61s/it]
 91%|█████████ | 465/512 [52:16<05:12,  6.64s/it]
                                                 
{'loss': '8.168e-06', 'grad_norm': '0.0004917', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 14070096, 'tokens/trainable': 212570, 'epoch': '1.816'}
+
 91%|█████████ | 465/512 [52:16<05:12,  6.64s/it]
 91%|█████████ | 466/512 [52:23<05:06,  6.65s/it]
                                                 
{'loss': '1.287e-05', 'grad_norm': '0.001592', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.93', 'tokens/total': 14100192, 'tokens/trainable': 213011, 'epoch': '1.82'}
+
 91%|█████████ | 466/512 [52:23<05:06,  6.65s/it]
 91%|█████████ | 467/512 [52:29<05:00,  6.67s/it]
                                                 
{'loss': '7.528e-06', 'grad_norm': '0.000542', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.58', 'tokens/total': 14130448, 'tokens/trainable': 213450, 'epoch': '1.824'}
+
 91%|█████████ | 467/512 [52:29<05:00,  6.67s/it]
 91%|█████████▏| 468/512 [52:36<04:53,  6.66s/it]
                                                 
{'loss': '4.258e-05', 'grad_norm': '0.0045', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.7', 'tokens/total': 14160480, 'tokens/trainable': 213898, 'epoch': '1.828'}
+
 91%|█████████▏| 468/512 [52:36<04:53,  6.66s/it]
 92%|█████████▏| 469/512 [52:43<04:46,  6.67s/it]
                                                 
{'loss': '0.000211', 'grad_norm': '0.1103', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.88', 'tokens/total': 14191152, 'tokens/trainable': 214366, 'epoch': '1.832'}
+
 92%|█████████▏| 469/512 [52:43<04:46,  6.67s/it]
 92%|█████████▏| 470/512 [52:49<04:34,  6.54s/it]
                                                 
{'loss': '2.79e-05', 'grad_norm': '0.002098', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.72', 'tokens/total': 14219392, 'tokens/trainable': 214800, 'epoch': '1.836'}
+
 92%|█████████▏| 470/512 [52:49<04:34,  6.54s/it]
 92%|█████████▏| 471/512 [52:56<04:31,  6.61s/it]
                                                 
{'loss': '1.734e-05', 'grad_norm': '0.001368', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.48', 'tokens/total': 14249920, 'tokens/trainable': 215292, 'epoch': '1.84'}
+
 92%|█████████▏| 471/512 [52:56<04:31,  6.61s/it]
 92%|█████████▏| 472/512 [53:02<04:25,  6.64s/it]
                                                 
{'loss': '1.396e-05', 'grad_norm': '0.0008574', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 14280048, 'tokens/trainable': 215741, 'epoch': '1.844'}
+
 92%|█████████▏| 472/512 [53:02<04:25,  6.64s/it]
 92%|█████████▏| 473/512 [53:09<04:20,  6.67s/it]
                                                 
{'loss': '3.098e-05', 'grad_norm': '0.003767', 'learning_rate': '1.149e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.68', 'tokens/total': 14310432, 'tokens/trainable': 216194, 'epoch': '1.848'}
+
 92%|█████████▏| 473/512 [53:09<04:20,  6.67s/it]
 93%|█████████▎| 474/512 [53:16<04:13,  6.67s/it]
                                                 
{'loss': '6.771e-06', 'grad_norm': '0.0003345', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 14340496, 'tokens/trainable': 216634, 'epoch': '1.852'}
+
 93%|█████████▎| 474/512 [53:16<04:13,  6.67s/it]
 93%|█████████▎| 475/512 [53:23<04:06,  6.67s/it]
                                                 
{'loss': '7.348e-06', 'grad_norm': '0.0003811', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '30.53', 'tokens/total': 14370720, 'tokens/trainable': 217040, 'epoch': '1.855'}
+
 93%|█████████▎| 475/512 [53:23<04:06,  6.67s/it]
 93%|█████████▎| 476/512 [53:29<04:01,  6.70s/it]
                                                 
{'loss': '1.269e-05', 'grad_norm': '0.001393', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 14401136, 'tokens/trainable': 217530, 'epoch': '1.859'}
+
 93%|█████████▎| 476/512 [53:29<04:01,  6.70s/it]
 93%|█████████▎| 477/512 [53:36<03:53,  6.69s/it]
                                                 
{'loss': '0.0003119', 'grad_norm': '0.02565', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 14431360, 'tokens/trainable': 217988, 'epoch': '1.863'}
+
 93%|█████████▎| 477/512 [53:36<03:53,  6.69s/it]
 93%|█████████▎| 478/512 [53:43<03:48,  6.73s/it]
                                                 
{'loss': '0.0003999', 'grad_norm': '0.08054', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.54', 'tokens/total': 14461968, 'tokens/trainable': 218461, 'epoch': '1.867'}
+
 93%|█████████▎| 478/512 [53:43<03:48,  6.73s/it]
 94%|█████████▎| 479/512 [53:49<03:41,  6.70s/it]
                                                 
{'loss': '0.000334', 'grad_norm': '0.05217', 'learning_rate': '1.108e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 14491984, 'tokens/trainable': 218909, 'epoch': '1.871'}
+
 94%|█████████▎| 479/512 [53:49<03:41,  6.70s/it]
 94%|█████████▍| 480/512 [53:56<03:34,  6.71s/it]
                                                 
{'loss': '0.0005761', 'grad_norm': '0.08784', 'learning_rate': '1.102e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.08', 'tokens/total': 14522544, 'tokens/trainable': 219350, 'epoch': '1.875'}
+
 94%|█████████▍| 480/512 [53:56<03:34,  6.71s/it][2026-08-18 14:58:56,154] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
+
 94%|█████████▍| 481/512 [54:04<03:42,  7.18s/it]
                                                 
{'loss': '0.0007811', 'grad_norm': '0.106', 'learning_rate': '1.096e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '29.52', 'tokens/total': 14552560, 'tokens/trainable': 219776, 'epoch': '1.879'}
+
 94%|█████████▍| 481/512 [54:04<03:42,  7.18s/it]
 94%|█████████▍| 482/512 [54:11<03:31,  7.06s/it]
                                                 
{'loss': '5.21e-06', 'grad_norm': '0.0002076', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 14583056, 'tokens/trainable': 220284, 'epoch': '1.883'}
+
 94%|█████████▍| 482/512 [54:11<03:31,  7.06s/it]
 94%|█████████▍| 483/512 [54:18<03:21,  6.94s/it]
                                                 
{'loss': '1.686e-05', 'grad_norm': '0.002103', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 14613152, 'tokens/trainable': 220714, 'epoch': '1.887'}
+
 94%|█████████▍| 483/512 [54:18<03:21,  6.94s/it]
 95%|█████████▍| 484/512 [54:25<03:12,  6.86s/it]
                                                 
{'loss': '9.006e-05', 'grad_norm': '0.009665', 'learning_rate': '1.079e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 14643328, 'tokens/trainable': 221156, 'epoch': '1.891'}
+
 95%|█████████▍| 484/512 [54:25<03:12,  6.86s/it]
 95%|█████████▍| 485/512 [54:31<03:03,  6.80s/it]
                                                 
{'loss': '1.069e-05', 'grad_norm': '0.001844', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 14673488, 'tokens/trainable': 221605, 'epoch': '1.895'}
+
 95%|█████████▍| 485/512 [54:31<03:03,  6.80s/it]
 95%|█████████▍| 486/512 [54:38<02:55,  6.76s/it]
                                                 
{'loss': '5.547e-06', 'grad_norm': '0.0007748', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.46', 'tokens/total': 14704000, 'tokens/trainable': 222042, 'epoch': '1.898'}
+
 95%|█████████▍| 486/512 [54:38<02:55,  6.76s/it]
 95%|█████████▌| 487/512 [54:45<02:48,  6.73s/it]
                                                 
{'loss': '6.492e-06', 'grad_norm': '0.0002923', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 14733984, 'tokens/trainable': 222474, 'epoch': '1.902'}
+
 95%|█████████▌| 487/512 [54:45<02:48,  6.73s/it]
 95%|█████████▌| 488/512 [54:51<02:41,  6.73s/it]
                                                 
{'loss': '6.497e-06', 'grad_norm': '0.0003241', 'learning_rate': '1.058e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.27', 'tokens/total': 14764400, 'tokens/trainable': 222967, 'epoch': '1.906'}
+
 95%|█████████▌| 488/512 [54:51<02:41,  6.73s/it]
 96%|█████████▌| 489/512 [54:58<02:34,  6.71s/it]
                                                 
{'loss': '1.286e-05', 'grad_norm': '0.001957', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 14794672, 'tokens/trainable': 223420, 'epoch': '1.91'}
+
 96%|█████████▌| 489/512 [54:58<02:34,  6.71s/it]
 96%|█████████▌| 490/512 [55:05<02:27,  6.70s/it]
                                                 
{'loss': '4.987e-06', 'grad_norm': '0.0006595', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '29.4', 'tokens/total': 14824784, 'tokens/trainable': 223864, 'epoch': '1.914'}
+
 96%|█████████▌| 490/512 [55:05<02:27,  6.70s/it]
 96%|█████████▌| 491/512 [55:11<02:20,  6.71s/it]
                                                 
{'loss': '3.641e-05', 'grad_norm': '0.005793', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 14855056, 'tokens/trainable': 224321, 'epoch': '1.918'}
+
 96%|█████████▌| 491/512 [55:11<02:20,  6.71s/it]
 96%|█████████▌| 492/512 [55:18<02:14,  6.72s/it]
                                                 
{'loss': '9.24e-05', 'grad_norm': '0.0105', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 14885392, 'tokens/trainable': 224802, 'epoch': '1.922'}
+
 96%|█████████▌| 492/512 [55:18<02:14,  6.72s/it]
 96%|█████████▋| 493/512 [55:25<02:07,  6.73s/it]
                                                 
{'loss': '6.912e-06', 'grad_norm': '0.0007256', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 14915760, 'tokens/trainable': 225284, 'epoch': '1.926'}
+
 96%|█████████▋| 493/512 [55:25<02:07,  6.73s/it]
 96%|█████████▋| 494/512 [55:32<02:01,  6.74s/it]
                                                 
{'loss': '9.241e-06', 'grad_norm': '0.001594', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 14946192, 'tokens/trainable': 225733, 'epoch': '1.93'}
+
 96%|█████████▋| 494/512 [55:32<02:01,  6.74s/it]
 97%|█████████▋| 495/512 [55:38<01:54,  6.71s/it]
                                                 
{'loss': '6.262e-06', 'grad_norm': '0.0005452', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.58', 'tokens/total': 14976272, 'tokens/trainable': 226172, 'epoch': '1.934'}
+
 97%|█████████▋| 495/512 [55:38<01:54,  6.71s/it]
 97%|█████████▋| 496/512 [55:45<01:47,  6.72s/it]
                                                 
{'loss': '1.628e-05', 'grad_norm': '0.001487', 'learning_rate': '1.027e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.85', 'tokens/total': 15006544, 'tokens/trainable': 226636, 'epoch': '1.938'}
+
 97%|█████████▋| 496/512 [55:45<01:47,  6.72s/it]
 97%|█████████▋| 497/512 [55:52<01:40,  6.71s/it]
                                                 
{'loss': '8.379e-05', 'grad_norm': '0.01586', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 15037040, 'tokens/trainable': 227067, 'epoch': '1.941'}
+
 97%|█████████▋| 497/512 [55:52<01:40,  6.71s/it]
 97%|█████████▋| 498/512 [55:58<01:33,  6.70s/it]
                                                 
{'loss': '6.234e-06', 'grad_norm': '0.0002518', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.65', 'tokens/total': 15067344, 'tokens/trainable': 227532, 'epoch': '1.945'}
+
 97%|█████████▋| 498/512 [55:58<01:33,  6.70s/it]
 97%|█████████▋| 499/512 [56:05<01:27,  6.71s/it]
                                                 
{'loss': '0.0001019', 'grad_norm': '0.02453', 'learning_rate': '1.018e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.57', 'tokens/total': 15097648, 'tokens/trainable': 228007, 'epoch': '1.949'}
+
 97%|█████████▋| 499/512 [56:05<01:27,  6.71s/it]
 98%|█████████▊| 500/512 [56:12<01:20,  6.71s/it]
                                                 
{'loss': '3.307e-05', 'grad_norm': '0.0113', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.89', 'tokens/total': 15128000, 'tokens/trainable': 228457, 'epoch': '1.953'}
+
 98%|█████████▊| 500/512 [56:12<01:20,  6.71s/it]
 98%|█████████▊| 501/512 [56:18<01:13,  6.70s/it]
                                                 
{'loss': '1.035e-05', 'grad_norm': '0.0004915', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 15158352, 'tokens/trainable': 228928, 'epoch': '1.957'}
+
 98%|█████████▊| 501/512 [56:18<01:13,  6.70s/it]
 98%|█████████▊| 502/512 [56:25<01:07,  6.73s/it]
                                                 
{'loss': '9.377e-06', 'grad_norm': '0.001202', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.17', 'tokens/total': 15189136, 'tokens/trainable': 229368, 'epoch': '1.961'}
+
 98%|█████████▊| 502/512 [56:25<01:07,  6.73s/it]
 98%|█████████▊| 503/512 [56:32<01:00,  6.71s/it]
                                                 
{'loss': '1.131e-05', 'grad_norm': '0.001354', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 15219616, 'tokens/trainable': 229815, 'epoch': '1.965'}
+
 98%|█████████▊| 503/512 [56:32<01:00,  6.71s/it]
 98%|█████████▊| 504/512 [56:39<00:53,  6.72s/it]
                                                 
{'loss': '0.0009784', 'grad_norm': '0.1499', 'learning_rate': '1.008e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '38.89', 'tokens/total': 15250032, 'tokens/trainable': 230319, 'epoch': '1.969'}
+
 98%|█████████▊| 504/512 [56:39<00:53,  6.72s/it]
 99%|█████████▊| 505/512 [56:45<00:47,  6.72s/it]
                                                 
{'loss': '5.266e-06', 'grad_norm': '0.0002902', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.53', 'tokens/total': 15280592, 'tokens/trainable': 230807, 'epoch': '1.973'}
+
 99%|█████████▊| 505/512 [56:45<00:47,  6.72s/it]
 99%|█████████▉| 506/512 [56:52<00:40,  6.72s/it]
                                                 
{'loss': '2.386e-05', 'grad_norm': '0.005411', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 15310816, 'tokens/trainable': 231257, 'epoch': '1.977'}
+
 99%|█████████▉| 506/512 [56:52<00:40,  6.72s/it]
 99%|█████████▉| 507/512 [56:59<00:33,  6.74s/it]
                                                 
{'loss': '6.083e-06', 'grad_norm': '0.0002188', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 15341424, 'tokens/trainable': 231746, 'epoch': '1.98'}
+
 99%|█████████▉| 507/512 [56:59<00:33,  6.74s/it]
 99%|█████████▉| 508/512 [57:05<00:26,  6.57s/it]
                                                 
{'loss': '7.37e-06', 'grad_norm': '0.000437', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.33', 'memory/max_allocated (GiB)': '33.33', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.03', 'tokens/total': 15369664, 'tokens/trainable': 232172, 'epoch': '1.984'}
+
 99%|█████████▉| 508/512 [57:05<00:26,  6.57s/it]
 99%|█████████▉| 509/512 [57:12<00:20,  6.69s/it]
                                                 
{'loss': '6.241e-06', 'grad_norm': '0.0003495', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.14', 'tokens/total': 15400496, 'tokens/trainable': 232680, 'epoch': '1.988'}
+
 99%|█████████▉| 509/512 [57:12<00:20,  6.69s/it]
100%|█████████▉| 510/512 [57:19<00:13,  6.71s/it]
                                                 
{'loss': '7.525e-06', 'grad_norm': '0.0002909', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 15430864, 'tokens/trainable': 233169, 'epoch': '1.992'}
+
100%|█████████▉| 510/512 [57:19<00:13,  6.71s/it]
100%|█████████▉| 511/512 [57:25<00:06,  6.70s/it]
                                                 
{'loss': '0.0003311', 'grad_norm': '0.05436', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 15461232, 'tokens/trainable': 233609, 'epoch': '1.996'}
+
100%|█████████▉| 511/512 [57:25<00:06,  6.70s/it]
100%|██████████| 512/512 [57:32<00:00,  6.74s/it]
                                                 
{'loss': '5.907e-06', 'grad_norm': '0.0002704', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 15491760, 'tokens/trainable': 234078, 'epoch': '2'}
+
100%|██████████| 512/512 [57:32<00:00,  6.74s/it][2026-08-18 15:02:32,328] [INFO] [axolotl.core.trainers.base._save:828] [PID:10630] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
+
                                                 
{'train_runtime': '3455', 'train_samples_per_second': '4.743', 'train_steps_per_second': '0.148', 'train_loss': '0.006451', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '36.16', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
+
100%|██████████| 512/512 [57:34<00:00,  6.74s/it]
100%|██████████| 512/512 [57:34<00:00,  6.75s/it]
+[2026-08-18 15:02:34,094] [INFO] [axolotl.train.save_trained_model:267] [PID:10630] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
+[2026-08-18 15:02:34,400] [INFO] [axolotl.train.save_trained_model:388] [PID:10630] Model successfully saved to /workspace/wave/training/checkpoints
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/health/training_started.json b/aft_wave_v2/charter_real_4x__charter0p2/training/health/training_started.json
index e39379bc5a40b90baf715ce762816a184ca7523f..099d23fc26fd6450b01d49fa93235ad8e71933f6 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/health/training_started.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/health/training_started.json
@@ -1,6 +1,6 @@
 {
   "loss": 0.1392,
   "observed": "first_optimizer_loss",
-  "observed_at": "2026-08-18T12:39:25+00:00",
+  "observed_at": "2026-08-18T14:05:07+00:00",
   "status": "training_started"
 }
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/run.json b/aft_wave_v2/charter_real_4x__charter0p2/training/run.json
index 9ce04979ee56c097cfeab3c748b39edd2a219f75..15ffae030bd378c75dd527f4d8b36fe4c824d47a 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/run.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/run.json
@@ -1,9 +1,9 @@
 {
   "run_name": "charter_real_4x__charter0p2",
-  "git_commit": "dc47d38dd7732e27f42285faab76fb29f6840b74",
+  "git_commit": "e0c479b41f5718f428d5a199ba67c513d7fb9574",
   "git_dirty": false,
   "host": "a8707da64105",
-  "started_at": "2026-08-18T12:37:20+00:00",
+  "started_at": "2026-08-18T14:03:45+00:00",
   "configs": {
     "axolotl": "/workspace/wave/training/config/axolotl.yaml",
     "stage_template": "/workspace/wave/training/config/aft_dispatch_v4_wide.yaml"
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/train.log b/aft_wave_v2/charter_real_4x__charter0p2/training/train.log
index 3ee309da7366d092dbc0fd80efa4af420ecf151a..a143192ca56f7ad4e6cd1fc6cfab3972e5e6bd73 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/train.log
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/train.log
@@ -1,8 +1,8 @@
-[2026-08-18 12:37:22,903] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:03:47,454] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-W0818 12:37:25.044000 2703 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:37:25.066000 2703 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:03:49.443000 10319 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:03:49.465000 10319 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
 
      #@@ #@@      @@# @@#
     @@  @@          @@  @@           =@@#                               @@                 #@    =@@#.
@@ -22,17 +22,17 @@ The following values were not passed to `accelerate launch` and had defaults use
 	`--mixed_precision` was set to a value of `'no'`
 	`--dynamo_backend` was set to a value of `'no'`
 To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
-[2026-08-18 12:37:36,881] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:03:59,410] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-W0818 12:37:39.759000 2978 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:37:39.780000 2978 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-[2026-08-18 12:37:42,702] [INFO] [axolotl.integrations.base] Attempting to load plugin: axolotl.integrations.liger.LigerPlugin
-[2026-08-18 12:37:42,707] [INFO] [axolotl.integrations.base] Plugin loaded successfully: axolotl.integrations.liger.LigerPlugin
-[2026-08-18 12:37:42,758] [WARNING] [axolotl.utils.schemas.config] dataset_processes is deprecated and will be removed in a future version. Please use dataset_num_proc instead.
-[2026-08-18 12:37:42,758] [WARNING] [axolotl.utils.schemas.config] Auto-enabling LoRA kernel optimizations for faster training. Please explicitly set `lora_*_kernel` config values to `false` to disable. See https://docs.axolotl.ai/docs/lora_optims.html for more info.
-[2026-08-18 12:37:42,759] [WARNING] [axolotl.utils.schemas.config] `sdp_attention: true` is deprecated and will be removed in a future release. Use `attn_implementation: sdpa` instead.
-[2026-08-18 12:37:43,089] [INFO] [axolotl.cli.config] config:
+W0818 14:04:02.282000 10630 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:02.303000 10630 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+[2026-08-18 14:04:04,698] [INFO] [axolotl.integrations.base] Attempting to load plugin: axolotl.integrations.liger.LigerPlugin
+[2026-08-18 14:04:04,701] [INFO] [axolotl.integrations.base] Plugin loaded successfully: axolotl.integrations.liger.LigerPlugin
+[2026-08-18 14:04:04,753] [WARNING] [axolotl.utils.schemas.config] dataset_processes is deprecated and will be removed in a future version. Please use dataset_num_proc instead.
+[2026-08-18 14:04:04,753] [WARNING] [axolotl.utils.schemas.config] Auto-enabling LoRA kernel optimizations for faster training. Please explicitly set `lora_*_kernel` config values to `false` to disable. See https://docs.axolotl.ai/docs/lora_optims.html for more info.
+[2026-08-18 14:04:04,754] [WARNING] [axolotl.utils.schemas.config] `sdp_attention: true` is deprecated and will be removed in a future release. Use `attn_implementation: sdpa` instead.
+[2026-08-18 14:04:05,093] [INFO] [axolotl.cli.config] config:
 {
   "activation_offloading": false,
   "adapter": "lora",
@@ -232,12 +232,11 @@ W0818 12:37:39.780000 2978 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMo
   "weight_decay": 0.01,
   "world_size": 1
 }
-[2026-08-18 12:37:49,775] [INFO] [axolotl.utils.data.shared] Unable to find prepared dataset in /workspace/wave/training/prepared/2384177d8bb0e5a62897059184c8b941
-[2026-08-18 12:37:49,775] [INFO] [axolotl.utils.data.sft] Loading raw datasets...
-[2026-08-18 12:37:49,775] [WARNING] [axolotl.utils.data.sft] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
-
Generating train split: 0 examples [00:00, ? examples/s]
Generating train split: 3995 examples [00:00, 21852.63 examples/s]
Generating train split: 8003 examples [00:00, 28524.63 examples/s]
Generating train split: 8192 examples [00:00, 27088.63 examples/s]
-[2026-08-18 12:37:51,201] [INFO] [axolotl.utils.data.wrappers] Loading dataset: /workspace/wave/data/datasets/aft_charter0p2.jsonl with base_type: chat_template and prompt_style: None
-[2026-08-18 12:37:51,212] [INFO] [axolotl.prompt_strategies.chat_template] Using chat template:
+[2026-08-18 14:04:08,372] [INFO] [axolotl.utils.data.shared] Unable to find prepared dataset in /workspace/wave/training/prepared/2384177d8bb0e5a62897059184c8b941
+[2026-08-18 14:04:08,372] [INFO] [axolotl.utils.data.sft] Loading raw datasets...
+[2026-08-18 14:04:08,372] [WARNING] [axolotl.utils.data.sft] Processing datasets during training can lead to VRAM instability. Please pre-process your dataset using `axolotl preprocess path/to/config.yml`.
+[2026-08-18 14:04:09,202] [INFO] [axolotl.utils.data.wrappers] Loading dataset: /workspace/wave/data/datasets/aft_charter0p2.jsonl with base_type: chat_template and prompt_style: None
+[2026-08-18 14:04:09,206] [INFO] [axolotl.prompt_strategies.chat_template] Using chat template:
 ---
 {{ bos_token }}
 {%- if messages[0]['role'] == 'system' -%}
@@ -288,598 +287,596 @@ W0818 12:37:39.780000 2978 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMo
 {%- endif -%}
 
 ---
-
Tokenizing Prompts (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▏        | 1000/8192 [00:07<00:55, 129.90 examples/s]
Tokenizing Prompts (num_proc=8):  12%|█▎        | 1024/8192 [00:07<00:54, 131.59 examples/s]
Tokenizing Prompts (num_proc=8):  25%|██▍       | 2024/8192 [00:11<00:29, 211.91 examples/s]
Tokenizing Prompts (num_proc=8):  37%|███▋      | 3048/8192 [00:14<00:20, 247.79 examples/s]
Tokenizing Prompts (num_proc=8):  50%|████▉     | 4072/8192 [00:17<00:15, 266.43 examples/s]
Tokenizing Prompts (num_proc=8):  62%|██████▏   | 5096/8192 [00:21<00:11, 280.75 examples/s]
Tokenizing Prompts (num_proc=8):  75%|███████▍  | 6120/8192 [00:24<00:07, 282.19 examples/s]
Tokenizing Prompts (num_proc=8):  87%|████████▋ | 7144/8192 [00:27<00:03, 296.99 examples/s]
Tokenizing Prompts (num_proc=8): 100%|█████████▉| 8168/8192 [00:31<00:00, 295.67 examples/s]
Tokenizing Prompts (num_proc=8): 100%|██████████| 8192/8192 [00:31<00:00, 258.35 examples/s]
-[2026-08-18 12:38:38,183] [INFO] [axolotl.utils.data.utils] min_input_len: 446
-[2026-08-18 12:38:38,183] [INFO] [axolotl.utils.data.utils] max_input_len: 967
-
Dropping Invalid Sequences (<None or >1280) (num_proc=8):   0%|          | 0/8192 [00:00<?, ? examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8):  12%|█▏        | 1000/8192 [00:00<00:01, 5059.58 examples/s]
Dropping Invalid Sequences (<None or >1280) (num_proc=8): 100%|██████████| 8192/8192 [00:00<00:00, 21541.94 examples/s]
-
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s][2026-08-18 12:38:41,112] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:22,575] [INFO] [axolotl.utils.data.utils] min_input_len: 446
+[2026-08-18 14:04:22,575] [INFO] [axolotl.utils.data.utils] max_input_len: 967
+
Saving the dataset (0/8 shards):   0%|          | 0/8192 [00:00<?, ? examples/s][2026-08-18 14:04:25,020] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,112] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,020] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,146] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,020] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,146] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,020] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,147] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,021] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,147] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,021] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,147] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,022] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,183] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,093] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-[2026-08-18 12:38:41,204] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
+[2026-08-18 14:04:25,104] [WARNING] [py.warnings] /usr/local/lib/python3.12/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.7.0) or chardet (6.0.0.post1)/charset_normalizer (3.4.3) doesn't match a supported version!
   warnings.warn(
 
-W0818 12:38:44.003000 3134 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.027000 3134 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.060000 3138 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.065000 3127 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.091000 3138 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.093000 3127 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.098000 3129 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.111000 3128 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.127000 3130 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.136000 3129 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.149000 3128 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.172000 3130 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.181000 3133 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.183000 3131 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.187000 3132 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.208000 3131 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.219000 3132 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-W0818 12:38:44.219000 3133 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
-
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:08<00:56, 127.71 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:08<00:56, 127.71 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:08<00:48, 127.71 examples/s]
Saving the dataset (3/8 shards):  38%|███▊      | 3072/8192 [00:08<00:40, 127.71 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:08<00:32, 127.71 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:08<00:24, 127.71 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:08<00:16, 127.71 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:08<00:08, 127.71 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:08<00:00, 127.71 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 896.57 examples/s]
-[2026-08-18 12:38:47,979] [INFO] [axolotl.utils.data.sft] Maximum number of steps set at 512
-[2026-08-18 12:38:59,266] [WARNING] [axolotl.loaders.patch_manager] Cannot patch self-attention - requires no dropout
-[2026-08-18 12:39:01,249] [INFO] [axolotl.integrations.liger.plugin] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
-
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  61%|██████▏   | 655/1066 [00:00<00:00, 6528.71it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 6327.25it/s]
-[2026-08-18 12:39:04,805] [INFO] [axolotl.loaders.model] Converting modules to torch.bfloat16
+W0818 14:04:27.847000 10720 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.862000 10716 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.879000 10720 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.889000 10718 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.908000 10716 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.920000 10717 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.929000 10718 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.936000 10725 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.955000 10721 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.968000 10715 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.970000 10717 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.981000 10725 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.985000 10714 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:27.998000 10715 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:28.008000 10721 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:28.015000 10719 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:28.019000 10714 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0818 14:04:28.052000 10719 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+
Saving the dataset (0/8 shards):  12%|█▎        | 1024/8192 [00:07<00:54, 131.44 examples/s]
Saving the dataset (1/8 shards):  12%|█▎        | 1024/8192 [00:07<00:54, 131.44 examples/s]
Saving the dataset (2/8 shards):  25%|██▌       | 2048/8192 [00:07<00:46, 131.44 examples/s]
Saving the dataset (3/8 shards):  50%|█████     | 4096/8192 [00:07<00:31, 131.44 examples/s]
Saving the dataset (4/8 shards):  50%|█████     | 4096/8192 [00:07<00:31, 131.44 examples/s]
Saving the dataset (5/8 shards):  62%|██████▎   | 5120/8192 [00:07<00:23, 131.44 examples/s]
Saving the dataset (5/8 shards):  75%|███████▍  | 6120/8192 [00:07<00:01, 1041.79 examples/s]
Saving the dataset (6/8 shards):  75%|███████▌  | 6144/8192 [00:07<00:01, 1041.79 examples/s]
Saving the dataset (7/8 shards):  88%|████████▊ | 7168/8192 [00:08<00:00, 1041.79 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:08<00:00, 1041.79 examples/s]
Saving the dataset (8/8 shards): 100%|██████████| 8192/8192 [00:09<00:00, 860.78 examples/s] 
+[2026-08-18 14:04:33,799] [INFO] [axolotl.utils.data.sft] Maximum number of steps set at 512
+[2026-08-18 14:04:44,292] [WARNING] [axolotl.loaders.patch_manager] Cannot patch self-attention - requires no dropout
+[2026-08-18 14:04:45,486] [INFO] [axolotl.integrations.liger.plugin] Applying LIGER to gemma3 with kwargs: {'rope': True, 'cross_entropy': None, 'fused_linear_cross_entropy': True, 'rms_norm': True, 'layer_norm': None, 'geglu': True}
+
Loading weights:   0%|          | 0/1066 [00:00<?, ?it/s]
Loading weights:  44%|████▍     | 474/1066 [00:00<00:00, 4739.18it/s]
Loading weights:  89%|████████▉ | 948/1066 [00:00<00:00, 4640.30it/s]
Loading weights: 100%|██████████| 1066/1066 [00:00<00:00, 4544.95it/s]
+[2026-08-18 14:04:49,110] [INFO] [axolotl.loaders.model] Converting modules to torch.bfloat16
 trainable params: 136,912,896 || all params: 12,324,237,936 || trainable%: 1.1109
-[2026-08-18 12:39:09,449] [INFO] [axolotl.monkeypatch.lora_kernels] LoRA kernels: dropout=0.05 enabled
-[2026-08-18 12:39:14,403] [INFO] [axolotl.train] Pre-saving adapter config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:14,403] [INFO] [axolotl.train] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:14,778] [INFO] [axolotl.train] Pre-saving model config to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:14,782] [INFO] [axolotl.train] Pre-saving processor to /workspace/wave/training/checkpoints...
-[2026-08-18 12:39:15,097] [INFO] [axolotl.train] Starting trainer...
+[2026-08-18 14:04:53,116] [INFO] [axolotl.monkeypatch.lora_kernels] LoRA kernels: dropout=0.05 enabled
+[2026-08-18 14:04:58,288] [INFO] [axolotl.train] Pre-saving adapter config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,289] [INFO] [axolotl.train] Pre-saving tokenizer to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,638] [INFO] [axolotl.train] Pre-saving model config to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,642] [INFO] [axolotl.train] Pre-saving processor to /workspace/wave/training/checkpoints...
+[2026-08-18 14:04:58,967] [INFO] [axolotl.train] Starting trainer...
 
  0%|          | 0/512 [00:00<?, ?it/s][transformers] `use_return_dict` is deprecated! Use `return_dict` instead!
-
  0%|          | 1/512 [00:10<1:26:05, 10.11s/it]
                                                 
{'loss': '0.1392', 'grad_norm': '1.368', 'learning_rate': '0', 'ppl': '1.149', 'memory/max_active (GiB)': '32.79', 'memory/max_allocated (GiB)': '32.79', 'memory/device_reserved (GiB)': '33.9', 'tokens/train_per_sec_per_gpu': '21.9', 'tokens/total': 30176, 'tokens/trainable': 417, 'epoch': '0.003906'}
-
  0%|          | 1/512 [00:10<1:26:05, 10.11s/it]
  0%|          | 2/512 [00:16<1:08:38,  8.08s/it]
                                                 
{'loss': '0.151', 'grad_norm': '1.227', 'learning_rate': '4e-06', 'ppl': '1.163', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '34.64', 'tokens/total': 60272, 'tokens/trainable': 873, 'epoch': '0.007812'}
-
  0%|          | 2/512 [00:16<1:08:38,  8.08s/it]
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
                                                 
{'loss': '0.1434', 'grad_norm': '1.363', 'learning_rate': '8e-06', 'ppl': '1.154', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 90512, 'tokens/trainable': 1353, 'epoch': '0.01172'}
-
  1%|          | 3/512 [00:23<1:03:18,  7.46s/it]
  1%|          | 4/512 [00:30<1:00:44,  7.17s/it]
                                                 
{'loss': '0.1327', 'grad_norm': '1.269', 'learning_rate': '1.2e-05', 'ppl': '1.142', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.73', 'tokens/train_per_sec_per_gpu': '31.2', 'tokens/total': 120944, 'tokens/trainable': 1778, 'epoch': '0.01562'}
-
  1%|          | 4/512 [00:30<1:00:44,  7.17s/it]
  1%|          | 5/512 [00:36<59:11,  7.01s/it]  
                                               
{'loss': '0.1425', 'grad_norm': '1.21', 'learning_rate': '1.6e-05', 'ppl': '1.153', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.12', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 151440, 'tokens/trainable': 2261, 'epoch': '0.01953'}
-
  1%|          | 5/512 [00:36<59:11,  7.01s/it]
  1%|          | 6/512 [00:43<58:15,  6.91s/it]
                                               
{'loss': '0.1099', 'grad_norm': '0.7939', 'learning_rate': '2e-05', 'ppl': '1.116', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 181984, 'tokens/trainable': 2705, 'epoch': '0.02344'}
-
  1%|          | 6/512 [00:43<58:15,  6.91s/it]
  1%|▏         | 7/512 [00:50<57:35,  6.84s/it]
                                               
{'loss': '0.0832', 'grad_norm': '0.9592', 'learning_rate': '2.4e-05', 'ppl': '1.087', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 212336, 'tokens/trainable': 3136, 'epoch': '0.02734'}
-
  1%|▏         | 7/512 [00:50<57:35,  6.84s/it]
  2%|▏         | 8/512 [00:57<57:08,  6.80s/it]
                                               
{'loss': '0.08284', 'grad_norm': '1.093', 'learning_rate': '2.8e-05', 'ppl': '1.086', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 242592, 'tokens/trainable': 3579, 'epoch': '0.03125'}
-
  2%|▏         | 8/512 [00:57<57:08,  6.80s/it]
  2%|▏         | 9/512 [01:03<56:41,  6.76s/it]
                                               
{'loss': '0.06837', 'grad_norm': '2.977', 'learning_rate': '3.2e-05', 'ppl': '1.071', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '29.09', 'tokens/total': 272784, 'tokens/trainable': 4000, 'epoch': '0.03516'}
-
  2%|▏         | 9/512 [01:03<56:41,  6.76s/it]
  2%|▏         | 10/512 [01:10<56:27,  6.75s/it]
                                                
{'loss': '0.05963', 'grad_norm': '2.392', 'learning_rate': '3.6e-05', 'ppl': '1.061', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '37.52', 'tokens/total': 303184, 'tokens/trainable': 4474, 'epoch': '0.03906'}
-
  2%|▏         | 10/512 [01:10<56:27,  6.75s/it]
  2%|▏         | 11/512 [01:17<56:17,  6.74s/it]
                                                
{'loss': '0.04087', 'grad_norm': '34.52', 'learning_rate': '4e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 333296, 'tokens/trainable': 4942, 'epoch': '0.04297'}
-
  2%|▏         | 11/512 [01:17<56:17,  6.74s/it]
  2%|▏         | 12/512 [01:23<56:09,  6.74s/it]
                                                
{'loss': '0.07224', 'grad_norm': '3.481', 'learning_rate': '4.4e-05', 'ppl': '1.075', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.17', 'tokens/total': 363840, 'tokens/trainable': 5424, 'epoch': '0.04688'}
-
  2%|▏         | 12/512 [01:23<56:09,  6.74s/it]
  3%|▎         | 13/512 [01:30<56:05,  6.74s/it]
                                                
{'loss': '0.05329', 'grad_norm': '3.392', 'learning_rate': '4.8e-05', 'ppl': '1.055', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 394112, 'tokens/trainable': 5855, 'epoch': '0.05078'}
-
  3%|▎         | 13/512 [01:30<56:05,  6.74s/it]
  3%|▎         | 14/512 [01:37<56:03,  6.75s/it]
                                                
{'loss': '0.04503', 'grad_norm': '1.194', 'learning_rate': '5.2e-05', 'ppl': '1.046', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.67', 'tokens/total': 424656, 'tokens/trainable': 6344, 'epoch': '0.05469'}
-
  3%|▎         | 14/512 [01:37<56:03,  6.75s/it]
  3%|▎         | 15/512 [01:44<55:56,  6.75s/it]
                                                
{'loss': '0.09748', 'grad_norm': '5.4', 'learning_rate': '5.6e-05', 'ppl': '1.102', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '36.01', 'tokens/total': 455152, 'tokens/trainable': 6825, 'epoch': '0.05859'}
-
  3%|▎         | 15/512 [01:44<55:56,  6.75s/it]
  3%|▎         | 16/512 [01:50<55:34,  6.72s/it]
                                                
{'loss': '0.1318', 'grad_norm': '5.269', 'learning_rate': '6e-05', 'ppl': '1.141', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '29.93', 'tokens/total': 485328, 'tokens/trainable': 7230, 'epoch': '0.0625'}
-
  3%|▎         | 16/512 [01:50<55:34,  6.72s/it]
  3%|▎         | 17/512 [01:57<55:32,  6.73s/it]
                                                
{'loss': '0.07405', 'grad_norm': '2.03', 'learning_rate': '6.4e-05', 'ppl': '1.077', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.85', 'tokens/total': 515744, 'tokens/trainable': 7699, 'epoch': '0.06641'}
-
  3%|▎         | 17/512 [01:57<55:32,  6.73s/it]
  4%|▎         | 18/512 [02:04<56:10,  6.82s/it]
                                                
{'loss': '0.0487', 'grad_norm': '1.814', 'learning_rate': '6.8e-05', 'ppl': '1.05', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '32.57', 'tokens/total': 546144, 'tokens/trainable': 8184, 'epoch': '0.07031'}
-
  4%|▎         | 18/512 [02:04<56:10,  6.82s/it]
  4%|▎         | 19/512 [02:11<55:50,  6.80s/it]
                                                
{'loss': '0.009342', 'grad_norm': '0.4662', 'learning_rate': '7.2e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 576560, 'tokens/trainable': 8675, 'epoch': '0.07422'}
-
  4%|▎         | 19/512 [02:11<55:50,  6.80s/it]
  4%|▍         | 20/512 [02:18<55:36,  6.78s/it]
                                                
{'loss': '0.04048', 'grad_norm': '1.167', 'learning_rate': '7.6e-05', 'ppl': '1.041', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.74', 'tokens/total': 607056, 'tokens/trainable': 9189, 'epoch': '0.07812'}
-
  4%|▍         | 20/512 [02:18<55:36,  6.78s/it]
  4%|▍         | 21/512 [02:24<55:22,  6.77s/it]
                                                
{'loss': '0.02852', 'grad_norm': '0.8362', 'learning_rate': '8e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 637488, 'tokens/trainable': 9614, 'epoch': '0.08203'}
-
  4%|▍         | 21/512 [02:24<55:22,  6.77s/it]
  4%|▍         | 22/512 [02:31<55:06,  6.75s/it]
                                                
{'loss': '0.04155', 'grad_norm': '0.8654', 'learning_rate': '8.4e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 667632, 'tokens/trainable': 10086, 'epoch': '0.08594'}
-
  4%|▍         | 22/512 [02:31<55:06,  6.75s/it]
  4%|▍         | 23/512 [02:38<55:27,  6.81s/it]
                                                
{'loss': '0.02058', 'grad_norm': '0.8493', 'learning_rate': '8.8e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.86', 'tokens/total': 697872, 'tokens/trainable': 10535, 'epoch': '0.08984'}
-
  4%|▍         | 23/512 [02:38<55:27,  6.81s/it]
  5%|▍         | 24/512 [02:45<55:12,  6.79s/it]
                                                
{'loss': '0.009413', 'grad_norm': '0.3281', 'learning_rate': '9.2e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.42', 'tokens/total': 728432, 'tokens/trainable': 10993, 'epoch': '0.09375'}
-
  5%|▍         | 24/512 [02:45<55:12,  6.79s/it]
  5%|▍         | 25/512 [02:51<53:41,  6.62s/it]
                                                
{'loss': '0.02935', 'grad_norm': '2.55', 'learning_rate': '9.6e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.78', 'tokens/total': 756656, 'tokens/trainable': 11417, 'epoch': '0.09766'}
-
  5%|▍         | 25/512 [02:51<53:41,  6.62s/it]
  5%|▌         | 26/512 [02:58<53:46,  6.64s/it]
                                                
{'loss': '0.01868', 'grad_norm': '0.8028', 'learning_rate': '0.0001', 'ppl': '1.019', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.21', 'tokens/total': 787120, 'tokens/trainable': 11874, 'epoch': '0.1016'}
-
  5%|▌         | 26/512 [02:58<53:46,  6.64s/it]
  5%|▌         | 27/512 [03:04<53:59,  6.68s/it]
                                                
{'loss': '0.05519', 'grad_norm': '1.805', 'learning_rate': '0.0001', 'ppl': '1.057', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 817504, 'tokens/trainable': 12316, 'epoch': '0.1055'}
-
  5%|▌         | 27/512 [03:04<53:59,  6.68s/it]
  5%|▌         | 28/512 [03:11<54:06,  6.71s/it]
                                                
{'loss': '0.02146', 'grad_norm': '1.037', 'learning_rate': '0.0001', 'ppl': '1.022', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 848128, 'tokens/trainable': 12800, 'epoch': '0.1094'}
-
  5%|▌         | 28/512 [03:11<54:06,  6.71s/it]
  6%|▌         | 29/512 [03:18<54:03,  6.72s/it]
                                                
{'loss': '0.03214', 'grad_norm': '2.161', 'learning_rate': '9.999e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 878448, 'tokens/trainable': 13270, 'epoch': '0.1133'}
-
  6%|▌         | 29/512 [03:18<54:03,  6.72s/it]
  6%|▌         | 30/512 [03:25<54:04,  6.73s/it]
                                                
{'loss': '0.02866', 'grad_norm': '1.293', 'learning_rate': '9.999e-05', 'ppl': '1.029', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.17', 'tokens/total': 909008, 'tokens/trainable': 13752, 'epoch': '0.1172'}
-
  6%|▌         | 30/512 [03:25<54:04,  6.73s/it]
  6%|▌         | 31/512 [03:31<54:01,  6.74s/it]
                                                
{'loss': '0.02288', 'grad_norm': '0.8399', 'learning_rate': '9.998e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 939312, 'tokens/trainable': 14218, 'epoch': '0.1211'}
-
  6%|▌         | 31/512 [03:31<54:01,  6.74s/it]
  6%|▋         | 32/512 [03:38<53:42,  6.71s/it]
                                                
{'loss': '0.006213', 'grad_norm': '0.2222', 'learning_rate': '9.997e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 969264, 'tokens/trainable': 14655, 'epoch': '0.125'}
-
  6%|▋         | 32/512 [03:38<53:42,  6.71s/it][2026-08-18 12:42:54,332] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
-
  6%|▋         | 33/512 [03:47<59:05,  7.40s/it]
                                                
{'loss': '0.01775', 'grad_norm': '0.8033', 'learning_rate': '9.995e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '33.19', 'tokens/total': 999792, 'tokens/trainable': 15087, 'epoch': '0.1289'}
-
  6%|▋         | 33/512 [03:47<59:05,  7.40s/it]
  7%|▋         | 34/512 [03:54<57:11,  7.18s/it]
                                                
{'loss': '0.01878', 'grad_norm': '0.3686', 'learning_rate': '9.994e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.81', 'tokens/total': 1029840, 'tokens/trainable': 15548, 'epoch': '0.1328'}
-
  7%|▋         | 34/512 [03:54<57:11,  7.18s/it]
  7%|▋         | 35/512 [04:01<56:08,  7.06s/it]
                                                
{'loss': '0.01864', 'grad_norm': '0.642', 'learning_rate': '9.992e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.6', 'tokens/total': 1060416, 'tokens/trainable': 16037, 'epoch': '0.1367'}
-
  7%|▋         | 35/512 [04:01<56:08,  7.06s/it]
  7%|▋         | 36/512 [04:07<55:03,  6.94s/it]
                                                
{'loss': '0.0156', 'grad_norm': '0.6261', 'learning_rate': '9.991e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '30.96', 'tokens/total': 1090464, 'tokens/trainable': 16475, 'epoch': '0.1406'}
-
  7%|▋         | 36/512 [04:07<55:03,  6.94s/it]
  7%|▋         | 37/512 [04:14<54:21,  6.87s/it]
                                                
{'loss': '0.01307', 'grad_norm': '0.72', 'learning_rate': '9.989e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 1120944, 'tokens/trainable': 16919, 'epoch': '0.1445'}
-
  7%|▋         | 37/512 [04:14<54:21,  6.87s/it]
  7%|▋         | 38/512 [04:21<53:49,  6.81s/it]
                                                
{'loss': '0.006183', 'grad_norm': '0.5011', 'learning_rate': '9.987e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.85', 'tokens/total': 1151360, 'tokens/trainable': 17333, 'epoch': '0.1484'}
-
  7%|▋         | 38/512 [04:21<53:49,  6.81s/it]
  8%|▊         | 39/512 [04:27<53:38,  6.80s/it]
                                                
{'loss': '0.009255', 'grad_norm': '0.5727', 'learning_rate': '9.984e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36', 'tokens/total': 1181728, 'tokens/trainable': 17806, 'epoch': '0.1523'}
-
  8%|▊         | 39/512 [04:27<53:38,  6.80s/it]
  8%|▊         | 40/512 [04:34<53:20,  6.78s/it]
                                                
{'loss': '0.01721', 'grad_norm': '0.8821', 'learning_rate': '9.982e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.53', 'tokens/total': 1212064, 'tokens/trainable': 18290, 'epoch': '0.1562'}
-
  8%|▊         | 40/512 [04:34<53:20,  6.78s/it]
  8%|▊         | 41/512 [04:41<53:03,  6.76s/it]
                                                
{'loss': '0.002518', 'grad_norm': '0.2254', 'learning_rate': '9.979e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.91', 'tokens/total': 1242592, 'tokens/trainable': 18747, 'epoch': '0.1602'}
-
  8%|▊         | 41/512 [04:41<53:03,  6.76s/it]
  8%|▊         | 42/512 [04:48<52:48,  6.74s/it]
                                                
{'loss': '0.02955', 'grad_norm': '1.097', 'learning_rate': '9.976e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 1272800, 'tokens/trainable': 19172, 'epoch': '0.1641'}
-
  8%|▊         | 42/512 [04:48<52:48,  6.74s/it]
  8%|▊         | 43/512 [04:54<52:50,  6.76s/it]
                                                
{'loss': '0.01922', 'grad_norm': '1.747', 'learning_rate': '9.973e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.17', 'tokens/total': 1303600, 'tokens/trainable': 19622, 'epoch': '0.168'}
-
  8%|▊         | 43/512 [04:54<52:50,  6.76s/it]
  9%|▊         | 44/512 [05:01<52:37,  6.75s/it]
                                                
{'loss': '0.009303', 'grad_norm': '0.8241', 'learning_rate': '9.97e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 1333936, 'tokens/trainable': 20073, 'epoch': '0.1719'}
-
  9%|▊         | 44/512 [05:01<52:37,  6.75s/it]
  9%|▉         | 45/512 [05:08<52:27,  6.74s/it]
                                                
{'loss': '0.01563', 'grad_norm': '0.9528', 'learning_rate': '9.966e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.39', 'tokens/total': 1364288, 'tokens/trainable': 20497, 'epoch': '0.1758'}
-
  9%|▉         | 45/512 [05:08<52:27,  6.74s/it]
  9%|▉         | 46/512 [05:14<52:15,  6.73s/it]
                                                
{'loss': '0.0121', 'grad_norm': '0.6404', 'learning_rate': '9.963e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 1394368, 'tokens/trainable': 20940, 'epoch': '0.1797'}
-
  9%|▉         | 46/512 [05:14<52:15,  6.73s/it]
  9%|▉         | 47/512 [05:21<52:20,  6.75s/it]
                                                
{'loss': '0.00568', 'grad_norm': '0.2612', 'learning_rate': '9.959e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 1425024, 'tokens/trainable': 21405, 'epoch': '0.1836'}
-
  9%|▉         | 47/512 [05:21<52:20,  6.75s/it]
  9%|▉         | 48/512 [05:28<52:05,  6.74s/it]
                                                
{'loss': '0.01393', 'grad_norm': '0.5887', 'learning_rate': '9.955e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.98', 'tokens/total': 1455392, 'tokens/trainable': 21828, 'epoch': '0.1875'}
-
  9%|▉         | 48/512 [05:28<52:05,  6.74s/it]
 10%|▉         | 49/512 [05:35<51:48,  6.71s/it]
                                                
{'loss': '0.02081', 'grad_norm': '1.133', 'learning_rate': '9.951e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.27', 'tokens/total': 1485616, 'tokens/trainable': 22241, 'epoch': '0.1914'}
-
 10%|▉         | 49/512 [05:35<51:48,  6.71s/it]
 10%|▉         | 50/512 [05:41<51:41,  6.71s/it]
                                                
{'loss': '0.01006', 'grad_norm': '0.6578', 'learning_rate': '9.946e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 1515984, 'tokens/trainable': 22688, 'epoch': '0.1953'}
-
 10%|▉         | 50/512 [05:41<51:41,  6.71s/it]
 10%|▉         | 51/512 [05:48<51:26,  6.70s/it]
                                                
{'loss': '0.0022', 'grad_norm': '0.2045', 'learning_rate': '9.942e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32', 'tokens/total': 1546224, 'tokens/trainable': 23136, 'epoch': '0.1992'}
-
 10%|▉         | 51/512 [05:48<51:26,  6.70s/it]
 10%|█         | 52/512 [05:55<51:27,  6.71s/it]
                                                
{'loss': '0.01885', 'grad_norm': '0.8622', 'learning_rate': '9.937e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.75', 'tokens/total': 1576528, 'tokens/trainable': 23631, 'epoch': '0.2031'}
-
 10%|█         | 52/512 [05:55<51:27,  6.71s/it]
 10%|█         | 53/512 [06:02<51:25,  6.72s/it]
                                                
{'loss': '0.01173', 'grad_norm': '0.8481', 'learning_rate': '9.932e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 1606768, 'tokens/trainable': 24096, 'epoch': '0.207'}
-
 10%|█         | 53/512 [06:02<51:25,  6.72s/it]
 11%|█         | 54/512 [06:08<51:10,  6.70s/it]
                                                
{'loss': '0.005829', 'grad_norm': '0.7554', 'learning_rate': '9.927e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.7', 'tokens/total': 1636640, 'tokens/trainable': 24569, 'epoch': '0.2109'}
-
 11%|█         | 54/512 [06:08<51:10,  6.70s/it]
 11%|█         | 55/512 [06:15<51:14,  6.73s/it]
                                                
{'loss': '0.0102', 'grad_norm': '0.7365', 'learning_rate': '9.921e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 1667136, 'tokens/trainable': 25067, 'epoch': '0.2148'}
-
 11%|█         | 55/512 [06:15<51:14,  6.73s/it]
 11%|█         | 56/512 [06:22<51:09,  6.73s/it]
                                                
{'loss': '0.008617', 'grad_norm': '0.9789', 'learning_rate': '9.916e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.35', 'tokens/total': 1697472, 'tokens/trainable': 25541, 'epoch': '0.2188'}
-
 11%|█         | 56/512 [06:22<51:09,  6.73s/it]
 11%|█         | 57/512 [06:28<51:01,  6.73s/it]
                                                
{'loss': '0.0009543', 'grad_norm': '0.1047', 'learning_rate': '9.91e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 1727680, 'tokens/trainable': 25965, 'epoch': '0.2227'}
-
 11%|█         | 57/512 [06:28<51:01,  6.73s/it]
 11%|█▏        | 58/512 [06:35<50:51,  6.72s/it]
                                                
{'loss': '0.01943', 'grad_norm': '1.643', 'learning_rate': '9.904e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.71', 'tokens/total': 1758192, 'tokens/trainable': 26409, 'epoch': '0.2266'}
-
 11%|█▏        | 58/512 [06:35<50:51,  6.72s/it]
 12%|█▏        | 59/512 [06:42<50:49,  6.73s/it]
                                                
{'loss': '0.01346', 'grad_norm': '1.175', 'learning_rate': '9.898e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.91', 'tokens/total': 1788816, 'tokens/trainable': 26883, 'epoch': '0.2305'}
-
 12%|█▏        | 59/512 [06:42<50:49,  6.73s/it]
 12%|█▏        | 60/512 [06:49<50:42,  6.73s/it]
                                                
{'loss': '0.006647', 'grad_norm': '0.5781', 'learning_rate': '9.892e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.6', 'tokens/total': 1819056, 'tokens/trainable': 27323, 'epoch': '0.2344'}
-
 12%|█▏        | 60/512 [06:49<50:42,  6.73s/it]
 12%|█▏        | 61/512 [06:55<50:36,  6.73s/it]
                                                
{'loss': '0.002089', 'grad_norm': '0.2438', 'learning_rate': '9.886e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '39.06', 'tokens/total': 1849664, 'tokens/trainable': 27830, 'epoch': '0.2383'}
-
 12%|█▏        | 61/512 [06:55<50:36,  6.73s/it]
 12%|█▏        | 62/512 [07:02<50:45,  6.77s/it]
                                                
{'loss': '0.008961', 'grad_norm': '0.7739', 'learning_rate': '9.879e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.12', 'tokens/total': 1880272, 'tokens/trainable': 28281, 'epoch': '0.2422'}
-
 12%|█▏        | 62/512 [07:02<50:45,  6.77s/it]
 12%|█▏        | 63/512 [07:09<50:33,  6.76s/it]
                                                
{'loss': '0.008773', 'grad_norm': '0.4645', 'learning_rate': '9.872e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.12', 'tokens/total': 1910752, 'tokens/trainable': 28742, 'epoch': '0.2461'}
-
 12%|█▏        | 63/512 [07:09<50:33,  6.76s/it]
 12%|█▎        | 64/512 [07:16<50:13,  6.73s/it]
                                                
{'loss': '0.005191', 'grad_norm': '0.3245', 'learning_rate': '9.865e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.51', 'tokens/total': 1940848, 'tokens/trainable': 29183, 'epoch': '0.25'}
-
 12%|█▎        | 64/512 [07:16<50:13,  6.73s/it][2026-08-18 12:46:31,785] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
-
 13%|█▎        | 65/512 [07:24<54:29,  7.32s/it]
                                                
{'loss': '0.002685', 'grad_norm': '0.1417', 'learning_rate': '9.858e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.41', 'tokens/total': 1971344, 'tokens/trainable': 29648, 'epoch': '0.2539'}
-
 13%|█▎        | 65/512 [07:24<54:29,  7.32s/it]
 13%|█▎        | 66/512 [07:31<52:57,  7.12s/it]
                                                
{'loss': '0.001938', 'grad_norm': '0.1456', 'learning_rate': '9.851e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.06', 'tokens/train_per_sec_per_gpu': '33.39', 'tokens/total': 2001712, 'tokens/trainable': 30075, 'epoch': '0.2578'}
-
 13%|█▎        | 66/512 [07:31<52:57,  7.12s/it]
 13%|█▎        | 67/512 [07:37<50:53,  6.86s/it]
                                                
{'loss': '0.002464', 'grad_norm': '0.1445', 'learning_rate': '9.844e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.46', 'tokens/total': 2029936, 'tokens/trainable': 30546, 'epoch': '0.2617'}
-
 13%|█▎        | 67/512 [07:37<50:53,  6.86s/it]
 13%|█▎        | 68/512 [07:44<50:31,  6.83s/it]
                                                
{'loss': '0.002092', 'grad_norm': '0.1162', 'learning_rate': '9.836e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.97', 'tokens/total': 2060384, 'tokens/trainable': 31025, 'epoch': '0.2656'}
-
 13%|█▎        | 68/512 [07:44<50:31,  6.83s/it]
 13%|█▎        | 69/512 [07:51<50:16,  6.81s/it]
                                                
{'loss': '0.01863', 'grad_norm': '1.026', 'learning_rate': '9.828e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 2090688, 'tokens/trainable': 31505, 'epoch': '0.2695'}
-
 13%|█▎        | 69/512 [07:51<50:16,  6.81s/it]
 14%|█▎        | 70/512 [07:57<49:53,  6.77s/it]
                                                
{'loss': '0.002588', 'grad_norm': '0.2051', 'learning_rate': '9.82e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '32.16', 'tokens/total': 2120848, 'tokens/trainable': 31956, 'epoch': '0.2734'}
-
 14%|█▎        | 70/512 [07:57<49:53,  6.77s/it]
 14%|█▍        | 71/512 [08:04<48:29,  6.60s/it]
                                                
{'loss': '0.001408', 'grad_norm': '0.09381', 'learning_rate': '9.812e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.35', 'memory/max_allocated (GiB)': '33.35', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.28', 'tokens/total': 2149120, 'tokens/trainable': 32374, 'epoch': '0.2773'}
-
 14%|█▍        | 71/512 [08:04<48:29,  6.60s/it]
 14%|█▍        | 72/512 [08:10<48:38,  6.63s/it]
                                                
{'loss': '0.004073', 'grad_norm': '0.2722', 'learning_rate': '9.803e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.41', 'tokens/total': 2179472, 'tokens/trainable': 32833, 'epoch': '0.2812'}
-
 14%|█▍        | 72/512 [08:10<48:38,  6.63s/it]
 14%|█▍        | 73/512 [08:17<48:40,  6.65s/it]
                                                
{'loss': '0.01158', 'grad_norm': '0.3583', 'learning_rate': '9.795e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.56', 'tokens/total': 2209984, 'tokens/trainable': 33304, 'epoch': '0.2852'}
-
 14%|█▍        | 73/512 [08:17<48:40,  6.65s/it]
 14%|█▍        | 74/512 [08:24<48:55,  6.70s/it]
                                                
{'loss': '0.01845', 'grad_norm': '1.028', 'learning_rate': '9.786e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.26', 'tokens/total': 2240512, 'tokens/trainable': 33771, 'epoch': '0.2891'}
-
 14%|█▍        | 74/512 [08:24<48:55,  6.70s/it]
 15%|█▍        | 75/512 [08:31<48:47,  6.70s/it]
                                                
{'loss': '0.008879', 'grad_norm': '0.3613', 'learning_rate': '9.777e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 2270608, 'tokens/trainable': 34186, 'epoch': '0.293'}
-
 15%|█▍        | 75/512 [08:31<48:47,  6.70s/it]
 15%|█▍        | 76/512 [08:37<48:50,  6.72s/it]
                                                
{'loss': '0.001569', 'grad_norm': '0.1639', 'learning_rate': '9.768e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.72', 'tokens/total': 2300864, 'tokens/trainable': 34667, 'epoch': '0.2969'}
-
 15%|█▍        | 76/512 [08:37<48:50,  6.72s/it]
 15%|█▌        | 77/512 [08:44<48:37,  6.71s/it]
                                                
{'loss': '0.02548', 'grad_norm': '2.788', 'learning_rate': '9.759e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.61', 'tokens/total': 2330960, 'tokens/trainable': 35078, 'epoch': '0.3008'}
-
 15%|█▌        | 77/512 [08:44<48:37,  6.71s/it]
 15%|█▌        | 78/512 [08:51<48:43,  6.74s/it]
                                                
{'loss': '0.00559', 'grad_norm': '0.3399', 'learning_rate': '9.749e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.71', 'tokens/total': 2361584, 'tokens/trainable': 35550, 'epoch': '0.3047'}
-
 15%|█▌        | 78/512 [08:51<48:43,  6.74s/it]
 15%|█▌        | 79/512 [08:57<48:32,  6.73s/it]
                                                
{'loss': '0.01145', 'grad_norm': '0.7557', 'learning_rate': '9.74e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.39', 'tokens/total': 2391936, 'tokens/trainable': 35977, 'epoch': '0.3086'}
-
 15%|█▌        | 79/512 [08:57<48:32,  6.73s/it]
 16%|█▌        | 80/512 [09:04<48:29,  6.73s/it]
                                                
{'loss': '0.01585', 'grad_norm': '0.4028', 'learning_rate': '9.73e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.08', 'tokens/total': 2422432, 'tokens/trainable': 36462, 'epoch': '0.3125'}
-
 16%|█▌        | 80/512 [09:04<48:29,  6.73s/it]
 16%|█▌        | 81/512 [09:11<48:18,  6.73s/it]
                                                
{'loss': '0.0111', 'grad_norm': '0.4202', 'learning_rate': '9.72e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 2452752, 'tokens/trainable': 36919, 'epoch': '0.3164'}
-
 16%|█▌        | 81/512 [09:11<48:18,  6.73s/it]
 16%|█▌        | 82/512 [09:17<47:08,  6.58s/it]
                                                
{'loss': '0.01844', 'grad_norm': '0.6295', 'learning_rate': '9.71e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.37', 'tokens/total': 2481216, 'tokens/trainable': 37379, 'epoch': '0.3203'}
-
 16%|█▌        | 82/512 [09:17<47:08,  6.58s/it]
 16%|█▌        | 83/512 [09:24<47:22,  6.63s/it]
                                                
{'loss': '0.001461', 'grad_norm': '0.1042', 'learning_rate': '9.699e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.98', 'tokens/total': 2511488, 'tokens/trainable': 37832, 'epoch': '0.3242'}
-
 16%|█▌        | 83/512 [09:24<47:22,  6.63s/it]
 16%|█▋        | 84/512 [09:31<47:30,  6.66s/it]
                                                
{'loss': '0.001734', 'grad_norm': '0.09028', 'learning_rate': '9.689e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.84', 'tokens/total': 2541984, 'tokens/trainable': 38303, 'epoch': '0.3281'}
-
 16%|█▋        | 84/512 [09:31<47:30,  6.66s/it]
 17%|█▋        | 85/512 [09:37<47:30,  6.67s/it]
                                                
{'loss': '0.004695', 'grad_norm': '0.5653', 'learning_rate': '9.678e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.54', 'tokens/total': 2572224, 'tokens/trainable': 38758, 'epoch': '0.332'}
-
 17%|█▋        | 85/512 [09:37<47:30,  6.67s/it]
 17%|█▋        | 86/512 [09:44<47:28,  6.69s/it]
                                                
{'loss': '0.002384', 'grad_norm': '0.1422', 'learning_rate': '9.667e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 2602704, 'tokens/trainable': 39179, 'epoch': '0.3359'}
-
 17%|█▋        | 86/512 [09:44<47:28,  6.69s/it]
 17%|█▋        | 87/512 [09:51<47:26,  6.70s/it]
                                                
{'loss': '0.01067', 'grad_norm': '0.6705', 'learning_rate': '9.656e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 2633072, 'tokens/trainable': 39679, 'epoch': '0.3398'}
-
 17%|█▋        | 87/512 [09:51<47:26,  6.70s/it]
 17%|█▋        | 88/512 [09:57<47:08,  6.67s/it]
                                                
{'loss': '0.01595', 'grad_norm': '0.7847', 'learning_rate': '9.645e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.69', 'tokens/total': 2663232, 'tokens/trainable': 40118, 'epoch': '0.3438'}
-
 17%|█▋        | 88/512 [09:57<47:08,  6.67s/it]
 17%|█▋        | 89/512 [10:04<47:03,  6.67s/it]
                                                
{'loss': '0.002848', 'grad_norm': '0.1816', 'learning_rate': '9.633e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 2693696, 'tokens/trainable': 40571, 'epoch': '0.3477'}
-
 17%|█▋        | 89/512 [10:04<47:03,  6.67s/it]
 18%|█▊        | 90/512 [10:11<46:59,  6.68s/it]
                                                
{'loss': '0.001987', 'grad_norm': '0.1428', 'learning_rate': '9.622e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.76', 'tokens/total': 2723984, 'tokens/trainable': 40997, 'epoch': '0.3516'}
-
 18%|█▊        | 90/512 [10:11<46:59,  6.68s/it]
 18%|█▊        | 91/512 [10:17<46:58,  6.69s/it]
                                                
{'loss': '0.007046', 'grad_norm': '0.525', 'learning_rate': '9.61e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.45', 'tokens/total': 2754240, 'tokens/trainable': 41462, 'epoch': '0.3555'}
-
 18%|█▊        | 91/512 [10:17<46:58,  6.69s/it]
 18%|█▊        | 92/512 [10:24<46:50,  6.69s/it]
                                                
{'loss': '0.006979', 'grad_norm': '0.5962', 'learning_rate': '9.598e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.81', 'tokens/total': 2784672, 'tokens/trainable': 41927, 'epoch': '0.3594'}
-
 18%|█▊        | 92/512 [10:24<46:50,  6.69s/it]
 18%|█▊        | 93/512 [10:31<46:48,  6.70s/it]
                                                
{'loss': '0.002276', 'grad_norm': '0.2272', 'learning_rate': '9.586e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.8', 'tokens/total': 2815120, 'tokens/trainable': 42397, 'epoch': '0.3633'}
-
 18%|█▊        | 93/512 [10:31<46:48,  6.70s/it]
 18%|█▊        | 94/512 [10:38<46:45,  6.71s/it]
                                                
{'loss': '0.0007734', 'grad_norm': '0.04159', 'learning_rate': '9.574e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 2845600, 'tokens/trainable': 42883, 'epoch': '0.3672'}
-
 18%|█▊        | 94/512 [10:38<46:45,  6.71s/it]
 19%|█▊        | 95/512 [10:44<46:44,  6.72s/it]
                                                
{'loss': '0.003535', 'grad_norm': '0.2429', 'learning_rate': '9.562e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.65', 'tokens/total': 2875952, 'tokens/trainable': 43381, 'epoch': '0.3711'}
-
 19%|█▊        | 95/512 [10:44<46:44,  6.72s/it]
 19%|█▉        | 96/512 [10:51<46:39,  6.73s/it]
                                                
{'loss': '0.0004532', 'grad_norm': '0.05062', 'learning_rate': '9.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 2906288, 'tokens/trainable': 43824, 'epoch': '0.375'}
-
 19%|█▉        | 96/512 [10:51<46:39,  6.73s/it][2026-08-18 12:50:07,343] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
-
 19%|█▉        | 97/512 [11:00<50:11,  7.26s/it]
                                                
{'loss': '0.0004007', 'grad_norm': '0.06917', 'learning_rate': '9.536e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '31.5', 'tokens/total': 2936752, 'tokens/trainable': 44255, 'epoch': '0.3789'}
-
 19%|█▉        | 97/512 [11:00<50:11,  7.26s/it]
 19%|█▉        | 98/512 [11:06<48:56,  7.09s/it]
                                                
{'loss': '0.002073', 'grad_norm': '0.2096', 'learning_rate': '9.523e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.19', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 2967072, 'tokens/trainable': 44739, 'epoch': '0.3828'}
-
 19%|█▉        | 98/512 [11:06<48:56,  7.09s/it]
 19%|█▉        | 99/512 [11:13<48:15,  7.01s/it]
                                                
{'loss': '0.0004412', 'grad_norm': '0.07389', 'learning_rate': '9.51e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '35.76', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 2997664, 'tokens/trainable': 45202, 'epoch': '0.3867'}
-
 19%|█▉        | 99/512 [11:13<48:15,  7.01s/it]
 20%|█▉        | 100/512 [11:20<47:29,  6.92s/it]
                                                 
{'loss': '0.00618', 'grad_norm': '0.4641', 'learning_rate': '9.497e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 3028032, 'tokens/trainable': 45646, 'epoch': '0.3906'}
-
 20%|█▉        | 100/512 [11:20<47:29,  6.92s/it]
 20%|█▉        | 101/512 [11:27<46:58,  6.86s/it]
                                                 
{'loss': '0.0001253', 'grad_norm': '0.01903', 'learning_rate': '9.484e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.04', 'tokens/total': 3058272, 'tokens/trainable': 46089, 'epoch': '0.3945'}
-
 20%|█▉        | 101/512 [11:27<46:58,  6.86s/it]
 20%|█▉        | 102/512 [11:33<46:37,  6.82s/it]
                                                 
{'loss': '0.03122', 'grad_norm': '3.644', 'learning_rate': '9.47e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 3088848, 'tokens/trainable': 46547, 'epoch': '0.3984'}
-
 20%|█▉        | 102/512 [11:33<46:37,  6.82s/it]
 20%|██        | 103/512 [11:40<46:16,  6.79s/it]
                                                 
{'loss': '0.01116', 'grad_norm': '1.234', 'learning_rate': '9.456e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '31.9', 'tokens/total': 3119424, 'tokens/trainable': 47014, 'epoch': '0.4023'}
-
 20%|██        | 103/512 [11:40<46:16,  6.79s/it]
 20%|██        | 104/512 [11:47<46:01,  6.77s/it]
                                                 
{'loss': '0.002164', 'grad_norm': '0.712', 'learning_rate': '9.442e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.65', 'tokens/total': 3149696, 'tokens/trainable': 47503, 'epoch': '0.4062'}
-
 20%|██        | 104/512 [11:47<46:01,  6.77s/it]
 21%|██        | 105/512 [11:53<45:50,  6.76s/it]
                                                 
{'loss': '0.0002605', 'grad_norm': '0.07053', 'learning_rate': '9.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.26', 'tokens/total': 3180064, 'tokens/trainable': 47996, 'epoch': '0.4102'}
-
 21%|██        | 105/512 [11:53<45:50,  6.76s/it]
 21%|██        | 106/512 [12:00<44:35,  6.59s/it]
                                                 
{'loss': '0.0164', 'grad_norm': '1.238', 'learning_rate': '9.414e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 3208320, 'tokens/trainable': 48447, 'epoch': '0.4141'}
-
 21%|██        | 106/512 [12:00<44:35,  6.59s/it]
 21%|██        | 107/512 [12:06<44:41,  6.62s/it]
                                                 
{'loss': '0.01001', 'grad_norm': '0.5694', 'learning_rate': '9.4e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.65', 'tokens/total': 3238416, 'tokens/trainable': 48889, 'epoch': '0.418'}
-
 21%|██        | 107/512 [12:06<44:41,  6.62s/it]
 21%|██        | 108/512 [12:13<44:46,  6.65s/it]
                                                 
{'loss': '0.002217', 'grad_norm': '0.5011', 'learning_rate': '9.385e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.84', 'tokens/total': 3268784, 'tokens/trainable': 49360, 'epoch': '0.4219'}
-
 21%|██        | 108/512 [12:13<44:46,  6.65s/it]
 21%|██▏       | 109/512 [12:20<44:48,  6.67s/it]
                                                 
{'loss': '0.006433', 'grad_norm': '0.6332', 'learning_rate': '9.37e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 3298912, 'tokens/trainable': 49827, 'epoch': '0.4258'}
-
 21%|██▏       | 109/512 [12:20<44:48,  6.67s/it]
 21%|██▏       | 110/512 [12:27<44:48,  6.69s/it]
                                                 
{'loss': '0.02577', 'grad_norm': '3.647', 'learning_rate': '9.355e-05', 'ppl': '1.026', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 3329232, 'tokens/trainable': 50267, 'epoch': '0.4297'}
-
 21%|██▏       | 110/512 [12:27<44:48,  6.69s/it]
 22%|██▏       | 111/512 [12:33<44:46,  6.70s/it]
                                                 
{'loss': '0.01398', 'grad_norm': '1.08', 'learning_rate': '9.34e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.32', 'tokens/total': 3359440, 'tokens/trainable': 50711, 'epoch': '0.4336'}
-
 22%|██▏       | 111/512 [12:33<44:46,  6.70s/it]
 22%|██▏       | 112/512 [12:40<44:38,  6.70s/it]
                                                 
{'loss': '0.008084', 'grad_norm': '0.3129', 'learning_rate': '9.325e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.54', 'tokens/total': 3389472, 'tokens/trainable': 51157, 'epoch': '0.4375'}
-
 22%|██▏       | 112/512 [12:40<44:38,  6.70s/it]
 22%|██▏       | 113/512 [12:47<44:36,  6.71s/it]
                                                 
{'loss': '0.01252', 'grad_norm': '0.5562', 'learning_rate': '9.31e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.67', 'tokens/total': 3419920, 'tokens/trainable': 51613, 'epoch': '0.4414'}
-
 22%|██▏       | 113/512 [12:47<44:36,  6.71s/it]
 22%|██▏       | 114/512 [12:53<44:26,  6.70s/it]
                                                 
{'loss': '0.01198', 'grad_norm': '0.3786', 'learning_rate': '9.294e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '29.04', 'tokens/total': 3450176, 'tokens/trainable': 52017, 'epoch': '0.4453'}
-
 22%|██▏       | 114/512 [12:53<44:26,  6.70s/it]
 22%|██▏       | 115/512 [13:00<44:22,  6.71s/it]
                                                 
{'loss': '0.003319', 'grad_norm': '0.1905', 'learning_rate': '9.278e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 3480544, 'tokens/trainable': 52506, 'epoch': '0.4492'}
-
 22%|██▏       | 115/512 [13:00<44:22,  6.71s/it]
 23%|██▎       | 116/512 [13:07<44:19,  6.72s/it]
                                                 
{'loss': '0.01132', 'grad_norm': '0.9839', 'learning_rate': '9.263e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '38.61', 'tokens/total': 3510848, 'tokens/trainable': 53008, 'epoch': '0.4531'}
-
 23%|██▎       | 116/512 [13:07<44:19,  6.72s/it]
 23%|██▎       | 117/512 [13:14<44:12,  6.72s/it]
                                                 
{'loss': '0.01193', 'grad_norm': '1.508', 'learning_rate': '9.247e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 3541296, 'tokens/trainable': 53467, 'epoch': '0.457'}
-
 23%|██▎       | 117/512 [13:14<44:12,  6.72s/it]
 23%|██▎       | 118/512 [13:20<44:09,  6.72s/it]
                                                 
{'loss': '0.002887', 'grad_norm': '0.1703', 'learning_rate': '9.23e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 3571936, 'tokens/trainable': 53923, 'epoch': '0.4609'}
-
 23%|██▎       | 118/512 [13:20<44:09,  6.72s/it]
 23%|██▎       | 119/512 [13:27<43:55,  6.71s/it]
                                                 
{'loss': '0.003508', 'grad_norm': '0.3885', 'learning_rate': '9.214e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 3602000, 'tokens/trainable': 54363, 'epoch': '0.4648'}
-
 23%|██▎       | 119/512 [13:27<43:55,  6.71s/it]
 23%|██▎       | 120/512 [13:34<43:48,  6.71s/it]
                                                 
{'loss': '0.005708', 'grad_norm': '0.2821', 'learning_rate': '9.198e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 3632288, 'tokens/trainable': 54839, 'epoch': '0.4688'}
-
 23%|██▎       | 120/512 [13:34<43:48,  6.71s/it]
 24%|██▎       | 121/512 [13:40<43:40,  6.70s/it]
                                                 
{'loss': '0.003157', 'grad_norm': '0.3879', 'learning_rate': '9.181e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.58', 'tokens/total': 3662608, 'tokens/trainable': 55267, 'epoch': '0.4727'}
-
 24%|██▎       | 121/512 [13:40<43:40,  6.70s/it]
 24%|██▍       | 122/512 [13:47<43:33,  6.70s/it]
                                                 
{'loss': '0.00323', 'grad_norm': '0.2678', 'learning_rate': '9.164e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.47', 'tokens/total': 3692864, 'tokens/trainable': 55700, 'epoch': '0.4766'}
-
 24%|██▍       | 122/512 [13:47<43:33,  6.70s/it]
 24%|██▍       | 123/512 [13:54<43:20,  6.69s/it]
                                                 
{'loss': '0.004257', 'grad_norm': '0.31', 'learning_rate': '9.147e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.38', 'tokens/total': 3722864, 'tokens/trainable': 56144, 'epoch': '0.4805'}
-
 24%|██▍       | 123/512 [13:54<43:20,  6.69s/it]
 24%|██▍       | 124/512 [14:00<43:14,  6.69s/it]
                                                 
{'loss': '0.0001001', 'grad_norm': '0.03114', 'learning_rate': '9.13e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.07', 'tokens/total': 3753056, 'tokens/trainable': 56589, 'epoch': '0.4844'}
-
 24%|██▍       | 124/512 [14:00<43:14,  6.69s/it]
 24%|██▍       | 125/512 [14:07<43:13,  6.70s/it]
                                                 
{'loss': '0.004621', 'grad_norm': '0.736', 'learning_rate': '9.113e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.64', 'tokens/total': 3783312, 'tokens/trainable': 57081, 'epoch': '0.4883'}
-
 24%|██▍       | 125/512 [14:07<43:13,  6.70s/it]
 25%|██▍       | 126/512 [14:14<43:04,  6.70s/it]
                                                 
{'loss': '0.0002661', 'grad_norm': '0.05499', 'learning_rate': '9.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 3813712, 'tokens/trainable': 57527, 'epoch': '0.4922'}
-
 25%|██▍       | 126/512 [14:14<43:04,  6.70s/it]
 25%|██▍       | 127/512 [14:21<43:04,  6.71s/it]
                                                 
{'loss': '0.003466', 'grad_norm': '0.416', 'learning_rate': '9.078e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 3844336, 'tokens/trainable': 58005, 'epoch': '0.4961'}
-
 25%|██▍       | 127/512 [14:21<43:04,  6.71s/it]
 25%|██▌       | 128/512 [14:27<41:58,  6.56s/it]
                                                 
{'loss': '0.002959', 'grad_norm': '0.4412', 'learning_rate': '9.06e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 3872688, 'tokens/trainable': 58454, 'epoch': '0.5'}
-
 25%|██▌       | 128/512 [14:27<41:58,  6.56s/it][2026-08-18 12:53:42,959] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
-
 25%|██▌       | 129/512 [14:35<45:35,  7.14s/it]
                                                 
{'loss': '0.002471', 'grad_norm': '0.2943', 'learning_rate': '9.043e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 3902640, 'tokens/trainable': 58896, 'epoch': '0.5039'}
-
 25%|██▌       | 129/512 [14:35<45:35,  7.14s/it]
 25%|██▌       | 130/512 [14:42<44:38,  7.01s/it]
                                                 
{'loss': '0.0001329', 'grad_norm': '0.02422', 'learning_rate': '9.025e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '35.79', 'tokens/total': 3933104, 'tokens/trainable': 59356, 'epoch': '0.5078'}
-
 25%|██▌       | 130/512 [14:42<44:38,  7.01s/it]
 26%|██▌       | 131/512 [14:49<43:58,  6.93s/it]
                                                 
{'loss': '0.01349', 'grad_norm': '3.24', 'learning_rate': '9.007e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '32.14', 'tokens/total': 3963184, 'tokens/trainable': 59792, 'epoch': '0.5117'}
-
 26%|██▌       | 131/512 [14:49<43:58,  6.93s/it]
 26%|██▌       | 132/512 [14:55<43:26,  6.86s/it]
                                                 
{'loss': '0.0009926', 'grad_norm': '0.1477', 'learning_rate': '8.988e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.66', 'tokens/total': 3993648, 'tokens/trainable': 60266, 'epoch': '0.5156'}
-
 26%|██▌       | 132/512 [14:55<43:26,  6.86s/it]
 26%|██▌       | 133/512 [15:02<43:04,  6.82s/it]
                                                 
{'loss': '0.005481', 'grad_norm': '1.215', 'learning_rate': '8.97e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 4024000, 'tokens/trainable': 60759, 'epoch': '0.5195'}
-
 26%|██▌       | 133/512 [15:02<43:04,  6.82s/it]
 26%|██▌       | 134/512 [15:08<41:49,  6.64s/it]
                                                 
{'loss': '0.001138', 'grad_norm': '0.2438', 'learning_rate': '8.951e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 4052480, 'tokens/trainable': 61204, 'epoch': '0.5234'}
-
 26%|██▌       | 134/512 [15:08<41:49,  6.64s/it]
 26%|██▋       | 135/512 [15:15<41:54,  6.67s/it]
                                                 
{'loss': '0.004932', 'grad_norm': '0.8191', 'learning_rate': '8.933e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.88', 'tokens/total': 4082864, 'tokens/trainable': 61684, 'epoch': '0.5273'}
-
 26%|██▋       | 135/512 [15:15<41:54,  6.67s/it]
 27%|██▋       | 136/512 [15:22<41:46,  6.67s/it]
                                                 
{'loss': '0.009538', 'grad_norm': '0.7444', 'learning_rate': '8.914e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '38.18', 'tokens/total': 4113040, 'tokens/trainable': 62159, 'epoch': '0.5312'}
-
 27%|██▋       | 136/512 [15:22<41:46,  6.67s/it]
 27%|██▋       | 137/512 [15:28<41:45,  6.68s/it]
                                                 
{'loss': '0.0001035', 'grad_norm': '0.0178', 'learning_rate': '8.895e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 4143344, 'tokens/trainable': 62649, 'epoch': '0.5352'}
-
 27%|██▋       | 137/512 [15:28<41:45,  6.68s/it]
 27%|██▋       | 138/512 [15:35<41:41,  6.69s/it]
                                                 
{'loss': '0.0006722', 'grad_norm': '0.1413', 'learning_rate': '8.876e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 4173648, 'tokens/trainable': 63100, 'epoch': '0.5391'}
-
 27%|██▋       | 138/512 [15:35<41:41,  6.69s/it]
 27%|██▋       | 139/512 [15:42<41:41,  6.71s/it]
                                                 
{'loss': '0.00706', 'grad_norm': '0.7884', 'learning_rate': '8.856e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.51', 'tokens/total': 4204080, 'tokens/trainable': 63588, 'epoch': '0.543'}
-
 27%|██▋       | 139/512 [15:42<41:41,  6.71s/it]
 27%|██▋       | 140/512 [15:48<40:32,  6.54s/it]
                                                 
{'loss': '0.04659', 'grad_norm': '2.152', 'learning_rate': '8.837e-05', 'ppl': '1.048', 'memory/max_active (GiB)': '33.42', 'memory/max_allocated (GiB)': '33.42', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 4232400, 'tokens/trainable': 64005, 'epoch': '0.5469'}
-
 27%|██▋       | 140/512 [15:48<40:32,  6.54s/it]
 28%|██▊       | 141/512 [15:55<40:47,  6.60s/it]
                                                 
{'loss': '0.0185', 'grad_norm': '0.9182', 'learning_rate': '8.817e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 4262624, 'tokens/trainable': 64462, 'epoch': '0.5508'}
-
 28%|██▊       | 141/512 [15:55<40:47,  6.60s/it]
 28%|██▊       | 142/512 [16:02<40:55,  6.64s/it]
                                                 
{'loss': '0.0007529', 'grad_norm': '0.04982', 'learning_rate': '8.798e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 4293168, 'tokens/trainable': 64900, 'epoch': '0.5547'}
-
 28%|██▊       | 142/512 [16:02<40:55,  6.64s/it]
 28%|██▊       | 143/512 [16:08<40:56,  6.66s/it]
                                                 
{'loss': '0.01302', 'grad_norm': '0.4962', 'learning_rate': '8.778e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.45', 'tokens/total': 4323248, 'tokens/trainable': 65379, 'epoch': '0.5586'}
-
 28%|██▊       | 143/512 [16:08<40:56,  6.66s/it]
 28%|██▊       | 144/512 [16:15<40:52,  6.66s/it]
                                                 
{'loss': '0.001344', 'grad_norm': '0.09981', 'learning_rate': '8.758e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 4353344, 'tokens/trainable': 65845, 'epoch': '0.5625'}
-
 28%|██▊       | 144/512 [16:15<40:52,  6.66s/it]
 28%|██▊       | 145/512 [16:22<40:49,  6.67s/it]
                                                 
{'loss': '0.005669', 'grad_norm': '0.25', 'learning_rate': '8.738e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 4383584, 'tokens/trainable': 66285, 'epoch': '0.5664'}
-
 28%|██▊       | 145/512 [16:22<40:49,  6.67s/it]
 29%|██▊       | 146/512 [16:28<40:49,  6.69s/it]
                                                 
{'loss': '0.03146', 'grad_norm': '0.8081', 'learning_rate': '8.718e-05', 'ppl': '1.032', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 4413856, 'tokens/trainable': 66736, 'epoch': '0.5703'}
-
 29%|██▊       | 146/512 [16:28<40:49,  6.69s/it]
 29%|██▊       | 147/512 [16:35<40:44,  6.70s/it]
                                                 
{'loss': '0.008184', 'grad_norm': '0.2932', 'learning_rate': '8.697e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.2', 'tokens/total': 4444112, 'tokens/trainable': 67150, 'epoch': '0.5742'}
-
 29%|██▊       | 147/512 [16:35<40:44,  6.70s/it]
 29%|██▉       | 148/512 [16:42<40:45,  6.72s/it]
                                                 
{'loss': '0.008484', 'grad_norm': '0.5703', 'learning_rate': '8.677e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 4474464, 'tokens/trainable': 67633, 'epoch': '0.5781'}
-
 29%|██▉       | 148/512 [16:42<40:45,  6.72s/it]
 29%|██▉       | 149/512 [16:49<40:39,  6.72s/it]
                                                 
{'loss': '0.002346', 'grad_norm': '0.1261', 'learning_rate': '8.656e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 4504880, 'tokens/trainable': 68077, 'epoch': '0.582'}
-
 29%|██▉       | 149/512 [16:49<40:39,  6.72s/it]
 29%|██▉       | 150/512 [16:55<40:28,  6.71s/it]
                                                 
{'loss': '0.0008666', 'grad_norm': '0.04483', 'learning_rate': '8.635e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.21', 'tokens/total': 4534976, 'tokens/trainable': 68545, 'epoch': '0.5859'}
-
 29%|██▉       | 150/512 [16:55<40:28,  6.71s/it]
 29%|██▉       | 151/512 [17:02<40:24,  6.72s/it]
                                                 
{'loss': '0.01069', 'grad_norm': '0.2727', 'learning_rate': '8.615e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.17', 'tokens/total': 4565088, 'tokens/trainable': 69011, 'epoch': '0.5898'}
-
 29%|██▉       | 151/512 [17:02<40:24,  6.72s/it]
 30%|██▉       | 152/512 [17:09<40:20,  6.72s/it]
                                                 
{'loss': '0.007179', 'grad_norm': '0.6334', 'learning_rate': '8.594e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 4595600, 'tokens/trainable': 69465, 'epoch': '0.5938'}
-
 30%|██▉       | 152/512 [17:09<40:20,  6.72s/it]
 30%|██▉       | 153/512 [17:15<40:18,  6.74s/it]
                                                 
{'loss': '0.005559', 'grad_norm': '0.6208', 'learning_rate': '8.572e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.83', 'tokens/total': 4626032, 'tokens/trainable': 69928, 'epoch': '0.5977'}
-
 30%|██▉       | 153/512 [17:15<40:18,  6.74s/it]
 30%|███       | 154/512 [17:22<40:09,  6.73s/it]
                                                 
{'loss': '0.0009', 'grad_norm': '0.1435', 'learning_rate': '8.551e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.64', 'tokens/total': 4656352, 'tokens/trainable': 70395, 'epoch': '0.6016'}
-
 30%|███       | 154/512 [17:22<40:09,  6.73s/it]
 30%|███       | 155/512 [17:28<39:05,  6.57s/it]
                                                 
{'loss': '0.003266', 'grad_norm': '0.3754', 'learning_rate': '8.53e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 4684672, 'tokens/trainable': 70829, 'epoch': '0.6055'}
-
 30%|███       | 155/512 [17:28<39:05,  6.57s/it]
 30%|███       | 156/512 [17:35<39:18,  6.62s/it]
                                                 
{'loss': '0.01512', 'grad_norm': '0.8183', 'learning_rate': '8.508e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.8', 'tokens/total': 4715216, 'tokens/trainable': 71287, 'epoch': '0.6094'}
-
 30%|███       | 156/512 [17:35<39:18,  6.62s/it]
 31%|███       | 157/512 [17:42<39:22,  6.65s/it]
                                                 
{'loss': '0.0002059', 'grad_norm': '0.05925', 'learning_rate': '8.487e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 4745376, 'tokens/trainable': 71749, 'epoch': '0.6133'}
-
 31%|███       | 157/512 [17:42<39:22,  6.65s/it]
 31%|███       | 158/512 [17:49<39:28,  6.69s/it]
                                                 
{'loss': '0.01122', 'grad_norm': '0.5881', 'learning_rate': '8.465e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 4775776, 'tokens/trainable': 72185, 'epoch': '0.6172'}
-
 31%|███       | 158/512 [17:49<39:28,  6.69s/it]
 31%|███       | 159/512 [17:55<39:18,  6.68s/it]
                                                 
{'loss': '0.001423', 'grad_norm': '0.188', 'learning_rate': '8.443e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.51', 'tokens/total': 4805856, 'tokens/trainable': 72604, 'epoch': '0.6211'}
-
 31%|███       | 159/512 [17:55<39:18,  6.68s/it]
 31%|███▏      | 160/512 [18:02<39:18,  6.70s/it]
                                                 
{'loss': '0.01892', 'grad_norm': '0.8338', 'learning_rate': '8.421e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.41', 'tokens/total': 4836304, 'tokens/trainable': 73061, 'epoch': '0.625'}
-
 31%|███▏      | 160/512 [18:02<39:18,  6.70s/it][2026-08-18 12:57:18,242] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
-
 31%|███▏      | 161/512 [18:11<42:48,  7.32s/it]
                                                 
{'loss': '0.001153', 'grad_norm': '0.1115', 'learning_rate': '8.399e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.97', 'tokens/total': 4866544, 'tokens/trainable': 73474, 'epoch': '0.6289'}
-
 31%|███▏      | 161/512 [18:11<42:48,  7.32s/it]
 32%|███▏      | 162/512 [18:18<41:41,  7.15s/it]
                                                 
{'loss': '0.00308', 'grad_norm': '0.3199', 'learning_rate': '8.376e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.61', 'tokens/total': 4897120, 'tokens/trainable': 73923, 'epoch': '0.6328'}
-
 32%|███▏      | 162/512 [18:18<41:41,  7.15s/it]
 32%|███▏      | 163/512 [18:24<40:49,  7.02s/it]
                                                 
{'loss': '0.002999', 'grad_norm': '0.2323', 'learning_rate': '8.354e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 4927440, 'tokens/trainable': 74379, 'epoch': '0.6367'}
-
 32%|███▏      | 163/512 [18:24<40:49,  7.02s/it]
 32%|███▏      | 164/512 [18:31<40:27,  6.97s/it]
                                                 
{'loss': '0.003661', 'grad_norm': '0.1695', 'learning_rate': '8.332e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 4958032, 'tokens/trainable': 74830, 'epoch': '0.6406'}
-
 32%|███▏      | 164/512 [18:31<40:27,  6.97s/it]
 32%|███▏      | 165/512 [18:38<39:53,  6.90s/it]
                                                 
{'loss': '0.001191', 'grad_norm': '0.0809', 'learning_rate': '8.309e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 4988368, 'tokens/trainable': 75262, 'epoch': '0.6445'}
-
 32%|███▏      | 165/512 [18:38<39:53,  6.90s/it]
 32%|███▏      | 166/512 [18:45<39:23,  6.83s/it]
                                                 
{'loss': '0.001098', 'grad_norm': '0.06333', 'learning_rate': '8.286e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.75', 'tokens/total': 5018672, 'tokens/trainable': 75703, 'epoch': '0.6484'}
-
 32%|███▏      | 166/512 [18:45<39:23,  6.83s/it]
 33%|███▎      | 167/512 [18:51<39:15,  6.83s/it]
                                                 
{'loss': '0.0008549', 'grad_norm': '0.05864', 'learning_rate': '8.263e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.26', 'tokens/total': 5049344, 'tokens/trainable': 76164, 'epoch': '0.6523'}
-
 33%|███▎      | 167/512 [18:51<39:15,  6.83s/it]
 33%|███▎      | 168/512 [18:58<38:59,  6.80s/it]
                                                 
{'loss': '0.002419', 'grad_norm': '0.192', 'learning_rate': '8.24e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.42', 'tokens/total': 5079856, 'tokens/trainable': 76623, 'epoch': '0.6562'}
-
 33%|███▎      | 168/512 [18:58<38:59,  6.80s/it]
 33%|███▎      | 169/512 [19:05<38:39,  6.76s/it]
                                                 
{'loss': '0.000676', 'grad_norm': '0.03395', 'learning_rate': '8.217e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 5110192, 'tokens/trainable': 77047, 'epoch': '0.6602'}
-
 33%|███▎      | 169/512 [19:05<38:39,  6.76s/it]
 33%|███▎      | 170/512 [19:12<38:33,  6.76s/it]
                                                 
{'loss': '0.0006985', 'grad_norm': '0.1052', 'learning_rate': '8.194e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 5140592, 'tokens/trainable': 77496, 'epoch': '0.6641'}
-
 33%|███▎      | 170/512 [19:12<38:33,  6.76s/it]
 33%|███▎      | 171/512 [19:18<38:20,  6.75s/it]
                                                 
{'loss': '0.0002491', 'grad_norm': '0.01598', 'learning_rate': '8.171e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.13', 'tokens/total': 5170848, 'tokens/trainable': 77956, 'epoch': '0.668'}
-
 33%|███▎      | 171/512 [19:18<38:20,  6.75s/it]
 34%|███▎      | 172/512 [19:25<38:14,  6.75s/it]
                                                 
{'loss': '0.002719', 'grad_norm': '0.2858', 'learning_rate': '8.147e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.75', 'tokens/total': 5201280, 'tokens/trainable': 78410, 'epoch': '0.6719'}
-
 34%|███▎      | 172/512 [19:25<38:14,  6.75s/it]
 34%|███▍      | 173/512 [19:32<37:59,  6.72s/it]
                                                 
{'loss': '0.0005204', 'grad_norm': '0.02717', 'learning_rate': '8.124e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.04', 'tokens/total': 5231712, 'tokens/trainable': 78849, 'epoch': '0.6758'}
-
 34%|███▍      | 173/512 [19:32<37:59,  6.72s/it]
 34%|███▍      | 174/512 [19:38<37:47,  6.71s/it]
                                                 
{'loss': '0.01529', 'grad_norm': '2.88', 'learning_rate': '8.1e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.62', 'memory/max_allocated (GiB)': '33.62', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.53', 'tokens/total': 5261440, 'tokens/trainable': 79266, 'epoch': '0.6797'}
-
 34%|███▍      | 174/512 [19:38<37:47,  6.71s/it]
 34%|███▍      | 175/512 [19:45<37:42,  6.71s/it]
                                                 
{'loss': '0.001033', 'grad_norm': '0.1005', 'learning_rate': '8.076e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.42', 'tokens/total': 5291728, 'tokens/trainable': 79736, 'epoch': '0.6836'}
-
 34%|███▍      | 175/512 [19:45<37:42,  6.71s/it]
 34%|███▍      | 176/512 [19:52<37:40,  6.73s/it]
                                                 
{'loss': '0.003816', 'grad_norm': '0.6506', 'learning_rate': '8.053e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.12', 'tokens/total': 5322240, 'tokens/trainable': 80214, 'epoch': '0.6875'}
-
 34%|███▍      | 176/512 [19:52<37:40,  6.73s/it]
 35%|███▍      | 177/512 [19:59<37:31,  6.72s/it]
                                                 
{'loss': '0.002027', 'grad_norm': '0.199', 'learning_rate': '8.029e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.51', 'tokens/total': 5352576, 'tokens/trainable': 80661, 'epoch': '0.6914'}
-
 35%|███▍      | 177/512 [19:59<37:31,  6.72s/it]
 35%|███▍      | 178/512 [20:05<37:25,  6.72s/it]
                                                 
{'loss': '0.0001421', 'grad_norm': '0.00993', 'learning_rate': '8.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.81', 'tokens/total': 5382784, 'tokens/trainable': 81129, 'epoch': '0.6953'}
-
 35%|███▍      | 178/512 [20:05<37:25,  6.72s/it]
 35%|███▍      | 179/512 [20:12<37:11,  6.70s/it]
                                                 
{'loss': '0.02322', 'grad_norm': '0.5209', 'learning_rate': '7.98e-05', 'ppl': '1.024', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 5412784, 'tokens/trainable': 81576, 'epoch': '0.6992'}
-
 35%|███▍      | 179/512 [20:12<37:11,  6.70s/it]
 35%|███▌      | 180/512 [20:19<37:07,  6.71s/it]
                                                 
{'loss': '0.0002464', 'grad_norm': '0.02281', 'learning_rate': '7.956e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.49', 'tokens/total': 5442960, 'tokens/trainable': 82049, 'epoch': '0.7031'}
-
 35%|███▌      | 180/512 [20:19<37:07,  6.71s/it]
 35%|███▌      | 181/512 [20:25<37:05,  6.72s/it]
                                                 
{'loss': '0.01167', 'grad_norm': '0.377', 'learning_rate': '7.932e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '38.75', 'tokens/total': 5473216, 'tokens/trainable': 82566, 'epoch': '0.707'}
-
 35%|███▌      | 181/512 [20:25<37:05,  6.72s/it]
 36%|███▌      | 182/512 [20:32<37:00,  6.73s/it]
                                                 
{'loss': '0.0002162', 'grad_norm': '0.06076', 'learning_rate': '7.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.83', 'tokens/total': 5503568, 'tokens/trainable': 83054, 'epoch': '0.7109'}
-
 36%|███▌      | 182/512 [20:32<37:00,  6.73s/it]
 36%|███▌      | 183/512 [20:39<36:52,  6.73s/it]
                                                 
{'loss': '0.0004136', 'grad_norm': '0.03374', 'learning_rate': '7.883e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 5533760, 'tokens/trainable': 83501, 'epoch': '0.7148'}
-
 36%|███▌      | 183/512 [20:39<36:52,  6.73s/it]
 36%|███▌      | 184/512 [20:46<36:45,  6.72s/it]
                                                 
{'loss': '0.00041', 'grad_norm': '0.03666', 'learning_rate': '7.858e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 5564240, 'tokens/trainable': 84007, 'epoch': '0.7188'}
-
 36%|███▌      | 184/512 [20:46<36:45,  6.72s/it]
 36%|███▌      | 185/512 [20:52<36:41,  6.73s/it]
                                                 
{'loss': '0.0003714', 'grad_norm': '0.03234', 'learning_rate': '7.833e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 5594448, 'tokens/trainable': 84491, 'epoch': '0.7227'}
-
 36%|███▌      | 185/512 [20:52<36:41,  6.73s/it]
 36%|███▋      | 186/512 [20:59<36:36,  6.74s/it]
                                                 
{'loss': '0.0003759', 'grad_norm': '0.03077', 'learning_rate': '7.808e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 5624816, 'tokens/trainable': 84952, 'epoch': '0.7266'}
-
 36%|███▋      | 186/512 [20:59<36:36,  6.74s/it]
 37%|███▋      | 187/512 [21:06<36:28,  6.74s/it]
                                                 
{'loss': '0.002438', 'grad_norm': '0.6632', 'learning_rate': '7.783e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 5655104, 'tokens/trainable': 85400, 'epoch': '0.7305'}
-
 37%|███▋      | 187/512 [21:06<36:28,  6.74s/it]
 37%|███▋      | 188/512 [21:13<36:23,  6.74s/it]
                                                 
{'loss': '0.002341', 'grad_norm': '0.2652', 'learning_rate': '7.758e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 5685360, 'tokens/trainable': 85889, 'epoch': '0.7344'}
-
 37%|███▋      | 188/512 [21:13<36:23,  6.74s/it]
 37%|███▋      | 189/512 [21:19<36:18,  6.75s/it]
                                                 
{'loss': '0.001637', 'grad_norm': '0.3055', 'learning_rate': '7.733e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.11', 'tokens/total': 5715648, 'tokens/trainable': 86396, 'epoch': '0.7383'}
-
 37%|███▋      | 189/512 [21:19<36:18,  6.75s/it]
 37%|███▋      | 190/512 [21:26<36:10,  6.74s/it]
                                                 
{'loss': '0.00207', 'grad_norm': '0.2344', 'learning_rate': '7.708e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.75', 'tokens/total': 5746240, 'tokens/trainable': 86867, 'epoch': '0.7422'}
-
 37%|███▋      | 190/512 [21:26<36:10,  6.74s/it]
 37%|███▋      | 191/512 [21:33<36:07,  6.75s/it]
                                                 
{'loss': '0.0006176', 'grad_norm': '0.09677', 'learning_rate': '7.683e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.03', 'tokens/total': 5776816, 'tokens/trainable': 87314, 'epoch': '0.7461'}
-
 37%|███▋      | 191/512 [21:33<36:07,  6.75s/it]
 38%|███▊      | 192/512 [21:39<35:55,  6.74s/it]
                                                 
{'loss': '0.000425', 'grad_norm': '0.05828', 'learning_rate': '7.657e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.29', 'tokens/total': 5807248, 'tokens/trainable': 87736, 'epoch': '0.75'}
-
 38%|███▊      | 192/512 [21:40<35:55,  6.74s/it][2026-08-18 13:00:55,710] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
-
 38%|███▊      | 193/512 [21:48<38:51,  7.31s/it]
                                                 
{'loss': '0.000321', 'grad_norm': '0.09241', 'learning_rate': '7.632e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 5837680, 'tokens/trainable': 88219, 'epoch': '0.7539'}
-
 38%|███▊      | 193/512 [21:48<38:51,  7.31s/it]
 38%|███▊      | 194/512 [21:55<37:51,  7.14s/it]
                                                 
{'loss': '7.824e-05', 'grad_norm': '0.005906', 'learning_rate': '7.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '36.11', 'tokens/total': 5868000, 'tokens/trainable': 88699, 'epoch': '0.7578'}
-
 38%|███▊      | 194/512 [21:55<37:51,  7.14s/it]
 38%|███▊      | 195/512 [22:01<36:11,  6.85s/it]
                                                 
{'loss': '0.001962', 'grad_norm': '0.4526', 'learning_rate': '7.58e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.36', 'tokens/total': 5896224, 'tokens/trainable': 89132, 'epoch': '0.7617'}
-
 38%|███▊      | 195/512 [22:01<36:11,  6.85s/it]
 38%|███▊      | 196/512 [22:08<35:52,  6.81s/it]
                                                 
{'loss': '0.000491', 'grad_norm': '0.03763', 'learning_rate': '7.555e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '37.21', 'tokens/total': 5926624, 'tokens/trainable': 89594, 'epoch': '0.7656'}
-
 38%|███▊      | 196/512 [22:08<35:52,  6.81s/it]
 38%|███▊      | 197/512 [22:14<34:49,  6.63s/it]
                                                 
{'loss': '0.005962', 'grad_norm': '0.3791', 'learning_rate': '7.529e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.34', 'memory/max_allocated (GiB)': '33.34', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '37.98', 'tokens/total': 5954944, 'tokens/trainable': 90042, 'epoch': '0.7695'}
-
 38%|███▊      | 197/512 [22:14<34:49,  6.63s/it]
 39%|███▊      | 198/512 [22:21<34:52,  6.66s/it]
                                                 
{'loss': '0.0004046', 'grad_norm': '0.06128', 'learning_rate': '7.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 5985200, 'tokens/trainable': 90503, 'epoch': '0.7734'}
-
 39%|███▊      | 198/512 [22:21<34:52,  6.66s/it]
 39%|███▉      | 199/512 [22:28<35:03,  6.72s/it]
                                                 
{'loss': '0.0001899', 'grad_norm': '0.01201', 'learning_rate': '7.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 6015952, 'tokens/trainable': 90982, 'epoch': '0.7773'}
-
 39%|███▉      | 199/512 [22:28<35:03,  6.72s/it]
 39%|███▉      | 200/512 [22:34<34:58,  6.73s/it]
                                                 
{'loss': '0.0003557', 'grad_norm': '0.03914', 'learning_rate': '7.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '39.22', 'tokens/total': 6046208, 'tokens/trainable': 91466, 'epoch': '0.7812'}
-
 39%|███▉      | 200/512 [22:34<34:58,  6.73s/it]
 39%|███▉      | 201/512 [22:41<34:46,  6.71s/it]
                                                 
{'loss': '0.0002456', 'grad_norm': '0.04544', 'learning_rate': '7.424e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 6076208, 'tokens/trainable': 91912, 'epoch': '0.7852'}
-
 39%|███▉      | 201/512 [22:41<34:46,  6.71s/it]
 39%|███▉      | 202/512 [22:48<34:38,  6.71s/it]
                                                 
{'loss': '0.009876', 'grad_norm': '0.6255', 'learning_rate': '7.398e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.79', 'tokens/total': 6106560, 'tokens/trainable': 92385, 'epoch': '0.7891'}
-
 39%|███▉      | 202/512 [22:48<34:38,  6.71s/it]
 40%|███▉      | 203/512 [22:54<34:30,  6.70s/it]
                                                 
{'loss': '0.01162', 'grad_norm': '0.7777', 'learning_rate': '7.372e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.48', 'tokens/total': 6136640, 'tokens/trainable': 92842, 'epoch': '0.793'}
-
 40%|███▉      | 203/512 [22:54<34:30,  6.70s/it]
 40%|███▉      | 204/512 [23:01<34:26,  6.71s/it]
                                                 
{'loss': '0.0005008', 'grad_norm': '0.07255', 'learning_rate': '7.345e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 6167184, 'tokens/trainable': 93323, 'epoch': '0.7969'}
-
 40%|███▉      | 204/512 [23:01<34:26,  6.71s/it]
 40%|████      | 205/512 [23:08<34:14,  6.69s/it]
                                                 
{'loss': '0.007061', 'grad_norm': '0.6082', 'learning_rate': '7.319e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.75', 'tokens/total': 6197376, 'tokens/trainable': 93756, 'epoch': '0.8008'}
-
 40%|████      | 205/512 [23:08<34:14,  6.69s/it]
 40%|████      | 206/512 [23:15<34:13,  6.71s/it]
                                                 
{'loss': '0.006006', 'grad_norm': '0.4931', 'learning_rate': '7.292e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.47', 'tokens/total': 6227600, 'tokens/trainable': 94235, 'epoch': '0.8047'}
-
 40%|████      | 206/512 [23:15<34:13,  6.71s/it]
 40%|████      | 207/512 [23:21<34:07,  6.71s/it]
                                                 
{'loss': '0.0006859', 'grad_norm': '0.06813', 'learning_rate': '7.266e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.48', 'tokens/total': 6257760, 'tokens/trainable': 94733, 'epoch': '0.8086'}
-
 40%|████      | 207/512 [23:21<34:07,  6.71s/it]
 41%|████      | 208/512 [23:28<34:00,  6.71s/it]
                                                 
{'loss': '0.001039', 'grad_norm': '0.06704', 'learning_rate': '7.239e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.68', 'tokens/total': 6288048, 'tokens/trainable': 95218, 'epoch': '0.8125'}
-
 41%|████      | 208/512 [23:28<34:00,  6.71s/it]
 41%|████      | 209/512 [23:35<33:48,  6.69s/it]
                                                 
{'loss': '0.003791', 'grad_norm': '0.2953', 'learning_rate': '7.212e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '29.17', 'tokens/total': 6318688, 'tokens/trainable': 95619, 'epoch': '0.8164'}
-
 41%|████      | 209/512 [23:35<33:48,  6.69s/it]
 41%|████      | 210/512 [23:41<33:43,  6.70s/it]
                                                 
{'loss': '0.002394', 'grad_norm': '0.3258', 'learning_rate': '7.185e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.32', 'tokens/total': 6348944, 'tokens/trainable': 96076, 'epoch': '0.8203'}
-
 41%|████      | 210/512 [23:41<33:43,  6.70s/it]
 41%|████      | 211/512 [23:48<33:33,  6.69s/it]
                                                 
{'loss': '0.006002', 'grad_norm': '0.4735', 'learning_rate': '7.158e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.28', 'tokens/total': 6378960, 'tokens/trainable': 96505, 'epoch': '0.8242'}
-
 41%|████      | 211/512 [23:48<33:33,  6.69s/it]
 41%|████▏     | 212/512 [23:55<33:29,  6.70s/it]
                                                 
{'loss': '0.003804', 'grad_norm': '0.1573', 'learning_rate': '7.131e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.57', 'tokens/total': 6409472, 'tokens/trainable': 96930, 'epoch': '0.8281'}
-
 41%|████▏     | 212/512 [23:55<33:29,  6.70s/it]
 42%|████▏     | 213/512 [24:01<33:24,  6.70s/it]
                                                 
{'loss': '0.005038', 'grad_norm': '0.1833', 'learning_rate': '7.104e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 6439904, 'tokens/trainable': 97361, 'epoch': '0.832'}
-
 42%|████▏     | 213/512 [24:01<33:24,  6.70s/it]
 42%|████▏     | 214/512 [24:08<33:22,  6.72s/it]
                                                 
{'loss': '0.002086', 'grad_norm': '0.1782', 'learning_rate': '7.077e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.26', 'tokens/total': 6470176, 'tokens/trainable': 97858, 'epoch': '0.8359'}
-
 42%|████▏     | 214/512 [24:08<33:22,  6.72s/it]
 42%|████▏     | 215/512 [24:15<33:10,  6.70s/it]
                                                 
{'loss': '0.005029', 'grad_norm': '0.2747', 'learning_rate': '7.05e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.41', 'tokens/total': 6500336, 'tokens/trainable': 98308, 'epoch': '0.8398'}
-
 42%|████▏     | 215/512 [24:15<33:10,  6.70s/it]
 42%|████▏     | 216/512 [24:22<33:02,  6.70s/it]
                                                 
{'loss': '0.005489', 'grad_norm': '0.4509', 'learning_rate': '7.022e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 6530688, 'tokens/trainable': 98781, 'epoch': '0.8438'}
-
 42%|████▏     | 216/512 [24:22<33:02,  6.70s/it]
 42%|████▏     | 217/512 [24:28<32:59,  6.71s/it]
                                                 
{'loss': '0.003419', 'grad_norm': '0.2319', 'learning_rate': '6.995e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.67', 'tokens/total': 6561040, 'tokens/trainable': 99266, 'epoch': '0.8477'}
-
 42%|████▏     | 217/512 [24:28<32:59,  6.71s/it]
 43%|████▎     | 218/512 [24:34<32:06,  6.55s/it]
                                                 
{'loss': '0.0009051', 'grad_norm': '0.06469', 'learning_rate': '6.968e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.6', 'tokens/total': 6589392, 'tokens/trainable': 99731, 'epoch': '0.8516'}
-
 43%|████▎     | 218/512 [24:34<32:06,  6.55s/it]
 43%|████▎     | 219/512 [24:41<32:12,  6.60s/it]
                                                 
{'loss': '0.001119', 'grad_norm': '0.06683', 'learning_rate': '6.94e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.66', 'tokens/total': 6619824, 'tokens/trainable': 100164, 'epoch': '0.8555'}
-
 43%|████▎     | 219/512 [24:41<32:12,  6.60s/it]
 43%|████▎     | 220/512 [24:48<32:19,  6.64s/it]
                                                 
{'loss': '0.001774', 'grad_norm': '0.1526', 'learning_rate': '6.913e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 6650224, 'tokens/trainable': 100620, 'epoch': '0.8594'}
-
 43%|████▎     | 220/512 [24:48<32:19,  6.64s/it]
 43%|████▎     | 221/512 [24:55<32:19,  6.67s/it]
                                                 
{'loss': '0.006786', 'grad_norm': '0.8795', 'learning_rate': '6.885e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 6680384, 'tokens/trainable': 101080, 'epoch': '0.8633'}
-
 43%|████▎     | 221/512 [24:55<32:19,  6.67s/it]
 43%|████▎     | 222/512 [25:01<32:16,  6.68s/it]
                                                 
{'loss': '0.001469', 'grad_norm': '0.1953', 'learning_rate': '6.857e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 6710896, 'tokens/trainable': 101546, 'epoch': '0.8672'}
-
 43%|████▎     | 222/512 [25:01<32:16,  6.68s/it]
 44%|████▎     | 223/512 [25:08<32:08,  6.67s/it]
                                                 
{'loss': '0.0001241', 'grad_norm': '0.01113', 'learning_rate': '6.83e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 6741280, 'tokens/trainable': 101987, 'epoch': '0.8711'}
-
 44%|████▎     | 223/512 [25:08<32:08,  6.67s/it]
 44%|████▍     | 224/512 [25:15<32:05,  6.69s/it]
                                                 
{'loss': '0.004402', 'grad_norm': '0.4161', 'learning_rate': '6.802e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 6771488, 'tokens/trainable': 102435, 'epoch': '0.875'}
-
 44%|████▍     | 224/512 [25:15<32:05,  6.69s/it][2026-08-18 13:04:30,905] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
-
 44%|████▍     | 225/512 [25:23<34:52,  7.29s/it]
                                                 
{'loss': '0.0005714', 'grad_norm': '0.2329', 'learning_rate': '6.774e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '28.91', 'tokens/total': 6801744, 'tokens/trainable': 102843, 'epoch': '0.8789'}
-
 44%|████▍     | 225/512 [25:23<34:52,  7.29s/it]
 44%|████▍     | 226/512 [25:30<34:07,  7.16s/it]
                                                 
{'loss': '0.0001228', 'grad_norm': '0.01683', 'learning_rate': '6.746e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.37', 'tokens/total': 6832432, 'tokens/trainable': 103342, 'epoch': '0.8828'}
-
 44%|████▍     | 226/512 [25:30<34:07,  7.16s/it]
 44%|████▍     | 227/512 [25:37<33:20,  7.02s/it]
                                                 
{'loss': '0.004477', 'grad_norm': '0.3537', 'learning_rate': '6.718e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.71', 'tokens/train_per_sec_per_gpu': '40.68', 'tokens/total': 6862448, 'tokens/trainable': 103844, 'epoch': '0.8867'}
-
 44%|████▍     | 227/512 [25:37<33:20,  7.02s/it]
 45%|████▍     | 228/512 [25:44<32:46,  6.93s/it]
                                                 
{'loss': '0.002163', 'grad_norm': '0.4744', 'learning_rate': '6.69e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 6892832, 'tokens/trainable': 104307, 'epoch': '0.8906'}
-
 45%|████▍     | 228/512 [25:44<32:46,  6.93s/it]
 45%|████▍     | 229/512 [25:50<32:18,  6.85s/it]
                                                 
{'loss': '0.004895', 'grad_norm': '0.4888', 'learning_rate': '6.662e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.43', 'tokens/total': 6922832, 'tokens/trainable': 104772, 'epoch': '0.8945'}
-
 45%|████▍     | 229/512 [25:50<32:18,  6.85s/it]
 45%|████▍     | 230/512 [25:57<32:01,  6.82s/it]
                                                 
{'loss': '7.635e-05', 'grad_norm': '0.01735', 'learning_rate': '6.634e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.83', 'tokens/total': 6953104, 'tokens/trainable': 105240, 'epoch': '0.8984'}
-
 45%|████▍     | 230/512 [25:57<32:01,  6.82s/it]
 45%|████▌     | 231/512 [26:04<31:45,  6.78s/it]
                                                 
{'loss': '0.003428', 'grad_norm': '0.3774', 'learning_rate': '6.606e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.75', 'tokens/total': 6983440, 'tokens/trainable': 105703, 'epoch': '0.9023'}
-
 45%|████▌     | 231/512 [26:04<31:45,  6.78s/it]
 45%|████▌     | 232/512 [26:10<31:31,  6.76s/it]
                                                 
{'loss': '1.835e-05', 'grad_norm': '0.0006717', 'learning_rate': '6.578e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 7013840, 'tokens/trainable': 106179, 'epoch': '0.9062'}
-
 45%|████▌     | 232/512 [26:10<31:31,  6.76s/it]
 46%|████▌     | 233/512 [26:17<31:23,  6.75s/it]
                                                 
{'loss': '2.816e-05', 'grad_norm': '0.001895', 'learning_rate': '6.55e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '37.96', 'tokens/total': 7044240, 'tokens/trainable': 106683, 'epoch': '0.9102'}
-
 46%|████▌     | 233/512 [26:17<31:23,  6.75s/it]
 46%|████▌     | 234/512 [26:24<31:09,  6.72s/it]
                                                 
{'loss': '0.0001036', 'grad_norm': '0.0416', 'learning_rate': '6.522e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 7074224, 'tokens/trainable': 107121, 'epoch': '0.9141'}
-
 46%|████▌     | 234/512 [26:24<31:09,  6.72s/it]
 46%|████▌     | 235/512 [26:31<31:00,  6.72s/it]
                                                 
{'loss': '5.149e-05', 'grad_norm': '0.005845', 'learning_rate': '6.493e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 7104672, 'tokens/trainable': 107561, 'epoch': '0.918'}
-
 46%|████▌     | 235/512 [26:31<31:00,  6.72s/it]
 46%|████▌     | 236/512 [26:37<30:53,  6.72s/it]
                                                 
{'loss': '2.886e-05', 'grad_norm': '0.002267', 'learning_rate': '6.465e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 7134784, 'tokens/trainable': 108018, 'epoch': '0.9219'}
-
 46%|████▌     | 236/512 [26:37<30:53,  6.72s/it]
 46%|████▋     | 237/512 [26:44<30:43,  6.70s/it]
                                                 
{'loss': '0.0009025', 'grad_norm': '0.2714', 'learning_rate': '6.437e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.28', 'tokens/total': 7164960, 'tokens/trainable': 108446, 'epoch': '0.9258'}
-
 46%|████▋     | 237/512 [26:44<30:43,  6.70s/it]
 46%|████▋     | 238/512 [26:51<30:32,  6.69s/it]
                                                 
{'loss': '0.0004671', 'grad_norm': '0.1335', 'learning_rate': '6.408e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.14', 'tokens/total': 7195344, 'tokens/trainable': 108866, 'epoch': '0.9297'}
-
 46%|████▋     | 238/512 [26:51<30:32,  6.69s/it]
 47%|████▋     | 239/512 [26:57<30:31,  6.71s/it]
                                                 
{'loss': '0.0006794', 'grad_norm': '0.1007', 'learning_rate': '6.38e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.8', 'tokens/total': 7225680, 'tokens/trainable': 109316, 'epoch': '0.9336'}
-
 47%|████▋     | 239/512 [26:57<30:31,  6.71s/it]
 47%|████▋     | 240/512 [27:04<30:26,  6.72s/it]
                                                 
{'loss': '0.0005133', 'grad_norm': '0.1203', 'learning_rate': '6.351e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.26', 'tokens/total': 7256112, 'tokens/trainable': 109787, 'epoch': '0.9375'}
-
 47%|████▋     | 240/512 [27:04<30:26,  6.72s/it]
 47%|████▋     | 241/512 [27:11<30:15,  6.70s/it]
                                                 
{'loss': '3.099e-05', 'grad_norm': '0.006275', 'learning_rate': '6.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.3', 'tokens/total': 7286384, 'tokens/trainable': 110246, 'epoch': '0.9414'}
-
 47%|████▋     | 241/512 [27:11<30:15,  6.70s/it]
 47%|████▋     | 242/512 [27:17<30:11,  6.71s/it]
                                                 
{'loss': '0.000395', 'grad_norm': '0.05609', 'learning_rate': '6.294e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.14', 'tokens/total': 7316688, 'tokens/trainable': 110667, 'epoch': '0.9453'}
-
 47%|████▋     | 242/512 [27:17<30:11,  6.71s/it]
 47%|████▋     | 243/512 [27:24<30:06,  6.71s/it]
                                                 
{'loss': '0.0002062', 'grad_norm': '0.05409', 'learning_rate': '6.266e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.09', 'tokens/total': 7346992, 'tokens/trainable': 111154, 'epoch': '0.9492'}
-
 47%|████▋     | 243/512 [27:24<30:06,  6.71s/it]
 48%|████▊     | 244/512 [27:31<30:03,  6.73s/it]
                                                 
{'loss': '0.0005747', 'grad_norm': '0.05567', 'learning_rate': '6.237e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 7377632, 'tokens/trainable': 111611, 'epoch': '0.9531'}
-
 48%|████▊     | 244/512 [27:31<30:03,  6.73s/it]
 48%|████▊     | 245/512 [27:38<29:52,  6.72s/it]
                                                 
{'loss': '0.0008788', 'grad_norm': '0.1613', 'learning_rate': '6.208e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 7407872, 'tokens/trainable': 112102, 'epoch': '0.957'}
-
 48%|████▊     | 245/512 [27:38<29:52,  6.72s/it]
 48%|████▊     | 246/512 [27:44<29:44,  6.71s/it]
                                                 
{'loss': '4.069e-05', 'grad_norm': '0.005659', 'learning_rate': '6.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.16', 'tokens/total': 7438448, 'tokens/trainable': 112569, 'epoch': '0.9609'}
-
 48%|████▊     | 246/512 [27:44<29:44,  6.71s/it]
 48%|████▊     | 247/512 [27:51<29:00,  6.57s/it]
                                                 
{'loss': '0.0005083', 'grad_norm': '0.1775', 'learning_rate': '6.151e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '37.68', 'tokens/total': 7466768, 'tokens/trainable': 113023, 'epoch': '0.9648'}
-
 48%|████▊     | 247/512 [27:51<29:00,  6.57s/it]
 48%|████▊     | 248/512 [27:57<29:09,  6.63s/it]
                                                 
{'loss': '3.471e-05', 'grad_norm': '0.002822', 'learning_rate': '6.122e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.17', 'tokens/total': 7497360, 'tokens/trainable': 113464, 'epoch': '0.9688'}
-
 48%|████▊     | 248/512 [27:57<29:09,  6.63s/it]
 49%|████▊     | 249/512 [28:04<29:07,  6.64s/it]
                                                 
{'loss': '8.433e-05', 'grad_norm': '0.03676', 'learning_rate': '6.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 7527504, 'tokens/trainable': 113908, 'epoch': '0.9727'}
-
 49%|████▊     | 249/512 [28:04<29:07,  6.64s/it]
 49%|████▉     | 250/512 [28:10<28:26,  6.51s/it]
                                                 
{'loss': '0.0006511', 'grad_norm': '0.1124', 'learning_rate': '6.065e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 7555808, 'tokens/trainable': 114348, 'epoch': '0.9766'}
-
 49%|████▉     | 250/512 [28:10<28:26,  6.51s/it]
 49%|████▉     | 251/512 [28:17<28:36,  6.58s/it]
                                                 
{'loss': '0.0002415', 'grad_norm': '0.04923', 'learning_rate': '6.036e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.15', 'tokens/total': 7586320, 'tokens/trainable': 114830, 'epoch': '0.9805'}
-
 49%|████▉     | 251/512 [28:17<28:36,  6.58s/it]
 49%|████▉     | 252/512 [28:24<28:41,  6.62s/it]
                                                 
{'loss': '0.005054', 'grad_norm': '0.4834', 'learning_rate': '6.007e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.1', 'tokens/total': 7616576, 'tokens/trainable': 115278, 'epoch': '0.9844'}
-
 49%|████▉     | 252/512 [28:24<28:41,  6.62s/it]
 49%|████▉     | 253/512 [28:30<28:46,  6.67s/it]
                                                 
{'loss': '8.867e-05', 'grad_norm': '0.01336', 'learning_rate': '5.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 7647152, 'tokens/trainable': 115699, 'epoch': '0.9883'}
-
 49%|████▉     | 253/512 [28:30<28:46,  6.67s/it]
 50%|████▉     | 254/512 [28:37<28:42,  6.68s/it]
                                                 
{'loss': '0.0006354', 'grad_norm': '0.09679', 'learning_rate': '5.949e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '30.73', 'tokens/total': 7677584, 'tokens/trainable': 116133, 'epoch': '0.9922'}
-
 50%|████▉     | 254/512 [28:37<28:42,  6.68s/it]
 50%|████▉     | 255/512 [28:44<28:37,  6.68s/it]
                                                 
{'loss': '0.001129', 'grad_norm': '0.514', 'learning_rate': '5.92e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '36.73', 'tokens/total': 7708016, 'tokens/trainable': 116591, 'epoch': '0.9961'}
-
 50%|████▉     | 255/512 [28:44<28:37,  6.68s/it]
 50%|█████     | 256/512 [28:51<28:46,  6.74s/it]
                                                 
{'loss': '7.111e-06', 'grad_norm': '0.000276', 'learning_rate': '5.891e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.04', 'tokens/total': 7738608, 'tokens/trainable': 117039, 'epoch': '1'}
-
 50%|█████     | 256/512 [28:51<28:46,  6.74s/it][2026-08-18 13:08:06,959] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
-
 50%|█████     | 257/512 [29:00<31:40,  7.45s/it]
                                                 
{'loss': '2.959e-05', 'grad_norm': '0.00545', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.24', 'tokens/total': 7768944, 'tokens/trainable': 117495, 'epoch': '1.004'}
-
 50%|█████     | 257/512 [29:00<31:40,  7.45s/it]
 50%|█████     | 258/512 [29:07<30:36,  7.23s/it]
                                                 
{'loss': '2.851e-05', 'grad_norm': '0.003981', 'learning_rate': '5.834e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.44', 'tokens/total': 7799232, 'tokens/trainable': 117925, 'epoch': '1.008'}
-
 50%|█████     | 258/512 [29:07<30:36,  7.23s/it]
 51%|█████     | 259/512 [29:13<29:50,  7.08s/it]
                                                 
{'loss': '1.161e-05', 'grad_norm': '0.001549', 'learning_rate': '5.805e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.58', 'tokens/total': 7829568, 'tokens/trainable': 118373, 'epoch': '1.012'}
-
 51%|█████     | 259/512 [29:13<29:50,  7.08s/it]
 51%|█████     | 260/512 [29:20<29:15,  6.97s/it]
                                                 
{'loss': '9.4e-06', 'grad_norm': '0.0003936', 'learning_rate': '5.776e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 7860000, 'tokens/trainable': 118816, 'epoch': '1.016'}
-
 51%|█████     | 260/512 [29:20<29:15,  6.97s/it]
 51%|█████     | 261/512 [29:27<28:47,  6.88s/it]
                                                 
{'loss': '0.000134', 'grad_norm': '0.02447', 'learning_rate': '5.747e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.8', 'tokens/total': 7890464, 'tokens/trainable': 119263, 'epoch': '1.02'}
-
 51%|█████     | 261/512 [29:27<28:47,  6.88s/it]
 51%|█████     | 262/512 [29:33<28:33,  6.85s/it]
                                                 
{'loss': '0.002393', 'grad_norm': '0.3619', 'learning_rate': '5.718e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.66', 'tokens/total': 7920944, 'tokens/trainable': 119772, 'epoch': '1.023'}
-
 51%|█████     | 262/512 [29:33<28:33,  6.85s/it]
 51%|█████▏    | 263/512 [29:40<28:18,  6.82s/it]
                                                 
{'loss': '0.000525', 'grad_norm': '0.1766', 'learning_rate': '5.689e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 7951296, 'tokens/trainable': 120269, 'epoch': '1.027'}
-
 51%|█████▏    | 263/512 [29:40<28:18,  6.82s/it]
 52%|█████▏    | 264/512 [29:47<28:07,  6.81s/it]
                                                 
{'loss': '1.115e-05', 'grad_norm': '0.0007879', 'learning_rate': '5.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.96', 'tokens/total': 7981856, 'tokens/trainable': 120746, 'epoch': '1.031'}
-
 52%|█████▏    | 264/512 [29:47<28:07,  6.81s/it]
 52%|█████▏    | 265/512 [29:54<27:51,  6.77s/it]
                                                 
{'loss': '4.71e-05', 'grad_norm': '0.008799', 'learning_rate': '5.631e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.22', 'tokens/total': 8012192, 'tokens/trainable': 121204, 'epoch': '1.035'}
-
 52%|█████▏    | 265/512 [29:54<27:51,  6.77s/it]
 52%|█████▏    | 266/512 [30:00<27:43,  6.76s/it]
                                                 
{'loss': '8.791e-06', 'grad_norm': '0.0005215', 'learning_rate': '5.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.23', 'tokens/total': 8042736, 'tokens/trainable': 121670, 'epoch': '1.039'}
-
 52%|█████▏    | 266/512 [30:00<27:43,  6.76s/it]
 52%|█████▏    | 267/512 [30:07<27:35,  6.76s/it]
                                                 
{'loss': '0.0008057', 'grad_norm': '0.1482', 'learning_rate': '5.573e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '40.04', 'tokens/total': 8073024, 'tokens/trainable': 122183, 'epoch': '1.043'}
-
 52%|█████▏    | 267/512 [30:07<27:35,  6.76s/it]
 52%|█████▏    | 268/512 [30:14<27:24,  6.74s/it]
                                                 
{'loss': '0.000323', 'grad_norm': '0.08615', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 8103280, 'tokens/trainable': 122645, 'epoch': '1.047'}
-
 52%|█████▏    | 268/512 [30:14<27:24,  6.74s/it]
 53%|█████▎    | 269/512 [30:21<27:15,  6.73s/it]
                                                 
{'loss': '1.134e-05', 'grad_norm': '0.0005172', 'learning_rate': '5.515e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 8133440, 'tokens/trainable': 123105, 'epoch': '1.051'}
-
 53%|█████▎    | 269/512 [30:21<27:15,  6.73s/it]
 53%|█████▎    | 270/512 [30:27<27:09,  6.73s/it]
                                                 
{'loss': '1.195e-05', 'grad_norm': '0.001041', 'learning_rate': '5.485e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 8163920, 'tokens/trainable': 123595, 'epoch': '1.055'}
-
 53%|█████▎    | 270/512 [30:27<27:09,  6.73s/it]
 53%|█████▎    | 271/512 [30:34<27:00,  6.73s/it]
                                                 
{'loss': '1.726e-05', 'grad_norm': '0.005696', 'learning_rate': '5.456e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.59', 'tokens/total': 8194032, 'tokens/trainable': 124077, 'epoch': '1.059'}
-
 53%|█████▎    | 271/512 [30:34<27:00,  6.73s/it]
 53%|█████▎    | 272/512 [30:41<26:56,  6.73s/it]
                                                 
{'loss': '1.061e-05', 'grad_norm': '0.0005264', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.65', 'tokens/total': 8224320, 'tokens/trainable': 124547, 'epoch': '1.062'}
-
 53%|█████▎    | 272/512 [30:41<26:56,  6.73s/it]
 53%|█████▎    | 273/512 [30:47<26:46,  6.72s/it]
                                                 
{'loss': '0.0003628', 'grad_norm': '0.08169', 'learning_rate': '5.398e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 8254368, 'tokens/trainable': 124977, 'epoch': '1.066'}
-
 53%|█████▎    | 273/512 [30:47<26:46,  6.72s/it]
 54%|█████▎    | 274/512 [30:54<26:37,  6.71s/it]
                                                 
{'loss': '1.123e-05', 'grad_norm': '0.0005203', 'learning_rate': '5.369e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.51', 'tokens/total': 8284944, 'tokens/trainable': 125438, 'epoch': '1.07'}
-
 54%|█████▎    | 274/512 [30:54<26:37,  6.71s/it]
 54%|█████▎    | 275/512 [31:01<26:32,  6.72s/it]
                                                 
{'loss': '1.476e-05', 'grad_norm': '0.00311', 'learning_rate': '5.34e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.47', 'tokens/total': 8315584, 'tokens/trainable': 125889, 'epoch': '1.074'}
-
 54%|█████▎    | 275/512 [31:01<26:32,  6.72s/it]
 54%|█████▍    | 276/512 [31:08<26:26,  6.72s/it]
                                                 
{'loss': '9.164e-05', 'grad_norm': '0.01906', 'learning_rate': '5.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 8345952, 'tokens/trainable': 126363, 'epoch': '1.078'}
-
 54%|█████▍    | 276/512 [31:08<26:26,  6.72s/it]
 54%|█████▍    | 277/512 [31:14<26:23,  6.74s/it]
                                                 
{'loss': '1.797e-05', 'grad_norm': '0.002002', 'learning_rate': '5.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 8376416, 'tokens/trainable': 126828, 'epoch': '1.082'}
-
 54%|█████▍    | 277/512 [31:14<26:23,  6.74s/it]
 54%|█████▍    | 278/512 [31:21<26:14,  6.73s/it]
                                                 
{'loss': '1.279e-05', 'grad_norm': '0.001124', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.16', 'tokens/total': 8406640, 'tokens/trainable': 127331, 'epoch': '1.086'}
-
 54%|█████▍    | 278/512 [31:21<26:14,  6.73s/it]
 54%|█████▍    | 279/512 [31:28<26:05,  6.72s/it]
                                                 
{'loss': '0.0005633', 'grad_norm': '0.07841', 'learning_rate': '5.224e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.34', 'tokens/total': 8436784, 'tokens/trainable': 127798, 'epoch': '1.09'}
-
 54%|█████▍    | 279/512 [31:28<26:05,  6.72s/it]
 55%|█████▍    | 280/512 [31:35<25:59,  6.72s/it]
                                                 
{'loss': '4.469e-05', 'grad_norm': '0.006657', 'learning_rate': '5.195e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 8467152, 'tokens/trainable': 128247, 'epoch': '1.094'}
-
 55%|█████▍    | 280/512 [31:35<25:59,  6.72s/it]
 55%|█████▍    | 281/512 [31:41<25:54,  6.73s/it]
                                                 
{'loss': '9.423e-06', 'grad_norm': '0.0006085', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 8497424, 'tokens/trainable': 128729, 'epoch': '1.098'}
-
 55%|█████▍    | 281/512 [31:41<25:54,  6.73s/it]
 55%|█████▌    | 282/512 [31:48<25:48,  6.73s/it]
                                                 
{'loss': '0.002226', 'grad_norm': '0.4024', 'learning_rate': '5.138e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 8527616, 'tokens/trainable': 129203, 'epoch': '1.102'}
-
 55%|█████▌    | 282/512 [31:48<25:48,  6.73s/it]
 55%|█████▌    | 283/512 [31:55<25:42,  6.74s/it]
                                                 
{'loss': '9.101e-06', 'grad_norm': '0.0008598', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 8557984, 'tokens/trainable': 129653, 'epoch': '1.105'}
-
 55%|█████▌    | 283/512 [31:55<25:42,  6.74s/it]
 55%|█████▌    | 284/512 [32:02<25:37,  6.74s/it]
                                                 
{'loss': '0.0004534', 'grad_norm': '0.1385', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.56', 'tokens/total': 8588384, 'tokens/trainable': 130100, 'epoch': '1.109'}
-
 55%|█████▌    | 284/512 [32:02<25:37,  6.74s/it]
 56%|█████▌    | 285/512 [32:08<25:28,  6.73s/it]
                                                 
{'loss': '0.0002941', 'grad_norm': '0.08453', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.44', 'tokens/total': 8618768, 'tokens/trainable': 130543, 'epoch': '1.113'}
-
 56%|█████▌    | 285/512 [32:08<25:28,  6.73s/it]
 56%|█████▌    | 286/512 [32:15<25:21,  6.73s/it]
                                                 
{'loss': '0.0001019', 'grad_norm': '0.01867', 'learning_rate': '5.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 8648976, 'tokens/trainable': 131000, 'epoch': '1.117'}
-
 56%|█████▌    | 286/512 [32:15<25:21,  6.73s/it]
 56%|█████▌    | 287/512 [32:22<25:13,  6.73s/it]
                                                 
{'loss': '1.198e-05', 'grad_norm': '0.00159', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 8679152, 'tokens/trainable': 131449, 'epoch': '1.121'}
-
 56%|█████▌    | 287/512 [32:22<25:13,  6.73s/it]
 56%|█████▋    | 288/512 [32:28<25:03,  6.71s/it]
                                                 
{'loss': '7.07e-06', 'grad_norm': '0.0004093', 'learning_rate': '4.964e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '35.94', 'tokens/total': 8709504, 'tokens/trainable': 131923, 'epoch': '1.125'}
-
 56%|█████▋    | 288/512 [32:28<25:03,  6.71s/it][2026-08-18 13:11:44,549] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
-
 56%|█████▋    | 289/512 [32:37<27:25,  7.38s/it]
                                                 
{'loss': '0.0006339', 'grad_norm': '0.1226', 'learning_rate': '4.935e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '33.11', 'tokens/total': 8739680, 'tokens/trainable': 132397, 'epoch': '1.129'}
-
 56%|█████▋    | 289/512 [32:37<27:25,  7.38s/it]
 57%|█████▋    | 290/512 [32:44<26:34,  7.18s/it]
                                                 
{'loss': '5.397e-06', 'grad_norm': '0.0002119', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 8770112, 'tokens/trainable': 132846, 'epoch': '1.133'}
-
 57%|█████▋    | 290/512 [32:44<26:34,  7.18s/it]
 57%|█████▋    | 291/512 [32:51<25:57,  7.05s/it]
                                                 
{'loss': '0.00135', 'grad_norm': '0.2777', 'learning_rate': '4.878e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '35.65', 'tokens/total': 8800368, 'tokens/trainable': 133347, 'epoch': '1.137'}
-
 57%|█████▋    | 291/512 [32:51<25:57,  7.05s/it]
 57%|█████▋    | 292/512 [32:57<25:27,  6.94s/it]
                                                 
{'loss': '8.635e-06', 'grad_norm': '0.001235', 'learning_rate': '4.849e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 8830688, 'tokens/trainable': 133779, 'epoch': '1.141'}
-
 57%|█████▋    | 292/512 [32:57<25:27,  6.94s/it]
 57%|█████▋    | 293/512 [33:04<25:08,  6.89s/it]
                                                 
{'loss': '4.307e-06', 'grad_norm': '0.0002618', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 8861008, 'tokens/trainable': 134251, 'epoch': '1.145'}
-
 57%|█████▋    | 293/512 [33:04<25:08,  6.89s/it]
 57%|█████▋    | 294/512 [33:11<24:56,  6.86s/it]
                                                 
{'loss': '4.467e-06', 'grad_norm': '0.0004677', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 8891760, 'tokens/trainable': 134683, 'epoch': '1.148'}
-
 57%|█████▋    | 294/512 [33:11<24:56,  6.86s/it]
 58%|█████▊    | 295/512 [33:18<24:40,  6.82s/it]
                                                 
{'loss': '4.914e-06', 'grad_norm': '0.0002164', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '35.51', 'tokens/total': 8922192, 'tokens/trainable': 135165, 'epoch': '1.152'}
-
 58%|█████▊    | 295/512 [33:18<24:40,  6.82s/it]
 58%|█████▊    | 296/512 [33:24<24:23,  6.78s/it]
                                                 
{'loss': '4.242e-06', 'grad_norm': '0.0002447', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 8952496, 'tokens/trainable': 135599, 'epoch': '1.156'}
-
 58%|█████▊    | 296/512 [33:24<24:23,  6.78s/it]
 58%|█████▊    | 297/512 [33:31<24:11,  6.75s/it]
                                                 
{'loss': '4.674e-06', 'grad_norm': '0.0002878', 'learning_rate': '4.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.56', 'tokens/total': 8982768, 'tokens/trainable': 136038, 'epoch': '1.16'}
-
 58%|█████▊    | 297/512 [33:31<24:11,  6.75s/it]
 58%|█████▊    | 298/512 [33:38<24:03,  6.74s/it]
                                                 
{'loss': '8.829e-06', 'grad_norm': '0.001556', 'learning_rate': '4.677e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 9013248, 'tokens/trainable': 136495, 'epoch': '1.164'}
-
 58%|█████▊    | 298/512 [33:38<24:03,  6.74s/it]
 58%|█████▊    | 299/512 [33:45<23:56,  6.75s/it]
                                                 
{'loss': '5.587e-06', 'grad_norm': '0.0002536', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 9043632, 'tokens/trainable': 136953, 'epoch': '1.168'}
-
 58%|█████▊    | 299/512 [33:45<23:56,  6.75s/it]
 59%|█████▊    | 300/512 [33:51<23:45,  6.73s/it]
                                                 
{'loss': '5.917e-06', 'grad_norm': '0.000319', 'learning_rate': '4.62e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 9073936, 'tokens/trainable': 137412, 'epoch': '1.172'}
-
 59%|█████▊    | 300/512 [33:51<23:45,  6.73s/it]
 59%|█████▉    | 301/512 [33:58<23:38,  6.72s/it]
                                                 
{'loss': '9.368e-06', 'grad_norm': '0.0005231', 'learning_rate': '4.592e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.62', 'tokens/total': 9104416, 'tokens/trainable': 137907, 'epoch': '1.176'}
-
 59%|█████▉    | 301/512 [33:58<23:38,  6.72s/it]
 59%|█████▉    | 302/512 [34:05<23:33,  6.73s/it]
                                                 
{'loss': '9.613e-06', 'grad_norm': '0.001005', 'learning_rate': '4.563e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 9134688, 'tokens/trainable': 138383, 'epoch': '1.18'}
-
 59%|█████▉    | 302/512 [34:05<23:33,  6.73s/it]
 59%|█████▉    | 303/512 [34:11<23:29,  6.74s/it]
                                                 
{'loss': '2.478e-05', 'grad_norm': '0.005482', 'learning_rate': '4.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '36.35', 'tokens/total': 9165088, 'tokens/trainable': 138842, 'epoch': '1.184'}
-
 59%|█████▉    | 303/512 [34:11<23:29,  6.74s/it]
 59%|█████▉    | 304/512 [34:18<23:17,  6.72s/it]
                                                 
{'loss': '5.224e-06', 'grad_norm': '0.0005683', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '30.61', 'tokens/total': 9195584, 'tokens/trainable': 139269, 'epoch': '1.188'}
-
 59%|█████▉    | 304/512 [34:18<23:17,  6.72s/it]
 60%|█████▉    | 305/512 [34:25<23:11,  6.72s/it]
                                                 
{'loss': '2.194e-05', 'grad_norm': '0.003497', 'learning_rate': '4.478e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.36', 'tokens/total': 9225760, 'tokens/trainable': 139736, 'epoch': '1.191'}
-
 60%|█████▉    | 305/512 [34:25<23:11,  6.72s/it]
 60%|█████▉    | 306/512 [34:32<23:04,  6.72s/it]
                                                 
{'loss': '4.178e-06', 'grad_norm': '0.000202', 'learning_rate': '4.45e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.91', 'tokens/total': 9256352, 'tokens/trainable': 140226, 'epoch': '1.195'}
-
 60%|█████▉    | 306/512 [34:32<23:04,  6.72s/it]
 60%|█████▉    | 307/512 [34:38<22:20,  6.54s/it]
                                                 
{'loss': '5.606e-06', 'grad_norm': '0.0004581', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.19', 'memory/max_allocated (GiB)': '33.19', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 9284256, 'tokens/trainable': 140664, 'epoch': '1.199'}
-
 60%|█████▉    | 307/512 [34:38<22:20,  6.54s/it]
 60%|██████    | 308/512 [34:44<22:29,  6.62s/it]
                                                 
{'loss': '7.164e-06', 'grad_norm': '0.001661', 'learning_rate': '4.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '29.85', 'tokens/total': 9314768, 'tokens/trainable': 141115, 'epoch': '1.203'}
-
 60%|██████    | 308/512 [34:45<22:29,  6.62s/it]
 60%|██████    | 309/512 [34:51<22:29,  6.65s/it]
                                                 
{'loss': '3.827e-06', 'grad_norm': '0.0001901', 'learning_rate': '4.366e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '28.88', 'tokens/total': 9344976, 'tokens/trainable': 141530, 'epoch': '1.207'}
-
 60%|██████    | 309/512 [34:51<22:29,  6.65s/it]
 61%|██████    | 310/512 [34:58<22:24,  6.66s/it]
                                                 
{'loss': '2.814e-05', 'grad_norm': '0.003887', 'learning_rate': '4.338e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 9375280, 'tokens/trainable': 141964, 'epoch': '1.211'}
-
 61%|██████    | 310/512 [34:58<22:24,  6.66s/it]
 61%|██████    | 311/512 [35:05<22:19,  6.67s/it]
                                                 
{'loss': '0.0001199', 'grad_norm': '0.02483', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 9405536, 'tokens/trainable': 142428, 'epoch': '1.215'}
-
 61%|██████    | 311/512 [35:05<22:19,  6.67s/it]
 61%|██████    | 312/512 [35:11<22:12,  6.66s/it]
                                                 
{'loss': '9.509e-06', 'grad_norm': '0.001796', 'learning_rate': '4.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '38.07', 'tokens/total': 9435696, 'tokens/trainable': 142894, 'epoch': '1.219'}
-
 61%|██████    | 312/512 [35:11<22:12,  6.66s/it]
 61%|██████    | 313/512 [35:18<22:09,  6.68s/it]
                                                 
{'loss': '8.858e-06', 'grad_norm': '0.0007182', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.3', 'tokens/total': 9466080, 'tokens/trainable': 143368, 'epoch': '1.223'}
-
 61%|██████    | 313/512 [35:18<22:09,  6.68s/it]
 61%|██████▏   | 314/512 [35:25<22:04,  6.69s/it]
                                                 
{'loss': '0.0151', 'grad_norm': '0.7545', 'learning_rate': '4.226e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 9496400, 'tokens/trainable': 143824, 'epoch': '1.227'}
-
 61%|██████▏   | 314/512 [35:25<22:04,  6.69s/it]
 62%|██████▏   | 315/512 [35:31<22:05,  6.73s/it]
                                                 
{'loss': '0.0007513', 'grad_norm': '0.281', 'learning_rate': '4.198e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 9526880, 'tokens/trainable': 144306, 'epoch': '1.23'}
-
 62%|██████▏   | 315/512 [35:31<22:05,  6.73s/it]
 62%|██████▏   | 316/512 [35:38<21:58,  6.73s/it]
                                                 
{'loss': '0.0005616', 'grad_norm': '0.7471', 'learning_rate': '4.17e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 9557280, 'tokens/trainable': 144763, 'epoch': '1.234'}
-
 62%|██████▏   | 316/512 [35:38<21:58,  6.73s/it]
 62%|██████▏   | 317/512 [35:45<21:52,  6.73s/it]
                                                 
{'loss': '5.734e-06', 'grad_norm': '0.0004947', 'learning_rate': '4.143e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.4', 'tokens/total': 9587712, 'tokens/trainable': 145255, 'epoch': '1.238'}
-
 62%|██████▏   | 317/512 [35:45<21:52,  6.73s/it]
 62%|██████▏   | 318/512 [35:52<21:45,  6.73s/it]
                                                 
{'loss': '2.625e-05', 'grad_norm': '0.003202', 'learning_rate': '4.115e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 9618064, 'tokens/trainable': 145714, 'epoch': '1.242'}
-
 62%|██████▏   | 318/512 [35:52<21:45,  6.73s/it]
 62%|██████▏   | 319/512 [35:58<21:39,  6.73s/it]
                                                 
{'loss': '0.0263', 'grad_norm': '1.172', 'learning_rate': '4.087e-05', 'ppl': '1.027', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.96', 'tokens/total': 9648336, 'tokens/trainable': 146200, 'epoch': '1.246'}
-
 62%|██████▏   | 319/512 [35:58<21:39,  6.73s/it]
 62%|██████▎   | 320/512 [36:05<21:31,  6.73s/it]
                                                 
{'loss': '8.492e-06', 'grad_norm': '0.00046', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 9678688, 'tokens/trainable': 146668, 'epoch': '1.25'}
-
 62%|██████▎   | 320/512 [36:05<21:31,  6.73s/it][2026-08-18 13:15:21,351] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
-
 63%|██████▎   | 321/512 [36:14<23:15,  7.31s/it]
                                                 
{'loss': '6.043e-05', 'grad_norm': '0.006735', 'learning_rate': '4.032e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.24', 'tokens/total': 9708848, 'tokens/trainable': 147114, 'epoch': '1.254'}
-
 63%|██████▎   | 321/512 [36:14<23:15,  7.31s/it]
 63%|██████▎   | 322/512 [36:20<22:33,  7.12s/it]
                                                 
{'loss': '0.0002499', 'grad_norm': '0.05488', 'learning_rate': '4.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 9739360, 'tokens/trainable': 147564, 'epoch': '1.258'}
-
 63%|██████▎   | 322/512 [36:20<22:33,  7.12s/it]
 63%|██████▎   | 323/512 [36:27<22:03,  7.00s/it]
                                                 
{'loss': '1.1e-05', 'grad_norm': '0.001182', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 9769776, 'tokens/trainable': 148013, 'epoch': '1.262'}
-
 63%|██████▎   | 323/512 [36:27<22:03,  7.00s/it]
 63%|██████▎   | 324/512 [36:34<21:44,  6.94s/it]
                                                 
{'loss': '0.0002116', 'grad_norm': '0.01802', 'learning_rate': '3.95e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.99', 'tokens/total': 9800096, 'tokens/trainable': 148470, 'epoch': '1.266'}
-
 63%|██████▎   | 324/512 [36:34<21:44,  6.94s/it]
 63%|██████▎   | 325/512 [36:41<21:22,  6.86s/it]
                                                 
{'loss': '0.01622', 'grad_norm': '1.591', 'learning_rate': '3.923e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.44', 'tokens/total': 9830304, 'tokens/trainable': 148926, 'epoch': '1.27'}
-
 63%|██████▎   | 325/512 [36:41<21:22,  6.86s/it]
 64%|██████▎   | 326/512 [36:47<21:10,  6.83s/it]
                                                 
{'loss': '3.801e-05', 'grad_norm': '0.004154', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.72', 'tokens/total': 9860608, 'tokens/trainable': 149420, 'epoch': '1.273'}
-
 64%|██████▎   | 326/512 [36:47<21:10,  6.83s/it]
 64%|██████▍   | 327/512 [36:54<20:57,  6.80s/it]
                                                 
{'loss': '5.527e-05', 'grad_norm': '0.002978', 'learning_rate': '3.869e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '39.32', 'tokens/total': 9890560, 'tokens/trainable': 149873, 'epoch': '1.277'}
-
 64%|██████▍   | 327/512 [36:54<20:57,  6.80s/it]
 64%|██████▍   | 328/512 [37:01<20:47,  6.78s/it]
                                                 
{'loss': '0.0003906', 'grad_norm': '0.02816', 'learning_rate': '3.842e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 9920880, 'tokens/trainable': 150347, 'epoch': '1.281'}
-
 64%|██████▍   | 328/512 [37:01<20:47,  6.78s/it]
 64%|██████▍   | 329/512 [37:08<20:37,  6.76s/it]
                                                 
{'loss': '0.0005861', 'grad_norm': '0.04141', 'learning_rate': '3.815e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.72', 'tokens/total': 9951280, 'tokens/trainable': 150822, 'epoch': '1.285'}
-
 64%|██████▍   | 329/512 [37:08<20:37,  6.76s/it]
 64%|██████▍   | 330/512 [37:14<20:29,  6.76s/it]
                                                 
{'loss': '9.081e-05', 'grad_norm': '0.004759', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 9981600, 'tokens/trainable': 151266, 'epoch': '1.289'}
-
 64%|██████▍   | 330/512 [37:14<20:29,  6.76s/it]
 65%|██████▍   | 331/512 [37:21<20:20,  6.74s/it]
                                                 
{'loss': '0.0005758', 'grad_norm': '0.04044', 'learning_rate': '3.761e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.11', 'tokens/total': 10012016, 'tokens/trainable': 151734, 'epoch': '1.293'}
-
 65%|██████▍   | 331/512 [37:21<20:20,  6.74s/it]
 65%|██████▍   | 332/512 [37:28<20:10,  6.72s/it]
                                                 
{'loss': '0.0004353', 'grad_norm': '0.06457', 'learning_rate': '3.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.99', 'tokens/total': 10042448, 'tokens/trainable': 152184, 'epoch': '1.297'}
-
 65%|██████▍   | 332/512 [37:28<20:10,  6.72s/it]
 65%|██████▌   | 333/512 [37:34<20:04,  6.73s/it]
                                                 
{'loss': '0.0005283', 'grad_norm': '0.05121', 'learning_rate': '3.708e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 10073024, 'tokens/trainable': 152623, 'epoch': '1.301'}
-
 65%|██████▌   | 333/512 [37:34<20:04,  6.73s/it]
 65%|██████▌   | 334/512 [37:41<19:53,  6.71s/it]
                                                 
{'loss': '0.0001399', 'grad_norm': '0.006798', 'learning_rate': '3.681e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.34', 'tokens/total': 10103504, 'tokens/trainable': 153040, 'epoch': '1.305'}
-
 65%|██████▌   | 334/512 [37:41<19:53,  6.71s/it]
 65%|██████▌   | 335/512 [37:47<19:21,  6.56s/it]
                                                 
{'loss': '0.003549', 'grad_norm': '0.2877', 'learning_rate': '3.655e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.7', 'tokens/total': 10131632, 'tokens/trainable': 153463, 'epoch': '1.309'}
-
 65%|██████▌   | 335/512 [37:47<19:21,  6.56s/it]
 66%|██████▌   | 336/512 [37:54<19:21,  6.60s/it]
                                                 
{'loss': '0.0007546', 'grad_norm': '0.04293', 'learning_rate': '3.628e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.22', 'tokens/total': 10161824, 'tokens/trainable': 153908, 'epoch': '1.312'}
-
 66%|██████▌   | 336/512 [37:54<19:21,  6.60s/it]
 66%|██████▌   | 337/512 [38:01<19:25,  6.66s/it]
                                                 
{'loss': '0.001123', 'grad_norm': '0.1319', 'learning_rate': '3.602e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.63', 'tokens/total': 10192448, 'tokens/trainable': 154387, 'epoch': '1.316'}
-
 66%|██████▌   | 337/512 [38:01<19:25,  6.66s/it]
 66%|██████▌   | 338/512 [38:08<19:23,  6.69s/it]
                                                 
{'loss': '0.0004925', 'grad_norm': '0.06429', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 10222704, 'tokens/trainable': 154836, 'epoch': '1.32'}
-
 66%|██████▌   | 338/512 [38:08<19:23,  6.69s/it]
 66%|██████▌   | 339/512 [38:14<19:22,  6.72s/it]
                                                 
{'loss': '8.633e-05', 'grad_norm': '0.006817', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.73', 'tokens/total': 10253168, 'tokens/trainable': 155261, 'epoch': '1.324'}
-
 66%|██████▌   | 339/512 [38:14<19:22,  6.72s/it]
 66%|██████▋   | 340/512 [38:21<19:12,  6.70s/it]
                                                 
{'loss': '0.0003515', 'grad_norm': '0.04607', 'learning_rate': '3.523e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.14', 'tokens/total': 10283552, 'tokens/trainable': 155674, 'epoch': '1.328'}
-
 66%|██████▋   | 340/512 [38:21<19:12,  6.70s/it]
 67%|██████▋   | 341/512 [38:28<19:08,  6.72s/it]
                                                 
{'loss': '0.0007426', 'grad_norm': '0.2491', 'learning_rate': '3.497e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.67', 'tokens/total': 10313872, 'tokens/trainable': 156164, 'epoch': '1.332'}
-
 67%|██████▋   | 341/512 [38:28<19:08,  6.72s/it]
 67%|██████▋   | 342/512 [38:35<19:02,  6.72s/it]
                                                 
{'loss': '0.0001773', 'grad_norm': '0.01124', 'learning_rate': '3.471e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.47', 'tokens/total': 10344256, 'tokens/trainable': 156611, 'epoch': '1.336'}
-
 67%|██████▋   | 342/512 [38:35<19:02,  6.72s/it]
 67%|██████▋   | 343/512 [38:41<18:54,  6.71s/it]
                                                 
{'loss': '0.0001897', 'grad_norm': '0.009496', 'learning_rate': '3.445e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.58', 'tokens/total': 10374544, 'tokens/trainable': 157073, 'epoch': '1.34'}
-
 67%|██████▋   | 343/512 [38:41<18:54,  6.71s/it]
 67%|██████▋   | 344/512 [38:48<18:49,  6.72s/it]
                                                 
{'loss': '0.0007583', 'grad_norm': '0.1076', 'learning_rate': '3.42e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 10405040, 'tokens/trainable': 157523, 'epoch': '1.344'}
-
 67%|██████▋   | 344/512 [38:48<18:49,  6.72s/it]
 67%|██████▋   | 345/512 [38:55<18:41,  6.72s/it]
                                                 
{'loss': '0.0015', 'grad_norm': '0.1594', 'learning_rate': '3.394e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.07', 'tokens/total': 10435424, 'tokens/trainable': 157964, 'epoch': '1.348'}
-
 67%|██████▋   | 345/512 [38:55<18:41,  6.72s/it]
 68%|██████▊   | 346/512 [39:01<18:33,  6.71s/it]
                                                 
{'loss': '3.834e-05', 'grad_norm': '0.001656', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '30.35', 'tokens/total': 10465760, 'tokens/trainable': 158390, 'epoch': '1.352'}
-
 68%|██████▊   | 346/512 [39:01<18:33,  6.71s/it]
 68%|██████▊   | 347/512 [39:08<18:24,  6.69s/it]
                                                 
{'loss': '0.0003919', 'grad_norm': '0.02406', 'learning_rate': '3.343e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.47', 'tokens/total': 10495904, 'tokens/trainable': 158848, 'epoch': '1.355'}
-
 68%|██████▊   | 347/512 [39:08<18:24,  6.69s/it]
 68%|██████▊   | 348/512 [39:15<18:21,  6.72s/it]
                                                 
{'loss': '0.0006162', 'grad_norm': '0.04579', 'learning_rate': '3.317e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.63', 'tokens/total': 10526624, 'tokens/trainable': 159270, 'epoch': '1.359'}
-
 68%|██████▊   | 348/512 [39:15<18:21,  6.72s/it]
 68%|██████▊   | 349/512 [39:21<18:13,  6.71s/it]
                                                 
{'loss': '0.001287', 'grad_norm': '0.1865', 'learning_rate': '3.292e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.46', 'tokens/total': 10556752, 'tokens/trainable': 159720, 'epoch': '1.363'}
-
 68%|██████▊   | 349/512 [39:21<18:13,  6.71s/it]
 68%|██████▊   | 350/512 [39:28<18:12,  6.74s/it]
                                                 
{'loss': '0.0002092', 'grad_norm': '0.01593', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.09', 'tokens/total': 10587264, 'tokens/trainable': 160187, 'epoch': '1.367'}
-
 68%|██████▊   | 350/512 [39:28<18:12,  6.74s/it]
 69%|██████▊   | 351/512 [39:35<18:04,  6.74s/it]
                                                 
{'loss': '0.0001318', 'grad_norm': '0.008725', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 10617504, 'tokens/trainable': 160645, 'epoch': '1.371'}
-
 69%|██████▊   | 351/512 [39:35<18:04,  6.74s/it]
 69%|██████▉   | 352/512 [39:42<17:56,  6.73s/it]
                                                 
{'loss': '0.0008062', 'grad_norm': '0.06378', 'learning_rate': '3.217e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.42', 'tokens/total': 10647760, 'tokens/trainable': 161102, 'epoch': '1.375'}
-
 69%|██████▉   | 352/512 [39:42<17:56,  6.73s/it][2026-08-18 13:18:57,958] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
-
 69%|██████▉   | 353/512 [39:50<19:17,  7.28s/it]
                                                 
{'loss': '7.239e-05', 'grad_norm': '0.00372', 'learning_rate': '3.192e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.86', 'tokens/total': 10678080, 'tokens/trainable': 161555, 'epoch': '1.379'}
-
 69%|██████▉   | 353/512 [39:50<19:17,  7.28s/it]
 69%|██████▉   | 354/512 [39:57<18:43,  7.11s/it]
                                                 
{'loss': '3.681e-05', 'grad_norm': '0.002151', 'learning_rate': '3.167e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.41', 'tokens/train_per_sec_per_gpu': '36.65', 'tokens/total': 10708544, 'tokens/trainable': 162020, 'epoch': '1.383'}
-
 69%|██████▉   | 354/512 [39:57<18:43,  7.11s/it]
 69%|██████▉   | 355/512 [40:04<18:17,  6.99s/it]
                                                 
{'loss': '4.372e-05', 'grad_norm': '0.002642', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '32.05', 'tokens/total': 10739024, 'tokens/trainable': 162448, 'epoch': '1.387'}
-
 69%|██████▉   | 355/512 [40:04<18:17,  6.99s/it]
 70%|██████▉   | 356/512 [40:10<17:53,  6.88s/it]
                                                 
{'loss': '7.116e-05', 'grad_norm': '0.003865', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.61', 'memory/max_allocated (GiB)': '33.61', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '28.2', 'tokens/total': 10769024, 'tokens/trainable': 162863, 'epoch': '1.391'}
-
 70%|██████▉   | 356/512 [40:10<17:53,  6.88s/it]
 70%|██████▉   | 357/512 [40:17<17:38,  6.83s/it]
                                                 
{'loss': '5.407e-05', 'grad_norm': '0.0179', 'learning_rate': '3.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 10799328, 'tokens/trainable': 163318, 'epoch': '1.395'}
-
 70%|██████▉   | 357/512 [40:17<17:38,  6.83s/it]
 70%|██████▉   | 358/512 [40:24<17:28,  6.81s/it]
                                                 
{'loss': '0.01498', 'grad_norm': '0.8479', 'learning_rate': '3.068e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 10829488, 'tokens/trainable': 163767, 'epoch': '1.398'}
-
 70%|██████▉   | 358/512 [40:24<17:28,  6.81s/it]
 70%|███████   | 359/512 [40:31<17:16,  6.78s/it]
                                                 
{'loss': '3.039e-05', 'grad_norm': '0.00157', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '31.78', 'tokens/total': 10859936, 'tokens/trainable': 164208, 'epoch': '1.402'}
-
 70%|███████   | 359/512 [40:31<17:16,  6.78s/it]
 70%|███████   | 360/512 [40:37<17:07,  6.76s/it]
                                                 
{'loss': '0.0001003', 'grad_norm': '0.01467', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.08', 'tokens/total': 10890400, 'tokens/trainable': 164655, 'epoch': '1.406'}
-
 70%|███████   | 360/512 [40:37<17:07,  6.76s/it]
 71%|███████   | 361/512 [40:44<16:56,  6.73s/it]
                                                 
{'loss': '0.00261', 'grad_norm': '0.3479', 'learning_rate': '2.995e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 10920624, 'tokens/trainable': 165100, 'epoch': '1.41'}
-
 71%|███████   | 361/512 [40:44<16:56,  6.73s/it]
 71%|███████   | 362/512 [40:51<16:51,  6.74s/it]
                                                 
{'loss': '6.117e-05', 'grad_norm': '0.004394', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.84', 'tokens/total': 10951136, 'tokens/trainable': 165587, 'epoch': '1.414'}
-
 71%|███████   | 362/512 [40:51<16:51,  6.74s/it]
 71%|███████   | 363/512 [40:57<16:41,  6.72s/it]
                                                 
{'loss': '0.0001775', 'grad_norm': '0.0162', 'learning_rate': '2.947e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.09', 'tokens/total': 10981344, 'tokens/trainable': 166034, 'epoch': '1.418'}
-
 71%|███████   | 363/512 [40:57<16:41,  6.72s/it]
 71%|███████   | 364/512 [41:04<16:34,  6.72s/it]
                                                 
{'loss': '0.0005034', 'grad_norm': '0.09478', 'learning_rate': '2.924e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 11011552, 'tokens/trainable': 166485, 'epoch': '1.422'}
-
 71%|███████   | 364/512 [41:04<16:34,  6.72s/it]
 71%|███████▏  | 365/512 [41:11<16:28,  6.72s/it]
                                                 
{'loss': '0.0001905', 'grad_norm': '0.0149', 'learning_rate': '2.9e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 11041872, 'tokens/trainable': 166973, 'epoch': '1.426'}
-
 71%|███████▏  | 365/512 [41:11<16:28,  6.72s/it]
 71%|███████▏  | 366/512 [41:18<16:20,  6.72s/it]
                                                 
{'loss': '0.0001006', 'grad_norm': '0.005556', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '29.1', 'tokens/total': 11072192, 'tokens/trainable': 167414, 'epoch': '1.43'}
-
 71%|███████▏  | 366/512 [41:18<16:20,  6.72s/it]
 72%|███████▏  | 367/512 [41:24<16:15,  6.72s/it]
                                                 
{'loss': '0.0005889', 'grad_norm': '0.2071', 'learning_rate': '2.853e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.11', 'tokens/total': 11102736, 'tokens/trainable': 167851, 'epoch': '1.434'}
-
 72%|███████▏  | 367/512 [41:24<16:15,  6.72s/it]
 72%|███████▏  | 368/512 [41:31<16:07,  6.72s/it]
                                                 
{'loss': '1.23e-05', 'grad_norm': '0.0007192', 'learning_rate': '2.829e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.48', 'tokens/total': 11133056, 'tokens/trainable': 168294, 'epoch': '1.438'}
-
 72%|███████▏  | 368/512 [41:31<16:07,  6.72s/it]
 72%|███████▏  | 369/512 [41:38<16:04,  6.74s/it]
                                                 
{'loss': '0.0002043', 'grad_norm': '0.03088', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.08', 'tokens/total': 11163568, 'tokens/trainable': 168769, 'epoch': '1.441'}
-
 72%|███████▏  | 369/512 [41:38<16:04,  6.74s/it]
 72%|███████▏  | 370/512 [41:45<15:59,  6.75s/it]
                                                 
{'loss': '0.0001146', 'grad_norm': '0.008817', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.59', 'tokens/total': 11193936, 'tokens/trainable': 169270, 'epoch': '1.445'}
-
 72%|███████▏  | 370/512 [41:45<15:59,  6.75s/it]
 72%|███████▏  | 371/512 [41:51<15:53,  6.77s/it]
                                                 
{'loss': '0.0001343', 'grad_norm': '0.02386', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 11224480, 'tokens/trainable': 169711, 'epoch': '1.449'}
-
 72%|███████▏  | 371/512 [41:51<15:53,  6.77s/it]
 73%|███████▎  | 372/512 [41:58<15:47,  6.77s/it]
                                                 
{'loss': '0.000151', 'grad_norm': '0.0105', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 11254912, 'tokens/trainable': 170185, 'epoch': '1.453'}
-
 73%|███████▎  | 372/512 [41:58<15:47,  6.77s/it]
 73%|███████▎  | 373/512 [42:05<15:39,  6.76s/it]
                                                 
{'loss': '0.0002608', 'grad_norm': '0.02697', 'learning_rate': '2.714e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 11285104, 'tokens/trainable': 170650, 'epoch': '1.457'}
-
 73%|███████▎  | 373/512 [42:05<15:39,  6.76s/it]
 73%|███████▎  | 374/512 [42:12<15:31,  6.75s/it]
                                                 
{'loss': '0.0001453', 'grad_norm': '0.02033', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '31.98', 'tokens/total': 11315600, 'tokens/trainable': 171126, 'epoch': '1.461'}
-
 73%|███████▎  | 374/512 [42:12<15:31,  6.75s/it]
 73%|███████▎  | 375/512 [42:18<15:21,  6.72s/it]
                                                 
{'loss': '0.0004557', 'grad_norm': '0.03326', 'learning_rate': '2.668e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 11345776, 'tokens/trainable': 171600, 'epoch': '1.465'}
-
 73%|███████▎  | 375/512 [42:18<15:21,  6.72s/it]
 73%|███████▎  | 376/512 [42:25<15:13,  6.71s/it]
                                                 
{'loss': '0.0001491', 'grad_norm': '0.01729', 'learning_rate': '2.646e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '28.09', 'tokens/total': 11376208, 'tokens/trainable': 172017, 'epoch': '1.469'}
-
 73%|███████▎  | 376/512 [42:25<15:13,  6.71s/it]
 74%|███████▎  | 377/512 [42:32<15:06,  6.72s/it]
                                                 
{'loss': '6.834e-05', 'grad_norm': '0.04457', 'learning_rate': '2.624e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.99', 'tokens/total': 11406512, 'tokens/trainable': 172493, 'epoch': '1.473'}
-
 74%|███████▎  | 377/512 [42:32<15:06,  6.72s/it]
 74%|███████▍  | 378/512 [42:38<15:00,  6.72s/it]
                                                 
{'loss': '3.245e-05', 'grad_norm': '0.002269', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.27', 'tokens/total': 11436544, 'tokens/trainable': 172951, 'epoch': '1.477'}
-
 74%|███████▍  | 378/512 [42:38<15:00,  6.72s/it]
 74%|███████▍  | 379/512 [42:45<14:53,  6.71s/it]
                                                 
{'loss': '5.487e-05', 'grad_norm': '0.003817', 'learning_rate': '2.579e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.63', 'tokens/total': 11467008, 'tokens/trainable': 173435, 'epoch': '1.48'}
-
 74%|███████▍  | 379/512 [42:45<14:53,  6.71s/it]
 74%|███████▍  | 380/512 [42:52<14:46,  6.71s/it]
                                                 
{'loss': '8.241e-05', 'grad_norm': '0.005269', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.4', 'tokens/total': 11497184, 'tokens/trainable': 173900, 'epoch': '1.484'}
-
 74%|███████▍  | 380/512 [42:52<14:46,  6.71s/it]
 74%|███████▍  | 381/512 [42:59<14:40,  6.72s/it]
                                                 
{'loss': '0.0001424', 'grad_norm': '0.01606', 'learning_rate': '2.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.12', 'tokens/total': 11527600, 'tokens/trainable': 174371, 'epoch': '1.488'}
-
 74%|███████▍  | 381/512 [42:59<14:40,  6.72s/it]
 75%|███████▍  | 382/512 [43:05<14:33,  6.72s/it]
                                                 
{'loss': '5.001e-05', 'grad_norm': '0.005853', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 11557808, 'tokens/trainable': 174826, 'epoch': '1.492'}
-
 75%|███████▍  | 382/512 [43:05<14:33,  6.72s/it]
 75%|███████▍  | 383/512 [43:12<14:27,  6.73s/it]
                                                 
{'loss': '4.21e-05', 'grad_norm': '0.003166', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 11588352, 'tokens/trainable': 175276, 'epoch': '1.496'}
-
 75%|███████▍  | 383/512 [43:12<14:27,  6.73s/it]
 75%|███████▌  | 384/512 [43:19<14:21,  6.73s/it]
                                                 
{'loss': '0.001039', 'grad_norm': '0.08839', 'learning_rate': '2.47e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.69', 'tokens/total': 11618640, 'tokens/trainable': 175720, 'epoch': '1.5'}
-
 75%|███████▌  | 384/512 [43:19<14:21,  6.73s/it][2026-08-18 13:22:34,933] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
-
 75%|███████▌  | 385/512 [43:27<15:26,  7.30s/it]
                                                 
{'loss': '0.000537', 'grad_norm': '0.09447', 'learning_rate': '2.449e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.23', 'tokens/total': 11649072, 'tokens/trainable': 176203, 'epoch': '1.504'}
-
 75%|███████▌  | 385/512 [43:27<15:26,  7.30s/it]
 75%|███████▌  | 386/512 [43:34<14:57,  7.12s/it]
                                                 
{'loss': '1.903e-05', 'grad_norm': '0.001336', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 11679552, 'tokens/trainable': 176651, 'epoch': '1.508'}
-
 75%|███████▌  | 386/512 [43:34<14:57,  7.12s/it]
 76%|███████▌  | 387/512 [43:41<14:35,  7.00s/it]
                                                 
{'loss': '7.681e-05', 'grad_norm': '0.01169', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 11709760, 'tokens/trainable': 177115, 'epoch': '1.512'}
-
 76%|███████▌  | 387/512 [43:41<14:35,  7.00s/it]
 76%|███████▌  | 388/512 [43:47<14:17,  6.92s/it]
                                                 
{'loss': '0.0007352', 'grad_norm': '0.09523', 'learning_rate': '2.385e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 11740288, 'tokens/trainable': 177601, 'epoch': '1.516'}
-
 76%|███████▌  | 388/512 [43:47<14:17,  6.92s/it]
 76%|███████▌  | 389/512 [43:54<14:02,  6.85s/it]
                                                 
{'loss': '6.468e-05', 'grad_norm': '0.005972', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 11770592, 'tokens/trainable': 178050, 'epoch': '1.52'}
-
 76%|███████▌  | 389/512 [43:54<14:02,  6.85s/it]
 76%|███████▌  | 390/512 [44:01<13:51,  6.82s/it]
                                                 
{'loss': '1.599e-05', 'grad_norm': '0.001025', 'learning_rate': '2.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.38', 'tokens/total': 11800752, 'tokens/trainable': 178531, 'epoch': '1.523'}
-
 76%|███████▌  | 390/512 [44:01<13:51,  6.82s/it]
 76%|███████▋  | 391/512 [44:08<13:42,  6.80s/it]
                                                 
{'loss': '0.0004187', 'grad_norm': '0.05442', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 11831360, 'tokens/trainable': 179000, 'epoch': '1.527'}
-
 76%|███████▋  | 391/512 [44:08<13:42,  6.80s/it]
 77%|███████▋  | 392/512 [44:14<13:30,  6.75s/it]
                                                 
{'loss': '1.532e-05', 'grad_norm': '0.002925', 'learning_rate': '2.303e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 11861552, 'tokens/trainable': 179457, 'epoch': '1.531'}
-
 77%|███████▋  | 392/512 [44:14<13:30,  6.75s/it]
 77%|███████▋  | 393/512 [44:21<13:20,  6.73s/it]
                                                 
{'loss': '0.002704', 'grad_norm': '1.64', 'learning_rate': '2.282e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.73', 'tokens/total': 11891904, 'tokens/trainable': 179890, 'epoch': '1.535'}
-
 77%|███████▋  | 393/512 [44:21<13:20,  6.73s/it]
 77%|███████▋  | 394/512 [44:28<13:12,  6.72s/it]
                                                 
{'loss': '3.968e-05', 'grad_norm': '0.003221', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 11922208, 'tokens/trainable': 180341, 'epoch': '1.539'}
-
 77%|███████▋  | 394/512 [44:28<13:12,  6.72s/it]
 77%|███████▋  | 395/512 [44:34<13:03,  6.70s/it]
                                                 
{'loss': '0.0001106', 'grad_norm': '0.0131', 'learning_rate': '2.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 11952672, 'tokens/trainable': 180762, 'epoch': '1.543'}
-
 77%|███████▋  | 395/512 [44:34<13:03,  6.70s/it]
 77%|███████▋  | 396/512 [44:41<12:58,  6.71s/it]
                                                 
{'loss': '8.803e-05', 'grad_norm': '0.01553', 'learning_rate': '2.222e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.04', 'tokens/total': 11983088, 'tokens/trainable': 181226, 'epoch': '1.547'}
-
 77%|███████▋  | 396/512 [44:41<12:58,  6.71s/it]
 78%|███████▊  | 397/512 [44:48<12:52,  6.72s/it]
                                                 
{'loss': '8.931e-05', 'grad_norm': '0.01608', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.15', 'tokens/total': 12013488, 'tokens/trainable': 181714, 'epoch': '1.551'}
-
 78%|███████▊  | 397/512 [44:48<12:52,  6.72s/it]
 78%|███████▊  | 398/512 [44:54<12:44,  6.70s/it]
                                                 
{'loss': '0.0001537', 'grad_norm': '0.05752', 'learning_rate': '2.183e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.52', 'tokens/total': 12043968, 'tokens/trainable': 182142, 'epoch': '1.555'}
-
 78%|███████▊  | 398/512 [44:54<12:44,  6.70s/it]
 78%|███████▊  | 399/512 [45:01<12:38,  6.71s/it]
                                                 
{'loss': '1.811e-05', 'grad_norm': '0.001174', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.82', 'tokens/total': 12074160, 'tokens/trainable': 182604, 'epoch': '1.559'}
-
 78%|███████▊  | 399/512 [45:01<12:38,  6.71s/it]
 78%|███████▊  | 400/512 [45:08<12:30,  6.70s/it]
                                                 
{'loss': '5.335e-05', 'grad_norm': '0.004808', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '25.28', 'tokens/total': 12104352, 'tokens/trainable': 183011, 'epoch': '1.562'}
-
 78%|███████▊  | 400/512 [45:08<12:30,  6.70s/it]
 78%|███████▊  | 401/512 [45:15<12:23,  6.70s/it]
                                                 
{'loss': '6.45e-05', 'grad_norm': '0.00652', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.69', 'tokens/total': 12134240, 'tokens/trainable': 183459, 'epoch': '1.566'}
-
 78%|███████▊  | 401/512 [45:15<12:23,  6.70s/it]
 79%|███████▊  | 402/512 [45:21<12:19,  6.72s/it]
                                                 
{'loss': '6.929e-05', 'grad_norm': '0.006228', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.61', 'tokens/total': 12164480, 'tokens/trainable': 183959, 'epoch': '1.57'}
-
 79%|███████▊  | 402/512 [45:21<12:19,  6.72s/it]
 79%|███████▊  | 403/512 [45:28<12:13,  6.73s/it]
                                                 
{'loss': '0.002115', 'grad_norm': '0.3261', 'learning_rate': '2.086e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.46', 'tokens/total': 12194640, 'tokens/trainable': 184431, 'epoch': '1.574'}
-
 79%|███████▊  | 403/512 [45:28<12:13,  6.73s/it]
 79%|███████▉  | 404/512 [45:35<12:06,  6.73s/it]
                                                 
{'loss': '4.601e-05', 'grad_norm': '0.007742', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.34', 'tokens/total': 12224960, 'tokens/trainable': 184867, 'epoch': '1.578'}
-
 79%|███████▉  | 404/512 [45:35<12:06,  6.73s/it]
 79%|███████▉  | 405/512 [45:42<11:58,  6.71s/it]
                                                 
{'loss': '5.004e-05', 'grad_norm': '0.004995', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.25', 'tokens/total': 12255008, 'tokens/trainable': 185300, 'epoch': '1.582'}
-
 79%|███████▉  | 405/512 [45:42<11:58,  6.71s/it]
 79%|███████▉  | 406/512 [45:48<11:51,  6.71s/it]
                                                 
{'loss': '3.442e-05', 'grad_norm': '0.004599', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 12285248, 'tokens/trainable': 185754, 'epoch': '1.586'}
-
 79%|███████▉  | 406/512 [45:48<11:51,  6.71s/it]
 79%|███████▉  | 407/512 [45:55<11:43,  6.70s/it]
                                                 
{'loss': '1.92e-05', 'grad_norm': '0.001146', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.8', 'tokens/total': 12315552, 'tokens/trainable': 186213, 'epoch': '1.59'}
-
 79%|███████▉  | 407/512 [45:55<11:43,  6.70s/it]
 80%|███████▉  | 408/512 [46:02<11:37,  6.70s/it]
                                                 
{'loss': '5.249e-05', 'grad_norm': '0.004688', 'learning_rate': '1.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.72', 'tokens/total': 12345904, 'tokens/trainable': 186659, 'epoch': '1.594'}
-
 80%|███████▉  | 408/512 [46:02<11:37,  6.70s/it]
 80%|███████▉  | 409/512 [46:08<11:30,  6.71s/it]
                                                 
{'loss': '3.172e-05', 'grad_norm': '0.002638', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 12376336, 'tokens/trainable': 187083, 'epoch': '1.598'}
-
 80%|███████▉  | 409/512 [46:08<11:30,  6.71s/it]
 80%|████████  | 410/512 [46:15<11:25,  6.72s/it]
                                                 
{'loss': '5.453e-05', 'grad_norm': '0.007202', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 12406880, 'tokens/trainable': 187530, 'epoch': '1.602'}
-
 80%|████████  | 410/512 [46:15<11:25,  6.72s/it]
 80%|████████  | 411/512 [46:22<11:18,  6.71s/it]
                                                 
{'loss': '3.531e-05', 'grad_norm': '0.00717', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 12437088, 'tokens/trainable': 187972, 'epoch': '1.605'}
-
 80%|████████  | 411/512 [46:22<11:18,  6.71s/it]
 80%|████████  | 412/512 [46:28<10:55,  6.55s/it]
                                                 
{'loss': '3.509e-05', 'grad_norm': '0.00389', 'learning_rate': '1.922e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '39.89', 'tokens/total': 12465440, 'tokens/trainable': 188417, 'epoch': '1.609'}
-
 80%|████████  | 412/512 [46:28<10:55,  6.55s/it]
 81%|████████  | 413/512 [46:35<10:54,  6.62s/it]
                                                 
{'loss': '0.005698', 'grad_norm': '0.5135', 'learning_rate': '1.904e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 12495968, 'tokens/trainable': 188868, 'epoch': '1.613'}
-
 81%|████████  | 413/512 [46:35<10:54,  6.62s/it]
 81%|████████  | 414/512 [46:41<10:51,  6.65s/it]
                                                 
{'loss': '1.132e-05', 'grad_norm': '0.00101', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.85', 'tokens/total': 12526128, 'tokens/trainable': 189337, 'epoch': '1.617'}
-
 81%|████████  | 414/512 [46:41<10:51,  6.65s/it]
 81%|████████  | 415/512 [46:48<10:45,  6.65s/it]
                                                 
{'loss': '7.355e-06', 'grad_norm': '0.0004971', 'learning_rate': '1.87e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.82', 'tokens/total': 12556256, 'tokens/trainable': 189750, 'epoch': '1.621'}
-
 81%|████████  | 415/512 [46:48<10:45,  6.65s/it]
 81%|████████▏ | 416/512 [46:55<10:41,  6.68s/it]
                                                 
{'loss': '1.419e-05', 'grad_norm': '0.0008849', 'learning_rate': '1.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 12586736, 'tokens/trainable': 190216, 'epoch': '1.625'}
-
 81%|████████▏ | 416/512 [46:55<10:41,  6.68s/it][2026-08-18 13:26:11,049] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
-
 81%|████████▏ | 417/512 [47:03<11:29,  7.26s/it]
                                                 
{'loss': '2.454e-05', 'grad_norm': '0.002204', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 12617168, 'tokens/trainable': 190708, 'epoch': '1.629'}
-
 81%|████████▏ | 417/512 [47:03<11:29,  7.26s/it]
 82%|████████▏ | 418/512 [47:10<11:05,  7.08s/it]
                                                 
{'loss': '0.0002044', 'grad_norm': '0.02701', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.85', 'tokens/total': 12647680, 'tokens/trainable': 191126, 'epoch': '1.633'}
-
 82%|████████▏ | 418/512 [47:10<11:05,  7.08s/it]
 82%|████████▏ | 419/512 [47:17<10:46,  6.95s/it]
                                                 
{'loss': '0.0007251', 'grad_norm': '0.1568', 'learning_rate': '1.802e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.3', 'tokens/total': 12677952, 'tokens/trainable': 191540, 'epoch': '1.637'}
-
 82%|████████▏ | 419/512 [47:17<10:46,  6.95s/it]
 82%|████████▏ | 420/512 [47:23<10:32,  6.87s/it]
                                                 
{'loss': '3.131e-05', 'grad_norm': '0.00331', 'learning_rate': '1.786e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '36.65', 'tokens/total': 12708416, 'tokens/trainable': 192014, 'epoch': '1.641'}
-
 82%|████████▏ | 420/512 [47:23<10:32,  6.87s/it]
 82%|████████▏ | 421/512 [47:30<10:21,  6.83s/it]
                                                 
{'loss': '1.931e-05', 'grad_norm': '0.001256', 'learning_rate': '1.77e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '31.28', 'tokens/total': 12738640, 'tokens/trainable': 192463, 'epoch': '1.645'}
-
 82%|████████▏ | 421/512 [47:30<10:21,  6.83s/it]
 82%|████████▏ | 422/512 [47:37<10:10,  6.79s/it]
                                                 
{'loss': '9.236e-05', 'grad_norm': '0.016', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 12769232, 'tokens/trainable': 192918, 'epoch': '1.648'}
-
 82%|████████▏ | 422/512 [47:37<10:10,  6.79s/it]
 83%|████████▎ | 423/512 [47:44<10:03,  6.78s/it]
                                                 
{'loss': '0.0002709', 'grad_norm': '0.04851', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.35', 'tokens/total': 12799648, 'tokens/trainable': 193425, 'epoch': '1.652'}
-
 83%|████████▎ | 423/512 [47:44<10:03,  6.78s/it]
 83%|████████▎ | 424/512 [47:50<09:55,  6.76s/it]
                                                 
{'loss': '6.991e-05', 'grad_norm': '0.01011', 'learning_rate': '1.722e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.23', 'tokens/total': 12829920, 'tokens/trainable': 193898, 'epoch': '1.656'}
-
 83%|████████▎ | 424/512 [47:50<09:55,  6.76s/it]
 83%|████████▎ | 425/512 [47:57<09:46,  6.74s/it]
                                                 
{'loss': '5.418e-05', 'grad_norm': '0.005467', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.24', 'tokens/total': 12860112, 'tokens/trainable': 194315, 'epoch': '1.66'}
-
 83%|████████▎ | 425/512 [47:57<09:46,  6.74s/it]
 83%|████████▎ | 426/512 [48:04<09:39,  6.74s/it]
                                                 
{'loss': '2.498e-05', 'grad_norm': '0.00158', 'learning_rate': '1.69e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 12890320, 'tokens/trainable': 194796, 'epoch': '1.664'}
-
 83%|████████▎ | 426/512 [48:04<09:39,  6.74s/it]
 83%|████████▎ | 427/512 [48:11<09:33,  6.74s/it]
                                                 
{'loss': '0.000107', 'grad_norm': '0.02416', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 12920656, 'tokens/trainable': 195262, 'epoch': '1.668'}
-
 83%|████████▎ | 427/512 [48:11<09:33,  6.74s/it]
 84%|████████▎ | 428/512 [48:17<09:26,  6.74s/it]
                                                 
{'loss': '7.484e-06', 'grad_norm': '0.0002809', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 12951040, 'tokens/trainable': 195680, 'epoch': '1.672'}
-
 84%|████████▎ | 428/512 [48:17<09:26,  6.74s/it]
 84%|████████▍ | 429/512 [48:24<09:17,  6.72s/it]
                                                 
{'loss': '1.992e-05', 'grad_norm': '0.001109', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.83', 'tokens/total': 12981232, 'tokens/trainable': 196145, 'epoch': '1.676'}
-
 84%|████████▍ | 429/512 [48:24<09:17,  6.72s/it]
 84%|████████▍ | 430/512 [48:31<09:11,  6.72s/it]
                                                 
{'loss': '0.0001097', 'grad_norm': '0.02079', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.39', 'tokens/total': 13011616, 'tokens/trainable': 196626, 'epoch': '1.68'}
-
 84%|████████▍ | 430/512 [48:31<09:11,  6.72s/it]
 84%|████████▍ | 431/512 [48:38<09:07,  6.76s/it]
                                                 
{'loss': '0.002679', 'grad_norm': '0.4408', 'learning_rate': '1.615e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '31.14', 'tokens/total': 13042208, 'tokens/trainable': 197067, 'epoch': '1.684'}
-
 84%|████████▍ | 431/512 [48:38<09:07,  6.76s/it]
 84%|████████▍ | 432/512 [48:44<09:00,  6.75s/it]
                                                 
{'loss': '2.082e-05', 'grad_norm': '0.002634', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.61', 'tokens/total': 13072800, 'tokens/trainable': 197565, 'epoch': '1.688'}
-
 84%|████████▍ | 432/512 [48:44<09:00,  6.75s/it]
 85%|████████▍ | 433/512 [48:51<08:52,  6.73s/it]
                                                 
{'loss': '1.44e-05', 'grad_norm': '0.004509', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 13103328, 'tokens/trainable': 198015, 'epoch': '1.691'}
-
 85%|████████▍ | 433/512 [48:51<08:52,  6.73s/it]
 85%|████████▍ | 434/512 [48:58<08:44,  6.73s/it]
                                                 
{'loss': '0.0001532', 'grad_norm': '0.01481', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 13133472, 'tokens/trainable': 198442, 'epoch': '1.695'}
-
 85%|████████▍ | 434/512 [48:58<08:44,  6.73s/it]
 85%|████████▍ | 435/512 [49:04<08:37,  6.72s/it]
                                                 
{'loss': '1.444e-05', 'grad_norm': '0.003779', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.3', 'tokens/total': 13164064, 'tokens/trainable': 198914, 'epoch': '1.699'}
-
 85%|████████▍ | 435/512 [49:04<08:37,  6.72s/it]
 85%|████████▌ | 436/512 [49:11<08:31,  6.73s/it]
                                                 
{'loss': '0.001711', 'grad_norm': '0.2497', 'learning_rate': '1.544e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.35', 'tokens/total': 13194464, 'tokens/trainable': 199355, 'epoch': '1.703'}
-
 85%|████████▌ | 436/512 [49:11<08:31,  6.73s/it]
 85%|████████▌ | 437/512 [49:18<08:24,  6.73s/it]
                                                 
{'loss': '2.94e-05', 'grad_norm': '0.00248', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 13224992, 'tokens/trainable': 199830, 'epoch': '1.707'}
-
 85%|████████▌ | 437/512 [49:18<08:24,  6.73s/it]
 86%|████████▌ | 438/512 [49:25<08:18,  6.74s/it]
                                                 
{'loss': '1.559e-05', 'grad_norm': '0.001478', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 13255296, 'tokens/trainable': 200338, 'epoch': '1.711'}
-
 86%|████████▌ | 438/512 [49:25<08:18,  6.74s/it]
 86%|████████▌ | 439/512 [49:31<08:09,  6.71s/it]
                                                 
{'loss': '1.468e-05', 'grad_norm': '0.001041', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.79', 'tokens/total': 13285344, 'tokens/trainable': 200785, 'epoch': '1.715'}
-
 86%|████████▌ | 439/512 [49:31<08:09,  6.71s/it]
 86%|████████▌ | 440/512 [49:38<08:04,  6.72s/it]
                                                 
{'loss': '8.095e-06', 'grad_norm': '0.0004247', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 13315632, 'tokens/trainable': 201267, 'epoch': '1.719'}
-
 86%|████████▌ | 440/512 [49:38<08:04,  6.72s/it]
 86%|████████▌ | 441/512 [49:45<07:57,  6.73s/it]
                                                 
{'loss': '1.743e-05', 'grad_norm': '0.00263', 'learning_rate': '1.477e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 13346048, 'tokens/trainable': 201734, 'epoch': '1.723'}
-
 86%|████████▌ | 441/512 [49:45<07:57,  6.73s/it]
 86%|████████▋ | 442/512 [49:51<07:50,  6.72s/it]
                                                 
{'loss': '8.656e-06', 'grad_norm': '0.0005576', 'learning_rate': '1.464e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 13376304, 'tokens/trainable': 202174, 'epoch': '1.727'}
-
 86%|████████▋ | 442/512 [49:51<07:50,  6.72s/it]
 87%|████████▋ | 443/512 [49:58<07:44,  6.73s/it]
                                                 
{'loss': '1.409e-05', 'grad_norm': '0.0009604', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.13', 'tokens/total': 13406720, 'tokens/trainable': 202612, 'epoch': '1.73'}
-
 87%|████████▋ | 443/512 [49:58<07:44,  6.73s/it]
 87%|████████▋ | 444/512 [50:05<07:33,  6.68s/it]
                                                 
{'loss': '4.584e-06', 'grad_norm': '0.0001988', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.96', 'tokens/total': 13436464, 'tokens/trainable': 203020, 'epoch': '1.734'}
-
 87%|████████▋ | 444/512 [50:05<07:33,  6.68s/it]
 87%|████████▋ | 445/512 [50:11<07:27,  6.68s/it]
                                                 
{'loss': '0.000278', 'grad_norm': '0.05945', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 13466672, 'tokens/trainable': 203479, 'epoch': '1.738'}
-
 87%|████████▋ | 445/512 [50:11<07:27,  6.68s/it]
 87%|████████▋ | 446/512 [50:18<07:21,  6.69s/it]
                                                 
{'loss': '3.424e-05', 'grad_norm': '0.005872', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 13496736, 'tokens/trainable': 203946, 'epoch': '1.742'}
-
 87%|████████▋ | 446/512 [50:18<07:21,  6.69s/it]
 87%|████████▋ | 447/512 [50:25<07:14,  6.69s/it]
                                                 
{'loss': '3.27e-05', 'grad_norm': '0.006014', 'learning_rate': '1.402e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '37.43', 'tokens/total': 13526928, 'tokens/trainable': 204382, 'epoch': '1.746'}
-
 87%|████████▋ | 447/512 [50:25<07:14,  6.69s/it]
 88%|████████▊ | 448/512 [50:31<07:08,  6.69s/it]
                                                 
{'loss': '0.0003225', 'grad_norm': '0.05103', 'learning_rate': '1.39e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 13557392, 'tokens/trainable': 204830, 'epoch': '1.75'}
-
 88%|████████▊ | 448/512 [50:31<07:08,  6.69s/it][2026-08-18 13:29:47,708] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
-
 88%|████████▊ | 449/512 [50:40<07:41,  7.32s/it]
                                                 
{'loss': '0.0008941', 'grad_norm': '0.1325', 'learning_rate': '1.378e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.51', 'tokens/total': 13587776, 'tokens/trainable': 205300, 'epoch': '1.754'}
-
 88%|████████▊ | 449/512 [50:40<07:41,  7.32s/it]
 88%|████████▊ | 450/512 [50:47<07:22,  7.14s/it]
                                                 
{'loss': '0.001293', 'grad_norm': '0.2464', 'learning_rate': '1.367e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '30.62', 'tokens/total': 13618112, 'tokens/trainable': 205726, 'epoch': '1.758'}
-
 88%|████████▊ | 450/512 [50:47<07:22,  7.14s/it]
 88%|████████▊ | 451/512 [50:54<07:08,  7.03s/it]
                                                 
{'loss': '6.57e-05', 'grad_norm': '0.01153', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '31.55', 'tokens/total': 13648512, 'tokens/trainable': 206163, 'epoch': '1.762'}
-
 88%|████████▊ | 451/512 [50:54<07:08,  7.03s/it]
 88%|████████▊ | 452/512 [51:00<06:55,  6.93s/it]
                                                 
{'loss': '0.0156', 'grad_norm': '1.082', 'learning_rate': '1.344e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.41', 'tokens/total': 13678704, 'tokens/trainable': 206619, 'epoch': '1.766'}
-
 88%|████████▊ | 452/512 [51:00<06:55,  6.93s/it]
 88%|████████▊ | 453/512 [51:07<06:46,  6.88s/it]
                                                 
{'loss': '1.203e-05', 'grad_norm': '0.0007327', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.18', 'tokens/total': 13709232, 'tokens/trainable': 207101, 'epoch': '1.77'}
-
 88%|████████▊ | 453/512 [51:07<06:46,  6.88s/it]
 89%|████████▊ | 454/512 [51:14<06:35,  6.83s/it]
                                                 
{'loss': '1.262e-05', 'grad_norm': '0.002392', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.69', 'tokens/total': 13739600, 'tokens/trainable': 207546, 'epoch': '1.773'}
-
 89%|████████▊ | 454/512 [51:14<06:35,  6.83s/it]
 89%|████████▉ | 455/512 [51:21<06:27,  6.79s/it]
                                                 
{'loss': '2.477e-05', 'grad_norm': '0.002097', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '34.25', 'tokens/total': 13769936, 'tokens/trainable': 207989, 'epoch': '1.777'}
-
 89%|████████▉ | 455/512 [51:21<06:27,  6.79s/it]
 89%|████████▉ | 456/512 [51:27<06:18,  6.76s/it]
                                                 
{'loss': '1.457e-05', 'grad_norm': '0.0007976', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 13800240, 'tokens/trainable': 208402, 'epoch': '1.781'}
-
 89%|████████▉ | 456/512 [51:27<06:18,  6.76s/it]
 89%|████████▉ | 457/512 [51:34<06:11,  6.75s/it]
                                                 
{'loss': '7.903e-05', 'grad_norm': '0.01138', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.61', 'tokens/total': 13830512, 'tokens/trainable': 208904, 'epoch': '1.785'}
-
 89%|████████▉ | 457/512 [51:34<06:11,  6.75s/it]
 89%|████████▉ | 458/512 [51:41<06:03,  6.74s/it]
                                                 
{'loss': '1.925e-05', 'grad_norm': '0.001357', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.64', 'tokens/total': 13860832, 'tokens/trainable': 209382, 'epoch': '1.789'}
-
 89%|████████▉ | 458/512 [51:41<06:03,  6.74s/it]
 90%|████████▉ | 459/512 [51:47<05:56,  6.72s/it]
                                                 
{'loss': '1.119e-05', 'grad_norm': '0.000474', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 13890992, 'tokens/trainable': 209833, 'epoch': '1.793'}
-
 90%|████████▉ | 459/512 [51:47<05:56,  6.72s/it]
 90%|████████▉ | 460/512 [51:54<05:48,  6.71s/it]
                                                 
{'loss': '0.000175', 'grad_norm': '0.04677', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.5', 'tokens/total': 13921424, 'tokens/trainable': 210284, 'epoch': '1.797'}
-
 90%|████████▉ | 460/512 [51:54<05:48,  6.71s/it]
 90%|█████████ | 461/512 [52:01<05:41,  6.69s/it]
                                                 
{'loss': '9.222e-05', 'grad_norm': '0.02466', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 13951504, 'tokens/trainable': 210769, 'epoch': '1.801'}
-
 90%|█████████ | 461/512 [52:01<05:41,  6.69s/it]
 90%|█████████ | 462/512 [52:07<05:26,  6.54s/it]
                                                 
{'loss': '1.07e-05', 'grad_norm': '0.000474', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 13979728, 'tokens/trainable': 211193, 'epoch': '1.805'}
-
 90%|█████████ | 462/512 [52:07<05:26,  6.54s/it]
 90%|█████████ | 463/512 [52:14<05:23,  6.60s/it]
                                                 
{'loss': '1.394e-05', 'grad_norm': '0.0007591', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 14009936, 'tokens/trainable': 211655, 'epoch': '1.809'}
-
 90%|█████████ | 463/512 [52:14<05:23,  6.60s/it]
 91%|█████████ | 464/512 [52:20<05:17,  6.61s/it]
                                                 
{'loss': '2.675e-05', 'grad_norm': '0.002109', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.28', 'tokens/total': 14039872, 'tokens/trainable': 212085, 'epoch': '1.812'}
-
 91%|█████████ | 464/512 [52:20<05:17,  6.61s/it]
 91%|█████████ | 465/512 [52:27<05:12,  6.65s/it]
                                                 
{'loss': '1.082e-05', 'grad_norm': '0.0009436', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.61', 'tokens/total': 14070096, 'tokens/trainable': 212570, 'epoch': '1.816'}
-
 91%|█████████ | 465/512 [52:27<05:12,  6.65s/it]
 91%|█████████ | 466/512 [52:34<05:06,  6.66s/it]
                                                 
{'loss': '5.798e-05', 'grad_norm': '0.01104', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.86', 'tokens/total': 14100192, 'tokens/trainable': 213011, 'epoch': '1.82'}
-
 91%|█████████ | 466/512 [52:34<05:06,  6.66s/it]
 91%|█████████ | 467/512 [52:41<05:00,  6.69s/it]
                                                 
{'loss': '2.876e-05', 'grad_norm': '0.008272', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.44', 'tokens/total': 14130448, 'tokens/trainable': 213450, 'epoch': '1.824'}
-
 91%|█████████ | 467/512 [52:41<05:00,  6.69s/it]
 91%|█████████▏| 468/512 [52:47<04:53,  6.68s/it]
                                                 
{'loss': '0.0001435', 'grad_norm': '0.02348', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.62', 'tokens/total': 14160480, 'tokens/trainable': 213898, 'epoch': '1.828'}
-
 91%|█████████▏| 468/512 [52:47<04:53,  6.68s/it]
 92%|█████████▏| 469/512 [52:54<04:47,  6.69s/it]
                                                 
{'loss': '2.56e-05', 'grad_norm': '0.002308', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.78', 'tokens/total': 14191152, 'tokens/trainable': 214366, 'epoch': '1.832'}
-
 92%|█████████▏| 469/512 [52:54<04:47,  6.69s/it]
 92%|█████████▏| 470/512 [53:00<04:35,  6.55s/it]
                                                 
{'loss': '2.02e-05', 'grad_norm': '0.001436', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.64', 'tokens/total': 14219392, 'tokens/trainable': 214800, 'epoch': '1.836'}
-
 92%|█████████▏| 470/512 [53:00<04:35,  6.55s/it]
 92%|█████████▏| 471/512 [53:07<04:31,  6.63s/it]
                                                 
{'loss': '4.487e-05', 'grad_norm': '0.0116', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 14249920, 'tokens/trainable': 215292, 'epoch': '1.84'}
-
 92%|█████████▏| 471/512 [53:07<04:31,  6.63s/it]
 92%|█████████▏| 472/512 [53:14<04:26,  6.65s/it]
                                                 
{'loss': '2.048e-05', 'grad_norm': '0.001163', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 14280048, 'tokens/trainable': 215741, 'epoch': '1.844'}
-
 92%|█████████▏| 472/512 [53:14<04:26,  6.65s/it]
 92%|█████████▏| 473/512 [53:20<04:20,  6.68s/it]
                                                 
{'loss': '5.009e-05', 'grad_norm': '0.006094', 'learning_rate': '1.149e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.64', 'tokens/total': 14310432, 'tokens/trainable': 216194, 'epoch': '1.848'}
-
 92%|█████████▏| 473/512 [53:20<04:20,  6.68s/it]
 93%|█████████▎| 474/512 [53:27<04:13,  6.68s/it]
                                                 
{'loss': '3.059e-05', 'grad_norm': '0.003623', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.3', 'tokens/total': 14340496, 'tokens/trainable': 216634, 'epoch': '1.852'}
-
 93%|█████████▎| 474/512 [53:27<04:13,  6.68s/it]
 93%|█████████▎| 475/512 [53:34<04:07,  6.68s/it]
                                                 
{'loss': '3.954e-05', 'grad_norm': '0.006924', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '30.48', 'tokens/total': 14370720, 'tokens/trainable': 217040, 'epoch': '1.855'}
-
 93%|█████████▎| 475/512 [53:34<04:07,  6.68s/it]
 93%|█████████▎| 476/512 [53:41<04:01,  6.71s/it]
                                                 
{'loss': '3.342e-05', 'grad_norm': '0.00236', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.55', 'tokens/total': 14401136, 'tokens/trainable': 217530, 'epoch': '1.859'}
-
 93%|█████████▎| 476/512 [53:41<04:01,  6.71s/it]
 93%|█████████▎| 477/512 [53:47<03:54,  6.70s/it]
                                                 
{'loss': '0.0001011', 'grad_norm': '0.009585', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.47', 'tokens/total': 14431360, 'tokens/trainable': 217988, 'epoch': '1.863'}
-
 93%|█████████▎| 477/512 [53:47<03:54,  6.70s/it]
 93%|█████████▎| 478/512 [53:54<03:49,  6.74s/it]
                                                 
{'loss': '0.0002209', 'grad_norm': '0.03044', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.47', 'tokens/total': 14461968, 'tokens/trainable': 218461, 'epoch': '1.867'}
-
 93%|█████████▎| 478/512 [53:54<03:49,  6.74s/it]
 94%|█████████▎| 479/512 [54:01<03:41,  6.71s/it]
                                                 
{'loss': '0.0002753', 'grad_norm': '0.03494', 'learning_rate': '1.108e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.77', 'tokens/total': 14491984, 'tokens/trainable': 218909, 'epoch': '1.871'}
-
 94%|█████████▎| 479/512 [54:01<03:41,  6.71s/it]
 94%|█████████▍| 480/512 [54:07<03:35,  6.73s/it]
                                                 
{'loss': '0.0002576', 'grad_norm': '0.301', 'learning_rate': '1.102e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 14522544, 'tokens/trainable': 219350, 'epoch': '1.875'}
-
 94%|█████████▍| 480/512 [54:07<03:35,  6.73s/it][2026-08-18 13:33:23,660] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
-
 94%|█████████▍| 481/512 [54:16<03:46,  7.30s/it]
                                                 
{'loss': '0.0002504', 'grad_norm': '0.07506', 'learning_rate': '1.096e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '29.47', 'tokens/total': 14552560, 'tokens/trainable': 219776, 'epoch': '1.879'}
-
 94%|█████████▍| 481/512 [54:16<03:46,  7.30s/it]
 94%|█████████▍| 482/512 [54:23<03:34,  7.14s/it]
                                                 
{'loss': '1.01e-05', 'grad_norm': '0.0008993', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 14583056, 'tokens/trainable': 220284, 'epoch': '1.883'}
-
 94%|█████████▍| 482/512 [54:23<03:34,  7.14s/it]
 94%|█████████▍| 483/512 [54:30<03:22,  7.00s/it]
                                                 
{'loss': '2.53e-05', 'grad_norm': '0.001677', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.01', 'tokens/total': 14613152, 'tokens/trainable': 220714, 'epoch': '1.887'}
-
 94%|█████████▍| 483/512 [54:30<03:22,  7.00s/it]
 95%|█████████▍| 484/512 [54:36<03:13,  6.91s/it]
                                                 
{'loss': '0.0001353', 'grad_norm': '0.01496', 'learning_rate': '1.079e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 14643328, 'tokens/trainable': 221156, 'epoch': '1.891'}
-
 95%|█████████▍| 484/512 [54:36<03:13,  6.91s/it]
 95%|█████████▍| 485/512 [54:43<03:04,  6.84s/it]
                                                 
{'loss': '1.105e-05', 'grad_norm': '0.004265', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.53', 'tokens/total': 14673488, 'tokens/trainable': 221605, 'epoch': '1.895'}
-
 95%|█████████▍| 485/512 [54:43<03:04,  6.84s/it]
 95%|█████████▍| 486/512 [54:50<02:56,  6.79s/it]
                                                 
{'loss': '2.202e-05', 'grad_norm': '0.002974', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 14704000, 'tokens/trainable': 222042, 'epoch': '1.898'}
-
 95%|█████████▍| 486/512 [54:50<02:56,  6.79s/it]
 95%|█████████▌| 487/512 [54:56<02:48,  6.76s/it]
                                                 
{'loss': '7.452e-05', 'grad_norm': '0.008692', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.95', 'tokens/total': 14733984, 'tokens/trainable': 222474, 'epoch': '1.902'}
-
 95%|█████████▌| 487/512 [54:56<02:48,  6.76s/it]
 95%|█████████▌| 488/512 [55:03<02:42,  6.75s/it]
                                                 
{'loss': '7.341e-06', 'grad_norm': '0.0005811', 'learning_rate': '1.058e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 14764400, 'tokens/trainable': 222967, 'epoch': '1.906'}
-
 95%|█████████▌| 488/512 [55:03<02:42,  6.75s/it]
 96%|█████████▌| 489/512 [55:10<02:34,  6.73s/it]
                                                 
{'loss': '3.556e-05', 'grad_norm': '0.004224', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.78', 'tokens/total': 14794672, 'tokens/trainable': 223420, 'epoch': '1.91'}
-
 96%|█████████▌| 489/512 [55:10<02:34,  6.73s/it]
 96%|█████████▌| 490/512 [55:16<02:27,  6.71s/it]
                                                 
{'loss': '2.799e-05', 'grad_norm': '0.01521', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '29.42', 'tokens/total': 14824784, 'tokens/trainable': 223864, 'epoch': '1.914'}
-
 96%|█████████▌| 490/512 [55:16<02:27,  6.71s/it]
 96%|█████████▌| 491/512 [55:23<02:21,  6.72s/it]
                                                 
{'loss': '4.807e-05', 'grad_norm': '0.02447', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 14855056, 'tokens/trainable': 224321, 'epoch': '1.918'}
-
 96%|█████████▌| 491/512 [55:23<02:21,  6.72s/it]
 96%|█████████▌| 492/512 [55:30<02:14,  6.73s/it]
                                                 
{'loss': '3.398e-05', 'grad_norm': '0.003633', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.29', 'tokens/total': 14885392, 'tokens/trainable': 224802, 'epoch': '1.922'}
-
 96%|█████████▌| 492/512 [55:30<02:14,  6.73s/it]
 96%|█████████▋| 493/512 [55:37<02:08,  6.74s/it]
                                                 
{'loss': '1.887e-05', 'grad_norm': '0.0009831', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.77', 'tokens/total': 14915760, 'tokens/trainable': 225284, 'epoch': '1.926'}
-
 96%|█████████▋| 493/512 [55:37<02:08,  6.74s/it]
 96%|█████████▋| 494/512 [55:43<02:01,  6.74s/it]
                                                 
{'loss': '3.033e-05', 'grad_norm': '0.002036', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 14946192, 'tokens/trainable': 225733, 'epoch': '1.93'}
-
 96%|█████████▋| 494/512 [55:43<02:01,  6.74s/it]
 97%|█████████▋| 495/512 [55:50<01:54,  6.72s/it]
                                                 
{'loss': '6.913e-06', 'grad_norm': '0.0002904', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 14976272, 'tokens/trainable': 226172, 'epoch': '1.934'}
-
 97%|█████████▋| 495/512 [55:50<01:54,  6.72s/it]
 97%|█████████▋| 496/512 [55:57<01:47,  6.73s/it]
                                                 
{'loss': '0.0007051', 'grad_norm': '0.1167', 'learning_rate': '1.027e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.77', 'tokens/total': 15006544, 'tokens/trainable': 226636, 'epoch': '1.938'}
-
 97%|█████████▋| 496/512 [55:57<01:47,  6.73s/it]
 97%|█████████▋| 497/512 [56:03<01:40,  6.72s/it]
                                                 
{'loss': '1.047e-05', 'grad_norm': '0.0008453', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.91', 'tokens/total': 15037040, 'tokens/trainable': 227067, 'epoch': '1.941'}
-
 97%|█████████▋| 497/512 [56:03<01:40,  6.72s/it]
 97%|█████████▋| 498/512 [56:10<01:34,  6.72s/it]
                                                 
{'loss': '0.0001397', 'grad_norm': '0.02769', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.58', 'tokens/total': 15067344, 'tokens/trainable': 227532, 'epoch': '1.945'}
-
 97%|█████████▋| 498/512 [56:10<01:34,  6.72s/it]
 97%|█████████▋| 499/512 [56:17<01:27,  6.72s/it]
                                                 
{'loss': '0.001548', 'grad_norm': '0.2443', 'learning_rate': '1.018e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 15097648, 'tokens/trainable': 228007, 'epoch': '1.949'}
-
 97%|█████████▋| 499/512 [56:17<01:27,  6.72s/it]
 98%|█████████▊| 500/512 [56:24<01:20,  6.72s/it]
                                                 
{'loss': '1.88e-05', 'grad_norm': '0.001633', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.84', 'tokens/total': 15128000, 'tokens/trainable': 228457, 'epoch': '1.953'}
-
 98%|█████████▊| 500/512 [56:24<01:20,  6.72s/it]
 98%|█████████▊| 501/512 [56:30<01:13,  6.71s/it]
                                                 
{'loss': '3.361e-05', 'grad_norm': '0.004103', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.19', 'tokens/total': 15158352, 'tokens/trainable': 228928, 'epoch': '1.957'}
-
 98%|█████████▊| 501/512 [56:30<01:13,  6.71s/it]
 98%|█████████▊| 502/512 [56:37<01:07,  6.74s/it]
                                                 
{'loss': '6.559e-06', 'grad_norm': '0.0003138', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 15189136, 'tokens/trainable': 229368, 'epoch': '1.961'}
-
 98%|█████████▊| 502/512 [56:37<01:07,  6.74s/it]
 98%|█████████▊| 503/512 [56:44<01:00,  6.72s/it]
                                                 
{'loss': '1.713e-05', 'grad_norm': '0.002323', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.75', 'tokens/total': 15219616, 'tokens/trainable': 229815, 'epoch': '1.965'}
-
 98%|█████████▊| 503/512 [56:44<01:00,  6.72s/it]
 98%|█████████▊| 504/512 [56:51<00:53,  6.73s/it]
                                                 
{'loss': '0.002064', 'grad_norm': '0.1925', 'learning_rate': '1.008e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '38.82', 'tokens/total': 15250032, 'tokens/trainable': 230319, 'epoch': '1.969'}
-
 98%|█████████▊| 504/512 [56:51<00:53,  6.73s/it]
 99%|█████████▊| 505/512 [56:57<00:47,  6.73s/it]
                                                 
{'loss': '0.0001955', 'grad_norm': '0.02836', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.42', 'tokens/total': 15280592, 'tokens/trainable': 230807, 'epoch': '1.973'}
-
 99%|█████████▊| 505/512 [56:57<00:47,  6.73s/it]
 99%|█████████▉| 506/512 [57:04<00:40,  6.73s/it]
                                                 
{'loss': '1.307e-05', 'grad_norm': '0.0009315', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.16', 'tokens/total': 15310816, 'tokens/trainable': 231257, 'epoch': '1.977'}
-
 99%|█████████▉| 506/512 [57:04<00:40,  6.73s/it]
 99%|█████████▉| 507/512 [57:11<00:33,  6.75s/it]
                                                 
{'loss': '0.0003411', 'grad_norm': '0.04385', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.31', 'tokens/total': 15341424, 'tokens/trainable': 231746, 'epoch': '1.98'}
-
 99%|█████████▉| 507/512 [57:11<00:33,  6.75s/it]
 99%|█████████▉| 508/512 [57:17<00:26,  6.58s/it]
                                                 
{'loss': '7.53e-06', 'grad_norm': '0.0003456', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.33', 'memory/max_allocated (GiB)': '33.33', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 15369664, 'tokens/trainable': 232172, 'epoch': '1.984'}
-
 99%|█████████▉| 508/512 [57:17<00:26,  6.58s/it]
 99%|█████████▉| 509/512 [57:24<00:20,  6.71s/it]
                                                 
{'loss': '0.000155', 'grad_norm': '0.03426', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.04', 'tokens/total': 15400496, 'tokens/trainable': 232680, 'epoch': '1.988'}
-
 99%|█████████▉| 509/512 [57:24<00:20,  6.71s/it]
100%|█████████▉| 510/512 [57:31<00:13,  6.73s/it]
                                                 
{'loss': '1.387e-05', 'grad_norm': '0.005425', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.02', 'tokens/total': 15430864, 'tokens/trainable': 233169, 'epoch': '1.992'}
-
100%|█████████▉| 510/512 [57:31<00:13,  6.73s/it]
100%|█████████▉| 511/512 [57:37<00:06,  6.72s/it]
                                                 
{'loss': '0.0001456', 'grad_norm': '0.04471', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 15461232, 'tokens/trainable': 233609, 'epoch': '1.996'}
-
100%|█████████▉| 511/512 [57:37<00:06,  6.72s/it]
100%|██████████| 512/512 [57:44<00:00,  6.76s/it]
                                                 
{'loss': '2.066e-05', 'grad_norm': '0.001178', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.16', 'tokens/total': 15491760, 'tokens/trainable': 234078, 'epoch': '2'}
-
100%|██████████| 512/512 [57:44<00:00,  6.76s/it][2026-08-18 13:37:00,528] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
-
                                                 
{'train_runtime': '3467', 'train_samples_per_second': '4.726', 'train_steps_per_second': '0.148', 'train_loss': '0.006838', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '36.16', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
-
100%|██████████| 512/512 [57:46<00:00,  6.76s/it]
100%|██████████| 512/512 [57:46<00:00,  6.77s/it]
-[2026-08-18 13:37:02,416] [INFO] [axolotl.train] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
-[2026-08-18 13:37:02,873] [INFO] [axolotl.train] Model successfully saved to /workspace/wave/training/checkpoints
+
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
                                                 
{'loss': '0.1392', 'grad_norm': '1.378', 'learning_rate': '0', 'ppl': '1.149', 'memory/max_active (GiB)': '32.79', 'memory/max_allocated (GiB)': '32.79', 'memory/device_reserved (GiB)': '33.9', 'tokens/train_per_sec_per_gpu': '27.09', 'tokens/total': 30176, 'tokens/trainable': 417, 'epoch': '0.003906'}
+
  0%|          | 1/512 [00:08<1:09:52,  8.20s/it]
  0%|          | 2/512 [00:14<1:01:59,  7.29s/it]
                                                 
{'loss': '0.151', 'grad_norm': '1.233', 'learning_rate': '4e-06', 'ppl': '1.163', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '34.63', 'tokens/total': 60272, 'tokens/trainable': 873, 'epoch': '0.007812'}
+
  0%|          | 2/512 [00:14<1:01:59,  7.29s/it]
  1%|          | 3/512 [00:21<59:40,  7.03s/it]  
                                               
{'loss': '0.1442', 'grad_norm': '5.272', 'learning_rate': '8e-06', 'ppl': '1.155', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '34.72', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 90512, 'tokens/trainable': 1353, 'epoch': '0.01172'}
+
  1%|          | 3/512 [00:21<59:40,  7.03s/it]
  1%|          | 4/512 [00:28<58:33,  6.92s/it]
                                               
{'loss': '0.1343', 'grad_norm': '1.396', 'learning_rate': '1.2e-05', 'ppl': '1.144', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.73', 'tokens/train_per_sec_per_gpu': '31.19', 'tokens/total': 120944, 'tokens/trainable': 1778, 'epoch': '0.01562'}
+
  1%|          | 4/512 [00:28<58:33,  6.92s/it]
  1%|          | 5/512 [00:35<57:44,  6.83s/it]
                                               
{'loss': '0.1441', 'grad_norm': '1.407', 'learning_rate': '1.6e-05', 'ppl': '1.155', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.12', 'tokens/train_per_sec_per_gpu': '36.8', 'tokens/total': 151440, 'tokens/trainable': 2261, 'epoch': '0.01953'}
+
  1%|          | 5/512 [00:35<57:44,  6.83s/it]
  1%|          | 6/512 [00:41<57:14,  6.79s/it]
                                               
{'loss': '0.111', 'grad_norm': '1.676', 'learning_rate': '2e-05', 'ppl': '1.117', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 181984, 'tokens/trainable': 2705, 'epoch': '0.02344'}
+
  1%|          | 6/512 [00:41<57:14,  6.79s/it]
  1%|▏         | 7/512 [00:48<56:53,  6.76s/it]
                                               
{'loss': '0.08474', 'grad_norm': '1.306', 'learning_rate': '2.4e-05', 'ppl': '1.088', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 212336, 'tokens/trainable': 3136, 'epoch': '0.02734'}
+
  1%|▏         | 7/512 [00:48<56:53,  6.76s/it]
  2%|▏         | 8/512 [00:55<56:36,  6.74s/it]
                                               
{'loss': '0.08459', 'grad_norm': '15.48', 'learning_rate': '2.8e-05', 'ppl': '1.088', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 242592, 'tokens/trainable': 3579, 'epoch': '0.03125'}
+
  2%|▏         | 8/512 [00:55<56:36,  6.74s/it]
  2%|▏         | 9/512 [01:01<56:18,  6.72s/it]
                                               
{'loss': '0.06938', 'grad_norm': '1.313', 'learning_rate': '3.2e-05', 'ppl': '1.072', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.12', 'tokens/train_per_sec_per_gpu': '29.12', 'tokens/total': 272784, 'tokens/trainable': 4000, 'epoch': '0.03516'}
+
  2%|▏         | 9/512 [01:01<56:18,  6.72s/it]
  2%|▏         | 10/512 [01:08<56:07,  6.71s/it]
                                                
{'loss': '0.04889', 'grad_norm': '1.293', 'learning_rate': '3.6e-05', 'ppl': '1.05', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '37.66', 'tokens/total': 303184, 'tokens/trainable': 4474, 'epoch': '0.03906'}
+
  2%|▏         | 10/512 [01:08<56:07,  6.71s/it]
  2%|▏         | 11/512 [01:15<55:59,  6.71s/it]
                                                
{'loss': '0.03002', 'grad_norm': '1.032', 'learning_rate': '4e-05', 'ppl': '1.03', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '33.15', 'tokens/total': 333296, 'tokens/trainable': 4942, 'epoch': '0.04297'}
+
  2%|▏         | 11/512 [01:15<55:59,  6.71s/it]
  2%|▏         | 12/512 [01:21<55:54,  6.71s/it]
                                                
{'loss': '0.05269', 'grad_norm': '1.494', 'learning_rate': '4.4e-05', 'ppl': '1.054', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 363840, 'tokens/trainable': 5424, 'epoch': '0.04688'}
+
  2%|▏         | 12/512 [01:21<55:54,  6.71s/it]
  3%|▎         | 13/512 [01:28<55:49,  6.71s/it]
                                                
{'loss': '0.06069', 'grad_norm': '2.024', 'learning_rate': '4.8e-05', 'ppl': '1.063', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 394112, 'tokens/trainable': 5855, 'epoch': '0.05078'}
+
  3%|▎         | 13/512 [01:28<55:49,  6.71s/it]
  3%|▎         | 14/512 [01:35<55:50,  6.73s/it]
                                                
{'loss': '0.03631', 'grad_norm': '1.517', 'learning_rate': '5.2e-05', 'ppl': '1.037', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '34.75', 'tokens/total': 424656, 'tokens/trainable': 6344, 'epoch': '0.05469'}
+
  3%|▎         | 14/512 [01:35<55:50,  6.73s/it]
  3%|▎         | 15/512 [01:42<55:44,  6.73s/it]
                                                
{'loss': '0.05663', 'grad_norm': '1.741', 'learning_rate': '5.6e-05', 'ppl': '1.058', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 455152, 'tokens/trainable': 6825, 'epoch': '0.05859'}
+
  3%|▎         | 15/512 [01:42<55:44,  6.73s/it]
  3%|▎         | 16/512 [01:48<55:23,  6.70s/it]
                                                
{'loss': '0.07075', 'grad_norm': '2.507', 'learning_rate': '6e-05', 'ppl': '1.073', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '30.03', 'tokens/total': 485328, 'tokens/trainable': 7230, 'epoch': '0.0625'}
+
  3%|▎         | 16/512 [01:48<55:23,  6.70s/it]
  3%|▎         | 17/512 [01:55<55:22,  6.71s/it]
                                                
{'loss': '0.041', 'grad_norm': '1.257', 'learning_rate': '6.4e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '35.91', 'tokens/total': 515744, 'tokens/trainable': 7699, 'epoch': '0.06641'}
+
  3%|▎         | 17/512 [01:55<55:22,  6.71s/it]
  4%|▎         | 18/512 [02:02<55:25,  6.73s/it]
                                                
{'loss': '0.02119', 'grad_norm': '36.94', 'learning_rate': '6.8e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.81', 'tokens/train_per_sec_per_gpu': '33.78', 'tokens/total': 546144, 'tokens/trainable': 8184, 'epoch': '0.07031'}
+
  4%|▎         | 18/512 [02:02<55:25,  6.73s/it]
  4%|▎         | 19/512 [02:08<55:19,  6.73s/it]
                                                
{'loss': '0.01386', 'grad_norm': '0.3777', 'learning_rate': '7.2e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 576560, 'tokens/trainable': 8675, 'epoch': '0.07422'}
+
  4%|▎         | 19/512 [02:08<55:19,  6.73s/it]
  4%|▍         | 20/512 [02:15<55:15,  6.74s/it]
                                                
{'loss': '0.0455', 'grad_norm': '1.427', 'learning_rate': '7.6e-05', 'ppl': '1.047', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.72', 'tokens/total': 607056, 'tokens/trainable': 9189, 'epoch': '0.07812'}
+
  4%|▍         | 20/512 [02:15<55:15,  6.74s/it]
  4%|▍         | 21/512 [02:22<55:08,  6.74s/it]
                                                
{'loss': '0.0321', 'grad_norm': '0.5615', 'learning_rate': '8e-05', 'ppl': '1.033', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.61', 'tokens/total': 637488, 'tokens/trainable': 9614, 'epoch': '0.08203'}
+
  4%|▍         | 21/512 [02:22<55:08,  6.74s/it]
  4%|▍         | 22/512 [02:29<54:56,  6.73s/it]
                                                
{'loss': '0.04279', 'grad_norm': '1.153', 'learning_rate': '8.4e-05', 'ppl': '1.044', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '36.11', 'tokens/total': 667632, 'tokens/trainable': 10086, 'epoch': '0.08594'}
+
  4%|▍         | 22/512 [02:29<54:56,  6.73s/it]
  4%|▍         | 23/512 [02:35<54:49,  6.73s/it]
                                                
{'loss': '0.01939', 'grad_norm': '0.9407', 'learning_rate': '8.8e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 697872, 'tokens/trainable': 10535, 'epoch': '0.08984'}
+
  4%|▍         | 23/512 [02:35<54:49,  6.73s/it]
  5%|▍         | 24/512 [02:42<54:43,  6.73s/it]
                                                
{'loss': '0.01258', 'grad_norm': '0.35', 'learning_rate': '9.2e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 728432, 'tokens/trainable': 10993, 'epoch': '0.09375'}
+
  5%|▍         | 24/512 [02:42<54:43,  6.73s/it]
  5%|▍         | 25/512 [02:48<53:20,  6.57s/it]
                                                
{'loss': '0.04129', 'grad_norm': '1.678', 'learning_rate': '9.6e-05', 'ppl': '1.042', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '39.78', 'tokens/total': 756656, 'tokens/trainable': 11417, 'epoch': '0.09766'}
+
  5%|▍         | 25/512 [02:48<53:20,  6.57s/it]
  5%|▌         | 26/512 [02:55<53:29,  6.60s/it]
                                                
{'loss': '0.008186', 'grad_norm': '0.5761', 'learning_rate': '0.0001', 'ppl': '1.008', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 787120, 'tokens/trainable': 11874, 'epoch': '0.1016'}
+
  5%|▌         | 26/512 [02:55<53:29,  6.60s/it]
  5%|▌         | 27/512 [03:02<53:42,  6.64s/it]
                                                
{'loss': '0.01083', 'grad_norm': '5.236', 'learning_rate': '0.0001', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '31.62', 'tokens/total': 817504, 'tokens/trainable': 12316, 'epoch': '0.1055'}
+
  5%|▌         | 27/512 [03:02<53:42,  6.64s/it]
  5%|▌         | 28/512 [03:09<53:51,  6.68s/it]
                                                
{'loss': '0.05158', 'grad_norm': '1.404', 'learning_rate': '0.0001', 'ppl': '1.053', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '37.76', 'tokens/total': 848128, 'tokens/trainable': 12800, 'epoch': '0.1094'}
+
  5%|▌         | 28/512 [03:09<53:51,  6.68s/it]
  6%|▌         | 29/512 [03:15<53:51,  6.69s/it]
                                                
{'loss': '0.05181', 'grad_norm': '3.018', 'learning_rate': '9.999e-05', 'ppl': '1.053', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.43', 'tokens/total': 878448, 'tokens/trainable': 13270, 'epoch': '0.1133'}
+
  6%|▌         | 29/512 [03:15<53:51,  6.69s/it]
  6%|▌         | 30/512 [03:22<53:53,  6.71s/it]
                                                
{'loss': '0.02502', 'grad_norm': '1.414', 'learning_rate': '9.999e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 909008, 'tokens/trainable': 13752, 'epoch': '0.1172'}
+
  6%|▌         | 30/512 [03:22<53:53,  6.71s/it]
  6%|▌         | 31/512 [03:29<53:50,  6.72s/it]
                                                
{'loss': '0.008201', 'grad_norm': '1.161', 'learning_rate': '9.998e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 939312, 'tokens/trainable': 14218, 'epoch': '0.1211'}
+
  6%|▌         | 31/512 [03:29<53:50,  6.72s/it]
  6%|▋         | 32/512 [03:35<53:32,  6.69s/it]
                                                
{'loss': '0.006646', 'grad_norm': '0.2631', 'learning_rate': '9.997e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '34.38', 'tokens/total': 969264, 'tokens/trainable': 14655, 'epoch': '0.125'}
+
  6%|▋         | 32/512 [03:35<53:32,  6.69s/it][2026-08-18 14:08:35,358] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-32
+
  6%|▋         | 33/512 [03:44<58:20,  7.31s/it]
                                                
{'loss': '0.01719', 'grad_norm': '0.7988', 'learning_rate': '9.995e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '37.82', 'tokens/train_per_sec_per_gpu': '33.23', 'tokens/total': 999792, 'tokens/trainable': 15087, 'epoch': '0.1289'}
+
  6%|▋         | 33/512 [03:44<58:20,  7.31s/it]
  7%|▋         | 34/512 [03:51<56:40,  7.11s/it]
                                                
{'loss': '0.02046', 'grad_norm': '0.507', 'learning_rate': '9.994e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.81', 'tokens/total': 1029840, 'tokens/trainable': 15548, 'epoch': '0.1328'}
+
  7%|▋         | 34/512 [03:51<56:40,  7.11s/it]
  7%|▋         | 35/512 [03:58<55:43,  7.01s/it]
                                                
{'loss': '0.02475', 'grad_norm': '0.8837', 'learning_rate': '9.992e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '33.69', 'tokens/total': 1060416, 'tokens/trainable': 16037, 'epoch': '0.1367'}
+
  7%|▋         | 35/512 [03:58<55:43,  7.01s/it]
  7%|▋         | 36/512 [04:04<54:45,  6.90s/it]
                                                
{'loss': '0.00774', 'grad_norm': '0.7192', 'learning_rate': '9.991e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.48', 'tokens/train_per_sec_per_gpu': '31', 'tokens/total': 1090464, 'tokens/trainable': 16475, 'epoch': '0.1406'}
+
  7%|▋         | 36/512 [04:04<54:45,  6.90s/it]
  7%|▋         | 37/512 [04:11<54:05,  6.83s/it]
                                                
{'loss': '0.02497', 'grad_norm': '0.8888', 'learning_rate': '9.989e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 1120944, 'tokens/trainable': 16919, 'epoch': '0.1445'}
+
  7%|▋         | 37/512 [04:11<54:05,  6.83s/it]
  7%|▋         | 38/512 [04:18<53:37,  6.79s/it]
                                                
{'loss': '0.02262', 'grad_norm': '1.009', 'learning_rate': '9.987e-05', 'ppl': '1.023', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '28.9', 'tokens/total': 1151360, 'tokens/trainable': 17333, 'epoch': '0.1484'}
+
  7%|▋         | 38/512 [04:18<53:37,  6.79s/it]
  8%|▊         | 39/512 [04:24<53:26,  6.78s/it]
                                                
{'loss': '0.009409', 'grad_norm': '0.5544', 'learning_rate': '9.984e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '36.09', 'tokens/total': 1181728, 'tokens/trainable': 17806, 'epoch': '0.1523'}
+
  8%|▊         | 39/512 [04:24<53:26,  6.78s/it]
  8%|▊         | 40/512 [04:31<53:11,  6.76s/it]
                                                
{'loss': '0.01569', 'grad_norm': '0.6828', 'learning_rate': '9.982e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.86', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 1212064, 'tokens/trainable': 18290, 'epoch': '0.1562'}
+
  8%|▊         | 40/512 [04:31<53:11,  6.76s/it]
  8%|▊         | 41/512 [04:38<52:56,  6.74s/it]
                                                
{'loss': '0.006071', 'grad_norm': '0.3203', 'learning_rate': '9.979e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.93', 'tokens/total': 1242592, 'tokens/trainable': 18747, 'epoch': '0.1602'}
+
  8%|▊         | 41/512 [04:38<52:56,  6.74s/it]
  8%|▊         | 42/512 [04:44<52:42,  6.73s/it]
                                                
{'loss': '0.0388', 'grad_norm': '1.351', 'learning_rate': '9.976e-05', 'ppl': '1.04', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 1272800, 'tokens/trainable': 19172, 'epoch': '0.1641'}
+
  8%|▊         | 42/512 [04:44<52:42,  6.73s/it]
  8%|▊         | 43/512 [04:51<52:45,  6.75s/it]
                                                
{'loss': '0.01924', 'grad_norm': '0.871', 'learning_rate': '9.973e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '34.04', 'memory/max_allocated (GiB)': '34.04', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.21', 'tokens/total': 1303600, 'tokens/trainable': 19622, 'epoch': '0.168'}
+
  8%|▊         | 43/512 [04:51<52:45,  6.75s/it]
  9%|▊         | 44/512 [04:58<52:31,  6.73s/it]
                                                
{'loss': '0.01226', 'grad_norm': '0.7336', 'learning_rate': '9.97e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 1333936, 'tokens/trainable': 20073, 'epoch': '0.1719'}
+
  9%|▊         | 44/512 [04:58<52:31,  6.73s/it]
  9%|▉         | 45/512 [05:05<52:20,  6.73s/it]
                                                
{'loss': '0.01287', 'grad_norm': '1.955', 'learning_rate': '9.966e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '31.46', 'tokens/total': 1364288, 'tokens/trainable': 20497, 'epoch': '0.1758'}
+
  9%|▉         | 45/512 [05:05<52:20,  6.73s/it]
  9%|▉         | 46/512 [05:11<52:10,  6.72s/it]
                                                
{'loss': '0.006207', 'grad_norm': '0.5353', 'learning_rate': '9.963e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.03', 'tokens/train_per_sec_per_gpu': '34.21', 'tokens/total': 1394368, 'tokens/trainable': 20940, 'epoch': '0.1797'}
+
  9%|▉         | 46/512 [05:11<52:10,  6.72s/it]
  9%|▉         | 47/512 [05:18<52:12,  6.74s/it]
                                                
{'loss': '0.006637', 'grad_norm': '0.6219', 'learning_rate': '9.959e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.94', 'tokens/total': 1425024, 'tokens/trainable': 21405, 'epoch': '0.1836'}
+
  9%|▉         | 47/512 [05:18<52:12,  6.74s/it]
  9%|▉         | 48/512 [05:25<51:58,  6.72s/it]
                                                
{'loss': '0.008137', 'grad_norm': '0.7588', 'learning_rate': '9.955e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.03', 'tokens/total': 1455392, 'tokens/trainable': 21828, 'epoch': '0.1875'}
+
  9%|▉         | 48/512 [05:25<51:58,  6.72s/it]
 10%|▉         | 49/512 [05:31<51:38,  6.69s/it]
                                                
{'loss': '0.0193', 'grad_norm': '0.6957', 'learning_rate': '9.951e-05', 'ppl': '1.019', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.43', 'tokens/total': 1485616, 'tokens/trainable': 22241, 'epoch': '0.1914'}
+
 10%|▉         | 49/512 [05:31<51:38,  6.69s/it]
 10%|▉         | 50/512 [05:38<51:34,  6.70s/it]
                                                
{'loss': '0.006366', 'grad_norm': '0.7796', 'learning_rate': '9.946e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 1515984, 'tokens/trainable': 22688, 'epoch': '0.1953'}
+
 10%|▉         | 50/512 [05:38<51:34,  6.70s/it]
 10%|▉         | 51/512 [05:45<51:21,  6.68s/it]
                                                
{'loss': '0.002192', 'grad_norm': '0.208', 'learning_rate': '9.942e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 1546224, 'tokens/trainable': 23136, 'epoch': '0.1992'}
+
 10%|▉         | 51/512 [05:45<51:21,  6.68s/it]
 10%|█         | 52/512 [05:51<51:19,  6.69s/it]
                                                
{'loss': '0.01217', 'grad_norm': '0.7914', 'learning_rate': '9.937e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.92', 'tokens/total': 1576528, 'tokens/trainable': 23631, 'epoch': '0.2031'}
+
 10%|█         | 52/512 [05:51<51:19,  6.69s/it]
 10%|█         | 53/512 [05:58<51:17,  6.70s/it]
                                                
{'loss': '0.00895', 'grad_norm': '0.7186', 'learning_rate': '9.932e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.05', 'tokens/total': 1606768, 'tokens/trainable': 24096, 'epoch': '0.207'}
+
 10%|█         | 53/512 [05:58<51:17,  6.70s/it]
 11%|█         | 54/512 [06:05<51:02,  6.69s/it]
                                                
{'loss': '0.01141', 'grad_norm': '0.8439', 'learning_rate': '9.927e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.77', 'tokens/total': 1636640, 'tokens/trainable': 24569, 'epoch': '0.2109'}
+
 11%|█         | 54/512 [06:05<51:02,  6.69s/it]
 11%|█         | 55/512 [06:12<51:06,  6.71s/it]
                                                
{'loss': '0.004429', 'grad_norm': '0.4795', 'learning_rate': '9.921e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 1667136, 'tokens/trainable': 25067, 'epoch': '0.2148'}
+
 11%|█         | 55/512 [06:12<51:06,  6.71s/it]
 11%|█         | 56/512 [06:18<51:01,  6.71s/it]
                                                
{'loss': '0.003403', 'grad_norm': '0.4258', 'learning_rate': '9.916e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.47', 'tokens/total': 1697472, 'tokens/trainable': 25541, 'epoch': '0.2188'}
+
 11%|█         | 56/512 [06:18<51:01,  6.71s/it]
 11%|█         | 57/512 [06:25<50:54,  6.71s/it]
                                                
{'loss': '0.01049', 'grad_norm': '0.7323', 'learning_rate': '9.91e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 1727680, 'tokens/trainable': 25965, 'epoch': '0.2227'}
+
 11%|█         | 57/512 [06:25<50:54,  6.71s/it]
 11%|█▏        | 58/512 [06:32<50:44,  6.71s/it]
                                                
{'loss': '0.004758', 'grad_norm': '2.37', 'learning_rate': '9.904e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.79', 'tokens/total': 1758192, 'tokens/trainable': 26409, 'epoch': '0.2266'}
+
 11%|█▏        | 58/512 [06:32<50:44,  6.71s/it]
 12%|█▏        | 59/512 [06:38<50:42,  6.72s/it]
                                                
{'loss': '0.002968', 'grad_norm': '1.733', 'learning_rate': '9.898e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.99', 'tokens/total': 1788816, 'tokens/trainable': 26883, 'epoch': '0.2305'}
+
 12%|█▏        | 59/512 [06:38<50:42,  6.72s/it]
 12%|█▏        | 60/512 [06:45<50:35,  6.72s/it]
                                                
{'loss': '0.005325', 'grad_norm': '4.904', 'learning_rate': '9.892e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.67', 'tokens/total': 1819056, 'tokens/trainable': 27323, 'epoch': '0.2344'}
+
 12%|█▏        | 60/512 [06:45<50:35,  6.72s/it]
 12%|█▏        | 61/512 [06:52<50:30,  6.72s/it]
                                                
{'loss': '0.0007892', 'grad_norm': '0.08419', 'learning_rate': '9.886e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '39.09', 'tokens/total': 1849664, 'tokens/trainable': 27830, 'epoch': '0.2383'}
+
 12%|█▏        | 61/512 [06:52<50:30,  6.72s/it]
 12%|█▏        | 62/512 [06:59<50:39,  6.75s/it]
                                                
{'loss': '0.01735', 'grad_norm': '0.7356', 'learning_rate': '9.879e-05', 'ppl': '1.018', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 1880272, 'tokens/trainable': 28281, 'epoch': '0.2422'}
+
 12%|█▏        | 62/512 [06:59<50:39,  6.75s/it]
 12%|█▏        | 63/512 [07:05<50:27,  6.74s/it]
                                                
{'loss': '0.01441', 'grad_norm': '0.9251', 'learning_rate': '9.872e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.16', 'tokens/total': 1910752, 'tokens/trainable': 28742, 'epoch': '0.2461'}
+
 12%|█▏        | 63/512 [07:05<50:27,  6.74s/it]
 12%|█▎        | 64/512 [07:12<50:08,  6.72s/it]
                                                
{'loss': '0.002559', 'grad_norm': '0.2563', 'learning_rate': '9.865e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.54', 'tokens/total': 1940848, 'tokens/trainable': 29183, 'epoch': '0.25'}
+
 12%|█▎        | 64/512 [07:12<50:08,  6.72s/it][2026-08-18 14:12:12,138] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-64
+
 13%|█▎        | 65/512 [07:21<54:03,  7.26s/it]
                                                
{'loss': '0.007829', 'grad_norm': '0.5092', 'learning_rate': '9.858e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.45', 'tokens/total': 1971344, 'tokens/trainable': 29648, 'epoch': '0.2539'}
+
 13%|█▎        | 65/512 [07:21<54:03,  7.26s/it]
 13%|█▎        | 66/512 [07:27<52:37,  7.08s/it]
                                                
{'loss': '0.001763', 'grad_norm': '0.5972', 'learning_rate': '9.851e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.06', 'tokens/train_per_sec_per_gpu': '33.44', 'tokens/total': 2001712, 'tokens/trainable': 30075, 'epoch': '0.2578'}
+
 13%|█▎        | 66/512 [07:27<52:37,  7.08s/it]
 13%|█▎        | 67/512 [07:34<50:38,  6.83s/it]
                                                
{'loss': '0.005683', 'grad_norm': '0.6683', 'learning_rate': '9.844e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '37.53', 'tokens/total': 2029936, 'tokens/trainable': 30546, 'epoch': '0.2617'}
+
 13%|█▎        | 67/512 [07:34<50:38,  6.83s/it]
 13%|█▎        | 68/512 [07:40<50:18,  6.80s/it]
                                                
{'loss': '0.01054', 'grad_norm': '2.3', 'learning_rate': '9.836e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 2060384, 'tokens/trainable': 31025, 'epoch': '0.2656'}
+
 13%|█▎        | 68/512 [07:40<50:18,  6.80s/it]
 13%|█▎        | 69/512 [07:47<50:03,  6.78s/it]
                                                
{'loss': '0.0167', 'grad_norm': '2.232', 'learning_rate': '9.828e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 2090688, 'tokens/trainable': 31505, 'epoch': '0.2695'}
+
 13%|█▎        | 69/512 [07:47<50:03,  6.78s/it]
 14%|█▎        | 70/512 [07:54<49:42,  6.75s/it]
                                                
{'loss': '0.005516', 'grad_norm': '0.455', 'learning_rate': '9.82e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 2120848, 'tokens/trainable': 31956, 'epoch': '0.2734'}
+
 14%|█▎        | 70/512 [07:54<49:42,  6.75s/it]
 14%|█▍        | 71/512 [08:00<48:21,  6.58s/it]
                                                
{'loss': '0.008853', 'grad_norm': '0.7483', 'learning_rate': '9.812e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.35', 'memory/max_allocated (GiB)': '33.35', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 2149120, 'tokens/trainable': 32374, 'epoch': '0.2773'}
+
 14%|█▍        | 71/512 [08:00<48:21,  6.58s/it]
 14%|█▍        | 72/512 [08:07<48:31,  6.62s/it]
                                                
{'loss': '0.001421', 'grad_norm': '0.4359', 'learning_rate': '9.803e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.64', 'tokens/train_per_sec_per_gpu': '34.46', 'tokens/total': 2179472, 'tokens/trainable': 32833, 'epoch': '0.2812'}
+
 14%|█▍        | 72/512 [08:07<48:31,  6.62s/it]
 14%|█▍        | 73/512 [08:13<48:32,  6.63s/it]
                                                
{'loss': '0.005964', 'grad_norm': '0.3272', 'learning_rate': '9.795e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.67', 'tokens/total': 2209984, 'tokens/trainable': 33304, 'epoch': '0.2852'}
+
 14%|█▍        | 73/512 [08:13<48:32,  6.63s/it]
 14%|█▍        | 74/512 [08:20<48:48,  6.69s/it]
                                                
{'loss': '0.005361', 'grad_norm': '0.5767', 'learning_rate': '9.786e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.33', 'tokens/total': 2240512, 'tokens/trainable': 33771, 'epoch': '0.2891'}
+
 14%|█▍        | 74/512 [08:20<48:48,  6.69s/it]
 15%|█▍        | 75/512 [08:27<48:40,  6.68s/it]
                                                
{'loss': '0.006908', 'grad_norm': '0.9308', 'learning_rate': '9.777e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.46', 'tokens/total': 2270608, 'tokens/trainable': 34186, 'epoch': '0.293'}
+
 15%|█▍        | 75/512 [08:27<48:40,  6.68s/it]
 15%|█▍        | 76/512 [08:33<48:42,  6.70s/it]
                                                
{'loss': '0.001703', 'grad_norm': '0.5172', 'learning_rate': '9.768e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.83', 'tokens/total': 2300864, 'tokens/trainable': 34667, 'epoch': '0.2969'}
+
 15%|█▍        | 76/512 [08:33<48:42,  6.70s/it]
 15%|█▌        | 77/512 [08:40<48:29,  6.69s/it]
                                                
{'loss': '0.01024', 'grad_norm': '4.432', 'learning_rate': '9.759e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 2330960, 'tokens/trainable': 35078, 'epoch': '0.3008'}
+
 15%|█▌        | 77/512 [08:40<48:29,  6.69s/it]
 15%|█▌        | 78/512 [08:47<48:35,  6.72s/it]
                                                
{'loss': '0.01689', 'grad_norm': '1.699', 'learning_rate': '9.749e-05', 'ppl': '1.017', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.8', 'tokens/total': 2361584, 'tokens/trainable': 35550, 'epoch': '0.3047'}
+
 15%|█▌        | 78/512 [08:47<48:35,  6.72s/it]
 15%|█▌        | 79/512 [08:54<48:24,  6.71s/it]
                                                
{'loss': '0.005958', 'grad_norm': '1.176', 'learning_rate': '9.74e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '29.46', 'tokens/total': 2391936, 'tokens/trainable': 35977, 'epoch': '0.3086'}
+
 15%|█▌        | 79/512 [08:54<48:24,  6.71s/it]
 16%|█▌        | 80/512 [09:00<48:20,  6.72s/it]
                                                
{'loss': '0.03401', 'grad_norm': '2.761', 'learning_rate': '9.73e-05', 'ppl': '1.035', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.2', 'tokens/total': 2422432, 'tokens/trainable': 36462, 'epoch': '0.3125'}
+
 16%|█▌        | 80/512 [09:00<48:20,  6.72s/it]
 16%|█▌        | 81/512 [09:07<48:11,  6.71s/it]
                                                
{'loss': '0.0001726', 'grad_norm': '0.0173', 'learning_rate': '9.72e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 2452752, 'tokens/trainable': 36919, 'epoch': '0.3164'}
+
 16%|█▌        | 81/512 [09:07<48:11,  6.71s/it]
 16%|█▌        | 82/512 [09:13<47:02,  6.56s/it]
                                                
{'loss': '0.008328', 'grad_norm': '0.6305', 'learning_rate': '9.71e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.44', 'memory/max_allocated (GiB)': '33.44', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.4', 'tokens/total': 2481216, 'tokens/trainable': 37379, 'epoch': '0.3203'}
+
 16%|█▌        | 82/512 [09:13<47:02,  6.56s/it]
 16%|█▌        | 83/512 [09:20<47:16,  6.61s/it]
                                                
{'loss': '0.0007961', 'grad_norm': '0.08372', 'learning_rate': '9.699e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.05', 'tokens/total': 2511488, 'tokens/trainable': 37832, 'epoch': '0.3242'}
+
 16%|█▌        | 83/512 [09:20<47:16,  6.61s/it]
 16%|█▋        | 84/512 [09:27<47:25,  6.65s/it]
                                                
{'loss': '0.001538', 'grad_norm': '0.4056', 'learning_rate': '9.689e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 2541984, 'tokens/trainable': 38303, 'epoch': '0.3281'}
+
 16%|█▋        | 84/512 [09:27<47:25,  6.65s/it]
 17%|█▋        | 85/512 [09:33<47:24,  6.66s/it]
                                                
{'loss': '0.0007728', 'grad_norm': '0.1657', 'learning_rate': '9.678e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '33.62', 'tokens/total': 2572224, 'tokens/trainable': 38758, 'epoch': '0.332'}
+
 17%|█▋        | 85/512 [09:33<47:24,  6.66s/it]
 17%|█▋        | 86/512 [09:40<47:22,  6.67s/it]
                                                
{'loss': '0.008873', 'grad_norm': '1.778', 'learning_rate': '9.667e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.62', 'tokens/total': 2602704, 'tokens/trainable': 39179, 'epoch': '0.3359'}
+
 17%|█▋        | 86/512 [09:40<47:22,  6.67s/it]
 17%|█▋        | 87/512 [09:47<47:21,  6.69s/it]
                                                
{'loss': '0.01371', 'grad_norm': '0.6024', 'learning_rate': '9.656e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '35.29', 'tokens/total': 2633072, 'tokens/trainable': 39679, 'epoch': '0.3398'}
+
 17%|█▋        | 87/512 [09:47<47:21,  6.69s/it]
 17%|█▋        | 88/512 [09:53<47:06,  6.67s/it]
                                                
{'loss': '0.02105', 'grad_norm': '3.774', 'learning_rate': '9.645e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 2663232, 'tokens/trainable': 40118, 'epoch': '0.3438'}
+
 17%|█▋        | 88/512 [09:53<47:06,  6.67s/it]
 17%|█▋        | 89/512 [10:00<47:02,  6.67s/it]
                                                
{'loss': '0.002541', 'grad_norm': '0.209', 'learning_rate': '9.633e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 2693696, 'tokens/trainable': 40571, 'epoch': '0.3477'}
+
 17%|█▋        | 89/512 [10:00<47:02,  6.67s/it]
 18%|█▊        | 90/512 [10:07<46:59,  6.68s/it]
                                                
{'loss': '0.003635', 'grad_norm': '0.3318', 'learning_rate': '9.622e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '30.75', 'tokens/total': 2723984, 'tokens/trainable': 40997, 'epoch': '0.3516'}
+
 18%|█▊        | 90/512 [10:07<46:59,  6.68s/it]
 18%|█▊        | 91/512 [10:14<46:57,  6.69s/it]
                                                
{'loss': '0.007066', 'grad_norm': '0.4436', 'learning_rate': '9.61e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 2754240, 'tokens/trainable': 41462, 'epoch': '0.3555'}
+
 18%|█▊        | 91/512 [10:14<46:57,  6.69s/it]
 18%|█▊        | 92/512 [10:20<46:50,  6.69s/it]
                                                
{'loss': '0.01106', 'grad_norm': '0.8875', 'learning_rate': '9.598e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.79', 'tokens/total': 2784672, 'tokens/trainable': 41927, 'epoch': '0.3594'}
+
 18%|█▊        | 92/512 [10:20<46:50,  6.69s/it]
 18%|█▊        | 93/512 [10:27<46:47,  6.70s/it]
                                                
{'loss': '0.02101', 'grad_norm': '1.71', 'learning_rate': '9.586e-05', 'ppl': '1.021', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '34.84', 'tokens/total': 2815120, 'tokens/trainable': 42397, 'epoch': '0.3633'}
+
 18%|█▊        | 93/512 [10:27<46:47,  6.70s/it]
 18%|█▊        | 94/512 [10:34<46:44,  6.71s/it]
                                                
{'loss': '0.0007664', 'grad_norm': '0.05346', 'learning_rate': '9.574e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 2845600, 'tokens/trainable': 42883, 'epoch': '0.3672'}
+
 18%|█▊        | 94/512 [10:34<46:44,  6.71s/it]
 19%|█▊        | 95/512 [10:40<46:43,  6.72s/it]
                                                
{'loss': '0.001947', 'grad_norm': '0.3566', 'learning_rate': '9.562e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '38.67', 'tokens/total': 2875952, 'tokens/trainable': 43381, 'epoch': '0.3711'}
+
 19%|█▊        | 95/512 [10:40<46:43,  6.72s/it]
 19%|█▉        | 96/512 [10:47<46:37,  6.72s/it]
                                                
{'loss': '0.01163', 'grad_norm': '0.6131', 'learning_rate': '9.549e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 2906288, 'tokens/trainable': 43824, 'epoch': '0.375'}
+
 19%|█▉        | 96/512 [10:47<46:37,  6.72s/it][2026-08-18 14:15:47,190] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-96
+
 19%|█▉        | 97/512 [10:56<50:07,  7.25s/it]
                                                
{'loss': '0.004611', 'grad_norm': '0.5786', 'learning_rate': '9.536e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.98', 'tokens/train_per_sec_per_gpu': '31.51', 'tokens/total': 2936752, 'tokens/trainable': 44255, 'epoch': '0.3789'}
+
 19%|█▉        | 97/512 [10:56<50:07,  7.25s/it]
 19%|█▉        | 98/512 [11:02<48:52,  7.08s/it]
                                                
{'loss': '0.003711', 'grad_norm': '0.4322', 'learning_rate': '9.523e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.19', 'tokens/train_per_sec_per_gpu': '37.74', 'tokens/total': 2967072, 'tokens/trainable': 44739, 'epoch': '0.3828'}
+
 19%|█▉        | 98/512 [11:02<48:52,  7.08s/it]
 19%|█▉        | 99/512 [11:09<48:10,  7.00s/it]
                                                
{'loss': '0.002609', 'grad_norm': '0.1657', 'learning_rate': '9.51e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '35.76', 'tokens/train_per_sec_per_gpu': '36.2', 'tokens/total': 2997664, 'tokens/trainable': 45202, 'epoch': '0.3867'}
+
 19%|█▉        | 99/512 [11:09<48:10,  7.00s/it]
 20%|█▉        | 100/512 [11:16<47:24,  6.91s/it]
                                                 
{'loss': '0.005652', 'grad_norm': '0.4386', 'learning_rate': '9.497e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.05', 'tokens/total': 3028032, 'tokens/trainable': 45646, 'epoch': '0.3906'}
+
 20%|█▉        | 100/512 [11:16<47:24,  6.91s/it]
 20%|█▉        | 101/512 [11:23<46:54,  6.85s/it]
                                                 
{'loss': '0.001526', 'grad_norm': '0.1223', 'learning_rate': '9.484e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 3058272, 'tokens/trainable': 46089, 'epoch': '0.3945'}
+
 20%|█▉        | 101/512 [11:23<46:54,  6.85s/it]
 20%|█▉        | 102/512 [11:29<46:33,  6.81s/it]
                                                 
{'loss': '0.007584', 'grad_norm': '0.7791', 'learning_rate': '9.47e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '32.55', 'tokens/total': 3088848, 'tokens/trainable': 46547, 'epoch': '0.3984'}
+
 20%|█▉        | 102/512 [11:29<46:33,  6.81s/it]
 20%|██        | 103/512 [11:36<46:12,  6.78s/it]
                                                 
{'loss': '0.002193', 'grad_norm': '0.1951', 'learning_rate': '9.456e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.99', 'tokens/train_per_sec_per_gpu': '31.96', 'tokens/total': 3119424, 'tokens/trainable': 47014, 'epoch': '0.4023'}
+
 20%|██        | 103/512 [11:36<46:12,  6.78s/it]
 20%|██        | 104/512 [11:43<45:57,  6.76s/it]
                                                 
{'loss': '0.007656', 'grad_norm': '0.3494', 'learning_rate': '9.442e-05', 'ppl': '1.008', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 3149696, 'tokens/trainable': 47503, 'epoch': '0.4062'}
+
 20%|██        | 104/512 [11:43<45:57,  6.76s/it]
 21%|██        | 105/512 [11:49<45:46,  6.75s/it]
                                                 
{'loss': '0.001468', 'grad_norm': '0.3206', 'learning_rate': '9.428e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.27', 'tokens/total': 3180064, 'tokens/trainable': 47996, 'epoch': '0.4102'}
+
 21%|██        | 105/512 [11:49<45:46,  6.75s/it]
 21%|██        | 106/512 [11:56<44:32,  6.58s/it]
                                                 
{'loss': '0.00197', 'grad_norm': '0.232', 'learning_rate': '9.414e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 3208320, 'tokens/trainable': 48447, 'epoch': '0.4141'}
+
 21%|██        | 106/512 [11:56<44:32,  6.58s/it]
 21%|██        | 107/512 [12:02<44:38,  6.61s/it]
                                                 
{'loss': '0.000285', 'grad_norm': '0.05074', 'learning_rate': '9.4e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.69', 'tokens/total': 3238416, 'tokens/trainable': 48889, 'epoch': '0.418'}
+
 21%|██        | 107/512 [12:02<44:38,  6.61s/it]
 21%|██        | 108/512 [12:09<44:42,  6.64s/it]
                                                 
{'loss': '0.002674', 'grad_norm': '0.6825', 'learning_rate': '9.385e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '37.89', 'tokens/total': 3268784, 'tokens/trainable': 49360, 'epoch': '0.4219'}
+
 21%|██        | 108/512 [12:09<44:42,  6.64s/it]
 21%|██▏       | 109/512 [12:16<44:45,  6.66s/it]
                                                 
{'loss': '0.001428', 'grad_norm': '0.2664', 'learning_rate': '9.37e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.78', 'tokens/total': 3298912, 'tokens/trainable': 49827, 'epoch': '0.4258'}
+
 21%|██▏       | 109/512 [12:16<44:45,  6.66s/it]
 21%|██▏       | 110/512 [12:22<44:43,  6.68s/it]
                                                 
{'loss': '0.0007402', 'grad_norm': '0.1222', 'learning_rate': '9.355e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.31', 'tokens/total': 3329232, 'tokens/trainable': 50267, 'epoch': '0.4297'}
+
 21%|██▏       | 110/512 [12:22<44:43,  6.68s/it]
 22%|██▏       | 111/512 [12:29<44:40,  6.69s/it]
                                                 
{'loss': '0.02026', 'grad_norm': '1.222', 'learning_rate': '9.34e-05', 'ppl': '1.02', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.41', 'tokens/total': 3359440, 'tokens/trainable': 50711, 'epoch': '0.4336'}
+
 22%|██▏       | 111/512 [12:29<44:40,  6.69s/it]
 22%|██▏       | 112/512 [12:36<44:33,  6.68s/it]
                                                 
{'loss': '0.01277', 'grad_norm': '1.257', 'learning_rate': '9.325e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.59', 'tokens/total': 3389472, 'tokens/trainable': 51157, 'epoch': '0.4375'}
+
 22%|██▏       | 112/512 [12:36<44:33,  6.68s/it]
 22%|██▏       | 113/512 [12:43<44:32,  6.70s/it]
                                                 
{'loss': '0.00542', 'grad_norm': '0.8144', 'learning_rate': '9.31e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 3419920, 'tokens/trainable': 51613, 'epoch': '0.4414'}
+
 22%|██▏       | 113/512 [12:43<44:32,  6.70s/it]
 22%|██▏       | 114/512 [12:49<44:22,  6.69s/it]
                                                 
{'loss': '0.004596', 'grad_norm': '1.351', 'learning_rate': '9.294e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '29.1', 'tokens/total': 3450176, 'tokens/trainable': 52017, 'epoch': '0.4453'}
+
 22%|██▏       | 114/512 [12:49<44:22,  6.69s/it]
 22%|██▏       | 115/512 [12:56<44:18,  6.70s/it]
                                                 
{'loss': '0.005163', 'grad_norm': '2.766', 'learning_rate': '9.278e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 3480544, 'tokens/trainable': 52506, 'epoch': '0.4492'}
+
 22%|██▏       | 115/512 [12:56<44:18,  6.70s/it]
 23%|██▎       | 116/512 [13:03<44:15,  6.71s/it]
                                                 
{'loss': '0.02482', 'grad_norm': '1.783', 'learning_rate': '9.263e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '38.63', 'tokens/total': 3510848, 'tokens/trainable': 53008, 'epoch': '0.4531'}
+
 23%|██▎       | 116/512 [13:03<44:15,  6.71s/it]
 23%|██▎       | 117/512 [13:09<44:08,  6.71s/it]
                                                 
{'loss': '0.01225', 'grad_norm': '0.868', 'learning_rate': '9.247e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.36', 'tokens/total': 3541296, 'tokens/trainable': 53467, 'epoch': '0.457'}
+
 23%|██▎       | 117/512 [13:09<44:08,  6.71s/it]
 23%|██▎       | 118/512 [13:16<44:06,  6.72s/it]
                                                 
{'loss': '0.01154', 'grad_norm': '0.8031', 'learning_rate': '9.23e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 3571936, 'tokens/trainable': 53923, 'epoch': '0.4609'}
+
 23%|██▎       | 118/512 [13:16<44:06,  6.72s/it]
 23%|██▎       | 119/512 [13:23<43:53,  6.70s/it]
                                                 
{'loss': '0.004753', 'grad_norm': '1.132', 'learning_rate': '9.214e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 3602000, 'tokens/trainable': 54363, 'epoch': '0.4648'}
+
 23%|██▎       | 119/512 [13:23<43:53,  6.70s/it]
 23%|██▎       | 120/512 [13:29<43:47,  6.70s/it]
                                                 
{'loss': '0.006918', 'grad_norm': '0.2578', 'learning_rate': '9.198e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 3632288, 'tokens/trainable': 54839, 'epoch': '0.4688'}
+
 23%|██▎       | 120/512 [13:29<43:47,  6.70s/it]
 24%|██▎       | 121/512 [13:36<43:39,  6.70s/it]
                                                 
{'loss': '0.003779', 'grad_norm': '0.3194', 'learning_rate': '9.181e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.59', 'tokens/total': 3662608, 'tokens/trainable': 55267, 'epoch': '0.4727'}
+
 24%|██▎       | 121/512 [13:36<43:39,  6.70s/it]
 24%|██▍       | 122/512 [13:43<43:31,  6.70s/it]
                                                 
{'loss': '0.004217', 'grad_norm': '0.3439', 'learning_rate': '9.164e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '30.52', 'tokens/total': 3692864, 'tokens/trainable': 55700, 'epoch': '0.4766'}
+
 24%|██▍       | 122/512 [13:43<43:31,  6.70s/it]
 24%|██▍       | 123/512 [13:49<43:19,  6.68s/it]
                                                 
{'loss': '0.003874', 'grad_norm': '0.4916', 'learning_rate': '9.147e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.39', 'tokens/total': 3722864, 'tokens/trainable': 56144, 'epoch': '0.4805'}
+
 24%|██▍       | 123/512 [13:49<43:19,  6.68s/it]
 24%|██▍       | 124/512 [13:56<43:11,  6.68s/it]
                                                 
{'loss': '0.00401', 'grad_norm': '0.3992', 'learning_rate': '9.13e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 3753056, 'tokens/trainable': 56589, 'epoch': '0.4844'}
+
 24%|██▍       | 124/512 [13:56<43:11,  6.68s/it]
 24%|██▍       | 125/512 [14:03<43:11,  6.70s/it]
                                                 
{'loss': '0.005606', 'grad_norm': '0.4556', 'learning_rate': '9.113e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '36.68', 'tokens/total': 3783312, 'tokens/trainable': 57081, 'epoch': '0.4883'}
+
 24%|██▍       | 125/512 [14:03<43:11,  6.70s/it]
 25%|██▍       | 126/512 [14:10<43:01,  6.69s/it]
                                                 
{'loss': '0.0005784', 'grad_norm': '0.04188', 'learning_rate': '9.096e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 3813712, 'tokens/trainable': 57527, 'epoch': '0.4922'}
+
 25%|██▍       | 126/512 [14:10<43:01,  6.69s/it]
 25%|██▍       | 127/512 [14:16<43:01,  6.71s/it]
                                                 
{'loss': '0.002429', 'grad_norm': '0.2182', 'learning_rate': '9.078e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '34.84', 'tokens/total': 3844336, 'tokens/trainable': 58005, 'epoch': '0.4961'}
+
 25%|██▍       | 127/512 [14:16<43:01,  6.71s/it]
 25%|██▌       | 128/512 [14:23<41:55,  6.55s/it]
                                                 
{'loss': '0.001604', 'grad_norm': '0.4637', 'learning_rate': '9.06e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '35.21', 'tokens/total': 3872688, 'tokens/trainable': 58454, 'epoch': '0.5'}
+
 25%|██▌       | 128/512 [14:23<41:55,  6.55s/it][2026-08-18 14:19:22,518] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-128
+
 25%|██▌       | 129/512 [14:31<45:18,  7.10s/it]
                                                 
{'loss': '0.0002776', 'grad_norm': '0.03312', 'learning_rate': '9.043e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36', 'tokens/train_per_sec_per_gpu': '31.97', 'tokens/total': 3902640, 'tokens/trainable': 58896, 'epoch': '0.5039'}
+
 25%|██▌       | 129/512 [14:31<45:18,  7.10s/it]
 25%|██▌       | 130/512 [14:38<44:26,  6.98s/it]
                                                 
{'loss': '8.183e-05', 'grad_norm': '0.0194', 'learning_rate': '9.025e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '34.68', 'tokens/train_per_sec_per_gpu': '35.8', 'tokens/total': 3933104, 'tokens/trainable': 59356, 'epoch': '0.5078'}
+
 25%|██▌       | 130/512 [14:38<44:26,  6.98s/it]
 26%|██▌       | 131/512 [14:44<43:48,  6.90s/it]
                                                 
{'loss': '0.005079', 'grad_norm': '0.3952', 'learning_rate': '9.007e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '34.92', 'tokens/train_per_sec_per_gpu': '32.21', 'tokens/total': 3963184, 'tokens/trainable': 59792, 'epoch': '0.5117'}
+
 26%|██▌       | 131/512 [14:44<43:48,  6.90s/it]
 26%|██▌       | 132/512 [14:51<43:18,  6.84s/it]
                                                 
{'loss': '0.0001526', 'grad_norm': '0.03258', 'learning_rate': '8.988e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.68', 'tokens/total': 3993648, 'tokens/trainable': 60266, 'epoch': '0.5156'}
+
 26%|██▌       | 132/512 [14:51<43:18,  6.84s/it]
 26%|██▌       | 133/512 [14:58<42:58,  6.80s/it]
                                                 
{'loss': '0.0003361', 'grad_norm': '0.1087', 'learning_rate': '8.97e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.44', 'tokens/total': 4024000, 'tokens/trainable': 60759, 'epoch': '0.5195'}
+
 26%|██▌       | 133/512 [14:58<42:58,  6.80s/it]
 26%|██▌       | 134/512 [15:04<41:43,  6.62s/it]
                                                 
{'loss': '0.0001435', 'grad_norm': '0.01607', 'learning_rate': '8.951e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.39', 'memory/max_allocated (GiB)': '33.39', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '31.76', 'tokens/total': 4052480, 'tokens/trainable': 61204, 'epoch': '0.5234'}
+
 26%|██▌       | 134/512 [15:04<41:43,  6.62s/it]
 26%|██▋       | 135/512 [15:11<41:49,  6.66s/it]
                                                 
{'loss': '0.003308', 'grad_norm': '0.5635', 'learning_rate': '8.933e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '35.93', 'tokens/total': 4082864, 'tokens/trainable': 61684, 'epoch': '0.5273'}
+
 26%|██▋       | 135/512 [15:11<41:49,  6.66s/it]
 27%|██▋       | 136/512 [15:17<41:41,  6.65s/it]
                                                 
{'loss': '0.0003551', 'grad_norm': '0.1293', 'learning_rate': '8.914e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.7', 'memory/max_allocated (GiB)': '33.7', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '38.22', 'tokens/total': 4113040, 'tokens/trainable': 62159, 'epoch': '0.5312'}
+
 27%|██▋       | 136/512 [15:17<41:41,  6.65s/it]
 27%|██▋       | 137/512 [15:24<41:42,  6.67s/it]
                                                 
{'loss': '0.0002933', 'grad_norm': '0.03812', 'learning_rate': '8.895e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.28', 'tokens/total': 4143344, 'tokens/trainable': 62649, 'epoch': '0.5352'}
+
 27%|██▋       | 137/512 [15:24<41:42,  6.67s/it]
 27%|██▋       | 138/512 [15:31<41:37,  6.68s/it]
                                                 
{'loss': '0.009558', 'grad_norm': '0.4708', 'learning_rate': '8.876e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.65', 'tokens/total': 4173648, 'tokens/trainable': 63100, 'epoch': '0.5391'}
+
 27%|██▋       | 138/512 [15:31<41:37,  6.68s/it]
 27%|██▋       | 139/512 [15:37<41:37,  6.70s/it]
                                                 
{'loss': '0.006467', 'grad_norm': '0.6402', 'learning_rate': '8.856e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.56', 'tokens/total': 4204080, 'tokens/trainable': 63588, 'epoch': '0.543'}
+
 27%|██▋       | 139/512 [15:37<41:37,  6.70s/it]
 27%|██▋       | 140/512 [15:44<40:29,  6.53s/it]
                                                 
{'loss': '0.01086', 'grad_norm': '0.9977', 'learning_rate': '8.837e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.42', 'memory/max_allocated (GiB)': '33.42', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 4232400, 'tokens/trainable': 64005, 'epoch': '0.5469'}
+
 27%|██▋       | 140/512 [15:44<40:29,  6.53s/it]
 28%|██▊       | 141/512 [15:50<40:43,  6.59s/it]
                                                 
{'loss': '0.002833', 'grad_norm': '0.931', 'learning_rate': '8.817e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.12', 'tokens/total': 4262624, 'tokens/trainable': 64462, 'epoch': '0.5508'}
+
 28%|██▊       | 141/512 [15:50<40:43,  6.59s/it]
 28%|██▊       | 142/512 [15:57<40:52,  6.63s/it]
                                                 
{'loss': '0.00226', 'grad_norm': '0.8449', 'learning_rate': '8.798e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.93', 'tokens/total': 4293168, 'tokens/trainable': 64900, 'epoch': '0.5547'}
+
 28%|██▊       | 142/512 [15:57<40:52,  6.63s/it]
 28%|██▊       | 143/512 [16:04<40:52,  6.65s/it]
                                                 
{'loss': '0.001097', 'grad_norm': '0.1501', 'learning_rate': '8.778e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '37.52', 'tokens/total': 4323248, 'tokens/trainable': 65379, 'epoch': '0.5586'}
+
 28%|██▊       | 143/512 [16:04<40:52,  6.65s/it]
 28%|██▊       | 144/512 [16:10<40:49,  6.66s/it]
                                                 
{'loss': '0.02481', 'grad_norm': '0.7081', 'learning_rate': '8.758e-05', 'ppl': '1.025', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 4353344, 'tokens/trainable': 65845, 'epoch': '0.5625'}
+
 28%|██▊       | 144/512 [16:10<40:49,  6.66s/it]
 28%|██▊       | 145/512 [16:17<40:46,  6.67s/it]
                                                 
{'loss': '0.01163', 'grad_norm': '0.8303', 'learning_rate': '8.738e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 4383584, 'tokens/trainable': 66285, 'epoch': '0.5664'}
+
 28%|██▊       | 145/512 [16:17<40:46,  6.67s/it]
 29%|██▊       | 146/512 [16:24<40:46,  6.68s/it]
                                                 
{'loss': '0.006236', 'grad_norm': '0.4801', 'learning_rate': '8.718e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 4413856, 'tokens/trainable': 66736, 'epoch': '0.5703'}
+
 29%|██▊       | 146/512 [16:24<40:46,  6.68s/it]
 29%|██▊       | 147/512 [16:31<40:40,  6.69s/it]
                                                 
{'loss': '0.004551', 'grad_norm': '0.2672', 'learning_rate': '8.697e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.26', 'tokens/total': 4444112, 'tokens/trainable': 67150, 'epoch': '0.5742'}
+
 29%|██▊       | 147/512 [16:31<40:40,  6.69s/it]
 29%|██▉       | 148/512 [16:37<40:41,  6.71s/it]
                                                 
{'loss': '0.005152', 'grad_norm': '0.2334', 'learning_rate': '8.677e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 4474464, 'tokens/trainable': 67633, 'epoch': '0.5781'}
+
 29%|██▉       | 148/512 [16:37<40:41,  6.71s/it]
 29%|██▉       | 149/512 [16:44<40:35,  6.71s/it]
                                                 
{'loss': '0.01038', 'grad_norm': '1.36', 'learning_rate': '8.656e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.82', 'tokens/total': 4504880, 'tokens/trainable': 68077, 'epoch': '0.582'}
+
 29%|██▉       | 149/512 [16:44<40:35,  6.71s/it]
 29%|██▉       | 150/512 [16:51<40:27,  6.70s/it]
                                                 
{'loss': '0.001625', 'grad_norm': '0.2275', 'learning_rate': '8.635e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.79', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 4534976, 'tokens/trainable': 68545, 'epoch': '0.5859'}
+
 29%|██▉       | 150/512 [16:51<40:27,  6.70s/it]
 29%|██▉       | 151/512 [16:57<40:21,  6.71s/it]
                                                 
{'loss': '0.02714', 'grad_norm': '0.6304', 'learning_rate': '8.615e-05', 'ppl': '1.028', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.26', 'tokens/total': 4565088, 'tokens/trainable': 69011, 'epoch': '0.5898'}
+
 29%|██▉       | 151/512 [16:57<40:21,  6.71s/it]
 30%|██▉       | 152/512 [17:04<40:17,  6.71s/it]
                                                 
{'loss': '0.006482', 'grad_norm': '0.3306', 'learning_rate': '8.594e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 4595600, 'tokens/trainable': 69465, 'epoch': '0.5938'}
+
 30%|██▉       | 152/512 [17:04<40:17,  6.71s/it]
 30%|██▉       | 153/512 [17:11<40:15,  6.73s/it]
                                                 
{'loss': '0.01488', 'grad_norm': '2.509', 'learning_rate': '8.572e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.89', 'tokens/total': 4626032, 'tokens/trainable': 69928, 'epoch': '0.5977'}
+
 30%|██▉       | 153/512 [17:11<40:15,  6.73s/it]
 30%|███       | 154/512 [17:18<40:06,  6.72s/it]
                                                 
{'loss': '0.001413', 'grad_norm': '0.08187', 'learning_rate': '8.551e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.67', 'tokens/total': 4656352, 'tokens/trainable': 70395, 'epoch': '0.6016'}
+
 30%|███       | 154/512 [17:18<40:06,  6.72s/it]
 30%|███       | 155/512 [17:24<39:02,  6.56s/it]
                                                 
{'loss': '0.001624', 'grad_norm': '0.203', 'learning_rate': '8.53e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.58', 'tokens/total': 4684672, 'tokens/trainable': 70829, 'epoch': '0.6055'}
+
 30%|███       | 155/512 [17:24<39:02,  6.56s/it]
 30%|███       | 156/512 [17:31<39:13,  6.61s/it]
                                                 
{'loss': '0.001468', 'grad_norm': '0.1048', 'learning_rate': '8.508e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.88', 'tokens/total': 4715216, 'tokens/trainable': 71287, 'epoch': '0.6094'}
+
 30%|███       | 156/512 [17:31<39:13,  6.61s/it]
 31%|███       | 157/512 [17:37<39:19,  6.65s/it]
                                                 
{'loss': '0.004101', 'grad_norm': '0.2741', 'learning_rate': '8.487e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.95', 'tokens/total': 4745376, 'tokens/trainable': 71749, 'epoch': '0.6133'}
+
 31%|███       | 157/512 [17:37<39:19,  6.65s/it]
 31%|███       | 158/512 [17:44<39:23,  6.68s/it]
                                                 
{'loss': '0.004408', 'grad_norm': '0.3079', 'learning_rate': '8.465e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.18', 'tokens/total': 4775776, 'tokens/trainable': 72185, 'epoch': '0.6172'}
+
 31%|███       | 158/512 [17:44<39:23,  6.68s/it]
 31%|███       | 159/512 [17:51<39:13,  6.67s/it]
                                                 
{'loss': '0.003525', 'grad_norm': '0.3344', 'learning_rate': '8.443e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 4805856, 'tokens/trainable': 72604, 'epoch': '0.6211'}
+
 31%|███       | 159/512 [17:51<39:13,  6.67s/it]
 31%|███▏      | 160/512 [17:57<39:12,  6.68s/it]
                                                 
{'loss': '0.01557', 'grad_norm': '0.6212', 'learning_rate': '8.421e-05', 'ppl': '1.016', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '31.56', 'tokens/total': 4836304, 'tokens/trainable': 73061, 'epoch': '0.625'}
+
 31%|███▏      | 160/512 [17:57<39:12,  6.68s/it][2026-08-18 14:22:57,346] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-160
+
 31%|███▏      | 161/512 [18:06<41:56,  7.17s/it]
                                                 
{'loss': '0.005395', 'grad_norm': '0.5695', 'learning_rate': '8.399e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.98', 'tokens/total': 4866544, 'tokens/trainable': 73474, 'epoch': '0.6289'}
+
 31%|███▏      | 161/512 [18:06<41:56,  7.17s/it]
 32%|███▏      | 162/512 [18:12<41:03,  7.04s/it]
                                                 
{'loss': '0.004907', 'grad_norm': '0.3221', 'learning_rate': '8.376e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.65', 'tokens/total': 4897120, 'tokens/trainable': 73923, 'epoch': '0.6328'}
+
 32%|███▏      | 162/512 [18:12<41:03,  7.04s/it]
 32%|███▏      | 163/512 [18:19<40:21,  6.94s/it]
                                                 
{'loss': '0.01177', 'grad_norm': '0.2563', 'learning_rate': '8.354e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 4927440, 'tokens/trainable': 74379, 'epoch': '0.6367'}
+
 32%|███▏      | 163/512 [18:19<40:21,  6.94s/it]
 32%|███▏      | 164/512 [18:26<40:05,  6.91s/it]
                                                 
{'loss': '0.01483', 'grad_norm': '0.3738', 'learning_rate': '8.332e-05', 'ppl': '1.015', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.45', 'tokens/total': 4958032, 'tokens/trainable': 74830, 'epoch': '0.6406'}
+
 32%|███▏      | 164/512 [18:26<40:05,  6.91s/it]
 32%|███▏      | 165/512 [18:33<39:34,  6.84s/it]
                                                 
{'loss': '0.00068', 'grad_norm': '0.0768', 'learning_rate': '8.309e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.73', 'tokens/total': 4988368, 'tokens/trainable': 75262, 'epoch': '0.6445'}
+
 32%|███▏      | 165/512 [18:33<39:34,  6.84s/it]
 32%|███▏      | 166/512 [18:39<39:08,  6.79s/it]
                                                 
{'loss': '0.0005167', 'grad_norm': '0.1594', 'learning_rate': '8.286e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.84', 'tokens/total': 5018672, 'tokens/trainable': 75703, 'epoch': '0.6484'}
+
 32%|███▏      | 166/512 [18:39<39:08,  6.79s/it]
 33%|███▎      | 167/512 [18:46<39:03,  6.79s/it]
                                                 
{'loss': '0.003161', 'grad_norm': '0.1887', 'learning_rate': '8.263e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.3', 'tokens/total': 5049344, 'tokens/trainable': 76164, 'epoch': '0.6523'}
+
 33%|███▎      | 167/512 [18:46<39:03,  6.79s/it]
 33%|███▎      | 168/512 [18:53<38:51,  6.78s/it]
                                                 
{'loss': '0.01094', 'grad_norm': '0.3047', 'learning_rate': '8.24e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 5079856, 'tokens/trainable': 76623, 'epoch': '0.6562'}
+
 33%|███▎      | 168/512 [18:53<38:51,  6.78s/it]
 33%|███▎      | 169/512 [18:59<38:32,  6.74s/it]
                                                 
{'loss': '0.009002', 'grad_norm': '0.3929', 'learning_rate': '8.217e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.39', 'tokens/total': 5110192, 'tokens/trainable': 77047, 'epoch': '0.6602'}
+
 33%|███▎      | 169/512 [18:59<38:32,  6.74s/it]
 33%|███▎      | 170/512 [19:06<38:25,  6.74s/it]
                                                 
{'loss': '0.001463', 'grad_norm': '0.09876', 'learning_rate': '8.194e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 5140592, 'tokens/trainable': 77496, 'epoch': '0.6641'}
+
 33%|███▎      | 170/512 [19:06<38:25,  6.74s/it]
 33%|███▎      | 171/512 [19:13<38:15,  6.73s/it]
                                                 
{'loss': '0.0004142', 'grad_norm': '0.02294', 'learning_rate': '8.171e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 5170848, 'tokens/trainable': 77956, 'epoch': '0.668'}
+
 33%|███▎      | 171/512 [19:13<38:15,  6.73s/it]
 34%|███▎      | 172/512 [19:20<38:10,  6.74s/it]
                                                 
{'loss': '0.006397', 'grad_norm': '0.2763', 'learning_rate': '8.147e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '29.78', 'tokens/total': 5201280, 'tokens/trainable': 78410, 'epoch': '0.6719'}
+
 34%|███▎      | 172/512 [19:20<38:10,  6.74s/it]
 34%|███▍      | 173/512 [19:26<37:56,  6.72s/it]
                                                 
{'loss': '0.001727', 'grad_norm': '0.07086', 'learning_rate': '8.124e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 5231712, 'tokens/trainable': 78849, 'epoch': '0.6758'}
+
 34%|███▍      | 173/512 [19:26<37:56,  6.72s/it]
 34%|███▍      | 174/512 [19:33<37:33,  6.67s/it]
                                                 
{'loss': '0.01194', 'grad_norm': '0.4524', 'learning_rate': '8.1e-05', 'ppl': '1.012', 'memory/max_active (GiB)': '33.62', 'memory/max_allocated (GiB)': '33.62', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.07', 'tokens/total': 5261440, 'tokens/trainable': 79266, 'epoch': '0.6797'}
+
 34%|███▍      | 174/512 [19:33<37:33,  6.67s/it]
 34%|███▍      | 175/512 [19:40<37:30,  6.68s/it]
                                                 
{'loss': '0.002157', 'grad_norm': '0.1293', 'learning_rate': '8.076e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.53', 'tokens/total': 5291728, 'tokens/trainable': 79736, 'epoch': '0.6836'}
+
 34%|███▍      | 175/512 [19:40<37:30,  6.68s/it]
 34%|███▍      | 176/512 [19:46<37:30,  6.70s/it]
                                                 
{'loss': '0.001708', 'grad_norm': '0.1321', 'learning_rate': '8.053e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.17', 'tokens/total': 5322240, 'tokens/trainable': 80214, 'epoch': '0.6875'}
+
 34%|███▍      | 176/512 [19:46<37:30,  6.70s/it]
 35%|███▍      | 177/512 [19:53<37:23,  6.70s/it]
                                                 
{'loss': '0.002082', 'grad_norm': '0.09734', 'learning_rate': '8.029e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.56', 'tokens/total': 5352576, 'tokens/trainable': 80661, 'epoch': '0.6914'}
+
 35%|███▍      | 177/512 [19:53<37:23,  6.70s/it]
 35%|███▍      | 178/512 [20:00<37:19,  6.71s/it]
                                                 
{'loss': '0.006122', 'grad_norm': '0.8079', 'learning_rate': '8.005e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.82', 'tokens/total': 5382784, 'tokens/trainable': 81129, 'epoch': '0.6953'}
+
 35%|███▍      | 178/512 [20:00<37:19,  6.71s/it]
 35%|███▍      | 179/512 [20:06<37:05,  6.68s/it]
                                                 
{'loss': '0.003698', 'grad_norm': '0.8061', 'learning_rate': '7.98e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 5412784, 'tokens/trainable': 81576, 'epoch': '0.6992'}
+
 35%|███▍      | 179/512 [20:06<37:05,  6.68s/it]
 35%|███▌      | 180/512 [20:13<37:02,  6.69s/it]
                                                 
{'loss': '0.0007224', 'grad_norm': '0.05354', 'learning_rate': '7.956e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 5442960, 'tokens/trainable': 82049, 'epoch': '0.7031'}
+
 35%|███▌      | 180/512 [20:13<37:02,  6.69s/it]
 35%|███▌      | 181/512 [20:20<37:01,  6.71s/it]
                                                 
{'loss': '0.01404', 'grad_norm': '0.8788', 'learning_rate': '7.932e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '38.79', 'tokens/total': 5473216, 'tokens/trainable': 82566, 'epoch': '0.707'}
+
 35%|███▌      | 181/512 [20:20<37:01,  6.71s/it]
 36%|███▌      | 182/512 [20:27<36:56,  6.72s/it]
                                                 
{'loss': '0.001835', 'grad_norm': '0.2329', 'learning_rate': '7.907e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 5503568, 'tokens/trainable': 83054, 'epoch': '0.7109'}
+
 36%|███▌      | 182/512 [20:27<36:56,  6.72s/it]
 36%|███▌      | 183/512 [20:33<36:48,  6.71s/it]
                                                 
{'loss': '0.001123', 'grad_norm': '0.08039', 'learning_rate': '7.883e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 5533760, 'tokens/trainable': 83501, 'epoch': '0.7148'}
+
 36%|███▌      | 183/512 [20:33<36:48,  6.71s/it]
 36%|███▌      | 184/512 [20:40<36:41,  6.71s/it]
                                                 
{'loss': '0.0004567', 'grad_norm': '0.02801', 'learning_rate': '7.858e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.48', 'tokens/total': 5564240, 'tokens/trainable': 84007, 'epoch': '0.7188'}
+
 36%|███▌      | 184/512 [20:40<36:41,  6.71s/it]
 36%|███▌      | 185/512 [20:47<36:36,  6.72s/it]
                                                 
{'loss': '0.0006747', 'grad_norm': '0.04384', 'learning_rate': '7.833e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.54', 'tokens/total': 5594448, 'tokens/trainable': 84491, 'epoch': '0.7227'}
+
 36%|███▌      | 185/512 [20:47<36:36,  6.72s/it]
 36%|███▋      | 186/512 [20:53<36:33,  6.73s/it]
                                                 
{'loss': '0.001428', 'grad_norm': '0.117', 'learning_rate': '7.808e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 5624816, 'tokens/trainable': 84952, 'epoch': '0.7266'}
+
 36%|███▋      | 186/512 [20:53<36:33,  6.73s/it]
 37%|███▋      | 187/512 [21:00<36:25,  6.72s/it]
                                                 
{'loss': '0.0002471', 'grad_norm': '0.05614', 'learning_rate': '7.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '32.92', 'tokens/total': 5655104, 'tokens/trainable': 85400, 'epoch': '0.7305'}
+
 37%|███▋      | 187/512 [21:00<36:25,  6.72s/it]
 37%|███▋      | 188/512 [21:07<36:20,  6.73s/it]
                                                 
{'loss': '0.004403', 'grad_norm': '0.3061', 'learning_rate': '7.758e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 5685360, 'tokens/trainable': 85889, 'epoch': '0.7344'}
+
 37%|███▋      | 188/512 [21:07<36:20,  6.73s/it]
 37%|███▋      | 189/512 [21:14<36:14,  6.73s/it]
                                                 
{'loss': '0.00549', 'grad_norm': '0.4333', 'learning_rate': '7.733e-05', 'ppl': '1.006', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 5715648, 'tokens/trainable': 86396, 'epoch': '0.7383'}
+
 37%|███▋      | 189/512 [21:14<36:14,  6.73s/it]
 37%|███▋      | 190/512 [21:20<36:05,  6.73s/it]
                                                 
{'loss': '0.006627', 'grad_norm': '0.4977', 'learning_rate': '7.708e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '33.85', 'tokens/total': 5746240, 'tokens/trainable': 86867, 'epoch': '0.7422'}
+
 37%|███▋      | 190/512 [21:20<36:05,  6.73s/it]
 37%|███▋      | 191/512 [21:27<36:02,  6.74s/it]
                                                 
{'loss': '0.009029', 'grad_norm': '1', 'learning_rate': '7.683e-05', 'ppl': '1.009', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '36.07', 'tokens/total': 5776816, 'tokens/trainable': 87314, 'epoch': '0.7461'}
+
 37%|███▋      | 191/512 [21:27<36:02,  6.74s/it]
 38%|███▊      | 192/512 [21:34<35:50,  6.72s/it]
                                                 
{'loss': '5.811e-05', 'grad_norm': '0.006457', 'learning_rate': '7.657e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '30.38', 'tokens/total': 5807248, 'tokens/trainable': 87736, 'epoch': '0.75'}
+
 38%|███▊      | 192/512 [21:34<35:50,  6.72s/it][2026-08-18 14:26:33,856] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-192
+
 38%|███▊      | 193/512 [21:42<38:34,  7.26s/it]
                                                 
{'loss': '0.001513', 'grad_norm': '0.6313', 'learning_rate': '7.632e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.02', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 5837680, 'tokens/trainable': 88219, 'epoch': '0.7539'}
+
 38%|███▊      | 193/512 [21:42<38:34,  7.26s/it]
 38%|███▊      | 194/512 [21:49<37:39,  7.10s/it]
                                                 
{'loss': '3.505e-05', 'grad_norm': '0.003887', 'learning_rate': '7.606e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '36.15', 'tokens/total': 5868000, 'tokens/trainable': 88699, 'epoch': '0.7578'}
+
 38%|███▊      | 194/512 [21:49<37:39,  7.10s/it]
 38%|███▊      | 195/512 [21:55<36:01,  6.82s/it]
                                                 
{'loss': '0.0001714', 'grad_norm': '0.05365', 'learning_rate': '7.58e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.39', 'tokens/total': 5896224, 'tokens/trainable': 89132, 'epoch': '0.7617'}
+
 38%|███▊      | 195/512 [21:55<36:01,  6.82s/it]
 38%|███▊      | 196/512 [22:02<35:45,  6.79s/it]
                                                 
{'loss': '0.007072', 'grad_norm': '1.521', 'learning_rate': '7.555e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '37.24', 'tokens/total': 5926624, 'tokens/trainable': 89594, 'epoch': '0.7656'}
+
 38%|███▊      | 196/512 [22:02<35:45,  6.79s/it]
 38%|███▊      | 197/512 [22:08<34:42,  6.61s/it]
                                                 
{'loss': '0.0001249', 'grad_norm': '0.01594', 'learning_rate': '7.529e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.34', 'memory/max_allocated (GiB)': '33.34', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '38.08', 'tokens/total': 5954944, 'tokens/trainable': 90042, 'epoch': '0.7695'}
+
 38%|███▊      | 197/512 [22:08<34:42,  6.61s/it]
 39%|███▊      | 198/512 [22:15<34:46,  6.65s/it]
                                                 
{'loss': '0.004654', 'grad_norm': '0.4873', 'learning_rate': '7.503e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 5985200, 'tokens/trainable': 90503, 'epoch': '0.7734'}
+
 39%|███▊      | 198/512 [22:15<34:46,  6.65s/it]
 39%|███▉      | 199/512 [22:22<34:59,  6.71s/it]
                                                 
{'loss': '0.0007038', 'grad_norm': '0.06984', 'learning_rate': '7.477e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.46', 'tokens/total': 6015952, 'tokens/trainable': 90982, 'epoch': '0.7773'}
+
 39%|███▉      | 199/512 [22:22<34:59,  6.71s/it]
 39%|███▉      | 200/512 [22:28<34:54,  6.71s/it]
                                                 
{'loss': '0.001063', 'grad_norm': '0.09597', 'learning_rate': '7.451e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '39.3', 'tokens/total': 6046208, 'tokens/trainable': 91466, 'epoch': '0.7812'}
+
 39%|███▉      | 200/512 [22:28<34:54,  6.71s/it]
 39%|███▉      | 201/512 [22:35<34:42,  6.70s/it]
                                                 
{'loss': '0.0004073', 'grad_norm': '0.03423', 'learning_rate': '7.424e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 6076208, 'tokens/trainable': 91912, 'epoch': '0.7852'}
+
 39%|███▉      | 201/512 [22:35<34:42,  6.70s/it]
 39%|███▉      | 202/512 [22:42<34:34,  6.69s/it]
                                                 
{'loss': '0.003431', 'grad_norm': '0.2397', 'learning_rate': '7.398e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.86', 'tokens/total': 6106560, 'tokens/trainable': 92385, 'epoch': '0.7891'}
+
 39%|███▉      | 202/512 [22:42<34:34,  6.69s/it]
 40%|███▉      | 203/512 [22:48<34:26,  6.69s/it]
                                                 
{'loss': '0.0216', 'grad_norm': '0.88', 'learning_rate': '7.372e-05', 'ppl': '1.022', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.55', 'tokens/total': 6136640, 'tokens/trainable': 92842, 'epoch': '0.793'}
+
 40%|███▉      | 203/512 [22:48<34:26,  6.69s/it]
 40%|███▉      | 204/512 [22:55<34:22,  6.70s/it]
                                                 
{'loss': '0.0006239', 'grad_norm': '0.1207', 'learning_rate': '7.345e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.97', 'tokens/total': 6167184, 'tokens/trainable': 93323, 'epoch': '0.7969'}
+
 40%|███▉      | 204/512 [22:55<34:22,  6.70s/it]
 40%|████      | 205/512 [23:02<34:10,  6.68s/it]
                                                 
{'loss': '0.002666', 'grad_norm': '0.3556', 'learning_rate': '7.319e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 6197376, 'tokens/trainable': 93756, 'epoch': '0.8008'}
+
 40%|████      | 205/512 [23:02<34:10,  6.68s/it]
 40%|████      | 206/512 [23:09<34:09,  6.70s/it]
                                                 
{'loss': '0.001572', 'grad_norm': '0.2445', 'learning_rate': '7.292e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '37.54', 'tokens/total': 6227600, 'tokens/trainable': 94235, 'epoch': '0.8047'}
+
 40%|████      | 206/512 [23:09<34:09,  6.70s/it]
 40%|████      | 207/512 [23:15<34:03,  6.70s/it]
                                                 
{'loss': '0.003192', 'grad_norm': '0.2322', 'learning_rate': '7.266e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '36.52', 'tokens/total': 6257760, 'tokens/trainable': 94733, 'epoch': '0.8086'}
+
 40%|████      | 207/512 [23:15<34:03,  6.70s/it]
 41%|████      | 208/512 [23:22<33:57,  6.70s/it]
                                                 
{'loss': '0.001887', 'grad_norm': '0.1369', 'learning_rate': '7.239e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.93', 'tokens/train_per_sec_per_gpu': '33.74', 'tokens/total': 6288048, 'tokens/trainable': 95218, 'epoch': '0.8125'}
+
 41%|████      | 208/512 [23:22<33:57,  6.70s/it]
 41%|████      | 209/512 [23:29<33:44,  6.68s/it]
                                                 
{'loss': '0.001802', 'grad_norm': '0.2183', 'learning_rate': '7.212e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '29.23', 'tokens/total': 6318688, 'tokens/trainable': 95619, 'epoch': '0.8164'}
+
 41%|████      | 209/512 [23:29<33:44,  6.68s/it]
 41%|████      | 210/512 [23:35<33:39,  6.69s/it]
                                                 
{'loss': '0.0125', 'grad_norm': '0.8775', 'learning_rate': '7.185e-05', 'ppl': '1.013', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.39', 'tokens/total': 6348944, 'tokens/trainable': 96076, 'epoch': '0.8203'}
+
 41%|████      | 210/512 [23:35<33:39,  6.69s/it]
 41%|████      | 211/512 [23:42<33:29,  6.68s/it]
                                                 
{'loss': '0.006944', 'grad_norm': '0.4041', 'learning_rate': '7.158e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.34', 'tokens/total': 6378960, 'tokens/trainable': 96505, 'epoch': '0.8242'}
+
 41%|████      | 211/512 [23:42<33:29,  6.68s/it]
 41%|████▏     | 212/512 [23:49<33:25,  6.68s/it]
                                                 
{'loss': '0.001738', 'grad_norm': '0.5415', 'learning_rate': '7.131e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.68', 'tokens/total': 6409472, 'tokens/trainable': 96930, 'epoch': '0.8281'}
+
 41%|████▏     | 212/512 [23:49<33:25,  6.68s/it]
 42%|████▏     | 213/512 [23:55<33:19,  6.69s/it]
                                                 
{'loss': '0.0002014', 'grad_norm': '0.01034', 'learning_rate': '7.104e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.72', 'tokens/total': 6439904, 'tokens/trainable': 97361, 'epoch': '0.832'}
+
 42%|████▏     | 213/512 [23:55<33:19,  6.69s/it]
 42%|████▏     | 214/512 [24:02<33:17,  6.70s/it]
                                                 
{'loss': '0.009724', 'grad_norm': '0.5834', 'learning_rate': '7.077e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 6470176, 'tokens/trainable': 97858, 'epoch': '0.8359'}
+
 42%|████▏     | 214/512 [24:02<33:17,  6.70s/it]
 42%|████▏     | 215/512 [24:09<33:06,  6.69s/it]
                                                 
{'loss': '0.004063', 'grad_norm': '0.3147', 'learning_rate': '7.05e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.47', 'tokens/total': 6500336, 'tokens/trainable': 98308, 'epoch': '0.8398'}
+
 42%|████▏     | 215/512 [24:09<33:06,  6.69s/it]
 42%|████▏     | 216/512 [24:15<32:58,  6.68s/it]
                                                 
{'loss': '0.0002783', 'grad_norm': '0.01362', 'learning_rate': '7.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.41', 'tokens/total': 6530688, 'tokens/trainable': 98781, 'epoch': '0.8438'}
+
 42%|████▏     | 216/512 [24:15<32:58,  6.68s/it]
 42%|████▏     | 217/512 [24:22<32:54,  6.69s/it]
                                                 
{'loss': '0.0002952', 'grad_norm': '0.02845', 'learning_rate': '6.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.76', 'tokens/total': 6561040, 'tokens/trainable': 99266, 'epoch': '0.8477'}
+
 42%|████▏     | 217/512 [24:22<32:54,  6.69s/it]
 43%|████▎     | 218/512 [24:28<32:02,  6.54s/it]
                                                 
{'loss': '0.002927', 'grad_norm': '0.1757', 'learning_rate': '6.968e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '39.65', 'tokens/total': 6589392, 'tokens/trainable': 99731, 'epoch': '0.8516'}
+
 43%|████▎     | 218/512 [24:28<32:02,  6.54s/it]
 43%|████▎     | 219/512 [24:35<32:09,  6.58s/it]
                                                 
{'loss': '0.005347', 'grad_norm': '0.2501', 'learning_rate': '6.94e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.71', 'tokens/total': 6619824, 'tokens/trainable': 100164, 'epoch': '0.8555'}
+
 43%|████▎     | 219/512 [24:35<32:09,  6.58s/it]
 43%|████▎     | 220/512 [24:42<32:16,  6.63s/it]
                                                 
{'loss': '0.00676', 'grad_norm': '0.366', 'learning_rate': '6.913e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '36.18', 'tokens/total': 6650224, 'tokens/trainable': 100620, 'epoch': '0.8594'}
+
 43%|████▎     | 220/512 [24:42<32:16,  6.63s/it]
 43%|████▎     | 221/512 [24:49<32:17,  6.66s/it]
                                                 
{'loss': '0.004021', 'grad_norm': '0.2871', 'learning_rate': '6.885e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '32.94', 'tokens/total': 6680384, 'tokens/trainable': 101080, 'epoch': '0.8633'}
+
 43%|████▎     | 221/512 [24:49<32:17,  6.66s/it]
 43%|████▎     | 222/512 [24:55<32:13,  6.67s/it]
                                                 
{'loss': '0.0006698', 'grad_norm': '0.05318', 'learning_rate': '6.857e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '37.8', 'tokens/total': 6710896, 'tokens/trainable': 101546, 'epoch': '0.8672'}
+
 43%|████▎     | 222/512 [24:55<32:13,  6.67s/it]
 44%|████▎     | 223/512 [25:02<32:06,  6.67s/it]
                                                 
{'loss': '0.0002042', 'grad_norm': '0.01632', 'learning_rate': '6.83e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '31.52', 'tokens/total': 6741280, 'tokens/trainable': 101987, 'epoch': '0.8711'}
+
 44%|████▎     | 223/512 [25:02<32:06,  6.67s/it]
 44%|████▍     | 224/512 [25:09<32:05,  6.69s/it]
                                                 
{'loss': '0.001851', 'grad_norm': '0.09936', 'learning_rate': '6.802e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '35.37', 'tokens/total': 6771488, 'tokens/trainable': 102435, 'epoch': '0.875'}
+
 44%|████▍     | 224/512 [25:09<32:05,  6.69s/it][2026-08-18 14:30:08,595] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-224
+
 44%|████▍     | 225/512 [25:17<34:28,  7.21s/it]
                                                 
{'loss': '0.0007717', 'grad_norm': '0.04545', 'learning_rate': '6.774e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.94', 'tokens/train_per_sec_per_gpu': '28.86', 'tokens/total': 6801744, 'tokens/trainable': 102843, 'epoch': '0.8789'}
+
 44%|████▍     | 225/512 [25:17<34:28,  7.21s/it]
 44%|████▍     | 226/512 [25:24<33:50,  7.10s/it]
                                                 
{'loss': '0.001548', 'grad_norm': '0.1694', 'learning_rate': '6.746e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.39', 'tokens/total': 6832432, 'tokens/trainable': 103342, 'epoch': '0.8828'}
+
 44%|████▍     | 226/512 [25:24<33:50,  7.10s/it]
 44%|████▍     | 227/512 [25:31<33:06,  6.97s/it]
                                                 
{'loss': '0.0002161', 'grad_norm': '0.01104', 'learning_rate': '6.718e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.71', 'tokens/train_per_sec_per_gpu': '40.81', 'tokens/total': 6862448, 'tokens/trainable': 103844, 'epoch': '0.8867'}
+
 44%|████▍     | 227/512 [25:31<33:06,  6.97s/it]
 45%|████▍     | 228/512 [25:37<32:37,  6.89s/it]
                                                 
{'loss': '0.007129', 'grad_norm': '0.3398', 'learning_rate': '6.69e-05', 'ppl': '1.007', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 6892832, 'tokens/trainable': 104307, 'epoch': '0.8906'}
+
 45%|████▍     | 228/512 [25:37<32:37,  6.89s/it]
 45%|████▍     | 229/512 [25:44<32:10,  6.82s/it]
                                                 
{'loss': '0.0003861', 'grad_norm': '0.02249', 'learning_rate': '6.662e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.69', 'memory/max_allocated (GiB)': '33.69', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.49', 'tokens/total': 6922832, 'tokens/trainable': 104772, 'epoch': '0.8945'}
+
 45%|████▍     | 229/512 [25:44<32:10,  6.82s/it]
 45%|████▍     | 230/512 [25:51<31:56,  6.80s/it]
                                                 
{'loss': '0.0002201', 'grad_norm': '0.03081', 'learning_rate': '6.634e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 6953104, 'tokens/trainable': 105240, 'epoch': '0.8984'}
+
 45%|████▍     | 230/512 [25:51<31:56,  6.80s/it]
 45%|████▌     | 231/512 [25:57<31:41,  6.77s/it]
                                                 
{'loss': '0.001256', 'grad_norm': '0.1375', 'learning_rate': '6.606e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.76', 'tokens/total': 6983440, 'tokens/trainable': 105703, 'epoch': '0.9023'}
+
 45%|████▌     | 231/512 [25:57<31:41,  6.77s/it]
 45%|████▌     | 232/512 [26:04<31:28,  6.74s/it]
                                                 
{'loss': '6.874e-05', 'grad_norm': '0.003156', 'learning_rate': '6.578e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.77', 'tokens/train_per_sec_per_gpu': '33.02', 'tokens/total': 7013840, 'tokens/trainable': 106179, 'epoch': '0.9062'}
+
 45%|████▌     | 232/512 [26:04<31:28,  6.74s/it]
 46%|████▌     | 233/512 [26:11<31:21,  6.74s/it]
                                                 
{'loss': '0.000162', 'grad_norm': '0.02122', 'learning_rate': '6.55e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38', 'tokens/total': 7044240, 'tokens/trainable': 106683, 'epoch': '0.9102'}
+
 46%|████▌     | 233/512 [26:11<31:21,  6.74s/it]
 46%|████▌     | 234/512 [26:17<31:06,  6.71s/it]
                                                 
{'loss': '0.0003435', 'grad_norm': '0.03298', 'learning_rate': '6.522e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 7074224, 'tokens/trainable': 107121, 'epoch': '0.9141'}
+
 46%|████▌     | 234/512 [26:17<31:06,  6.71s/it]
 46%|████▌     | 235/512 [26:24<30:58,  6.71s/it]
                                                 
{'loss': '0.0006108', 'grad_norm': '0.08474', 'learning_rate': '6.493e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.7', 'tokens/total': 7104672, 'tokens/trainable': 107561, 'epoch': '0.918'}
+
 46%|████▌     | 235/512 [26:24<30:58,  6.71s/it]
 46%|████▌     | 236/512 [26:31<30:49,  6.70s/it]
                                                 
{'loss': '0.0001098', 'grad_norm': '0.012', 'learning_rate': '6.465e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 7134784, 'tokens/trainable': 108018, 'epoch': '0.9219'}
+
 46%|████▌     | 236/512 [26:31<30:49,  6.70s/it]
 46%|████▋     | 237/512 [26:37<30:39,  6.69s/it]
                                                 
{'loss': '6.363e-05', 'grad_norm': '0.01181', 'learning_rate': '6.437e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.32', 'tokens/total': 7164960, 'tokens/trainable': 108446, 'epoch': '0.9258'}
+
 46%|████▋     | 237/512 [26:37<30:39,  6.69s/it]
 46%|████▋     | 238/512 [26:44<30:30,  6.68s/it]
                                                 
{'loss': '0.0002591', 'grad_norm': '0.05564', 'learning_rate': '6.408e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.15', 'tokens/total': 7195344, 'tokens/trainable': 108866, 'epoch': '0.9297'}
+
 46%|████▋     | 238/512 [26:44<30:30,  6.68s/it]
 47%|████▋     | 239/512 [26:51<30:28,  6.70s/it]
                                                 
{'loss': '0.004054', 'grad_norm': '0.7226', 'learning_rate': '6.38e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 7225680, 'tokens/trainable': 109316, 'epoch': '0.9336'}
+
 47%|████▋     | 239/512 [26:51<30:28,  6.70s/it]
 47%|████▋     | 240/512 [26:58<30:24,  6.71s/it]
                                                 
{'loss': '0.0004681', 'grad_norm': '0.0682', 'learning_rate': '6.351e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.27', 'tokens/total': 7256112, 'tokens/trainable': 109787, 'epoch': '0.9375'}
+
 47%|████▋     | 240/512 [26:58<30:24,  6.71s/it]
 47%|████▋     | 241/512 [27:04<30:13,  6.69s/it]
                                                 
{'loss': '0.001304', 'grad_norm': '0.174', 'learning_rate': '6.323e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.35', 'tokens/total': 7286384, 'tokens/trainable': 110246, 'epoch': '0.9414'}
+
 47%|████▋     | 241/512 [27:04<30:13,  6.69s/it]
 47%|████▋     | 242/512 [27:11<30:07,  6.70s/it]
                                                 
{'loss': '0.0004871', 'grad_norm': '0.05733', 'learning_rate': '6.294e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '29.22', 'tokens/total': 7316688, 'tokens/trainable': 110667, 'epoch': '0.9453'}
+
 47%|████▋     | 242/512 [27:11<30:07,  6.70s/it]
 47%|████▋     | 243/512 [27:18<30:02,  6.70s/it]
                                                 
{'loss': '0.01397', 'grad_norm': '0.804', 'learning_rate': '6.266e-05', 'ppl': '1.014', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '38.13', 'tokens/total': 7346992, 'tokens/trainable': 111154, 'epoch': '0.9492'}
+
 47%|████▋     | 243/512 [27:18<30:02,  6.70s/it]
 48%|████▊     | 244/512 [27:24<29:59,  6.72s/it]
                                                 
{'loss': '0.0001144', 'grad_norm': '0.0327', 'learning_rate': '6.237e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 7377632, 'tokens/trainable': 111611, 'epoch': '0.9531'}
+
 48%|████▊     | 244/512 [27:24<29:59,  6.72s/it]
 48%|████▊     | 245/512 [27:31<29:48,  6.70s/it]
                                                 
{'loss': '4.604e-05', 'grad_norm': '0.006486', 'learning_rate': '6.208e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 7407872, 'tokens/trainable': 112102, 'epoch': '0.957'}
+
 48%|████▊     | 245/512 [27:31<29:48,  6.70s/it]
 48%|████▊     | 246/512 [27:38<29:40,  6.69s/it]
                                                 
{'loss': '0.0002825', 'grad_norm': '0.04384', 'learning_rate': '6.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '33.24', 'tokens/total': 7438448, 'tokens/trainable': 112569, 'epoch': '0.9609'}
+
 48%|████▊     | 246/512 [27:38<29:40,  6.69s/it]
 48%|████▊     | 247/512 [27:44<28:57,  6.56s/it]
                                                 
{'loss': '0.002453', 'grad_norm': '0.17', 'learning_rate': '6.151e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '37.71', 'tokens/total': 7466768, 'tokens/trainable': 113023, 'epoch': '0.9648'}
+
 48%|████▊     | 247/512 [27:44<28:57,  6.56s/it]
 48%|████▊     | 248/512 [27:51<29:08,  6.62s/it]
                                                 
{'loss': '0.0001876', 'grad_norm': '0.01652', 'learning_rate': '6.122e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.12', 'tokens/total': 7497360, 'tokens/trainable': 113464, 'epoch': '0.9688'}
+
 48%|████▊     | 248/512 [27:51<29:08,  6.62s/it]
 49%|████▊     | 249/512 [27:57<29:05,  6.64s/it]
                                                 
{'loss': '0.002968', 'grad_norm': '0.2792', 'learning_rate': '6.093e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 7527504, 'tokens/trainable': 113908, 'epoch': '0.9727'}
+
 49%|████▊     | 249/512 [27:57<29:05,  6.64s/it]
 49%|████▉     | 250/512 [28:04<28:24,  6.51s/it]
                                                 
{'loss': '0.005398', 'grad_norm': '0.4556', 'learning_rate': '6.065e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.3', 'tokens/total': 7555808, 'tokens/trainable': 114348, 'epoch': '0.9766'}
+
 49%|████▉     | 250/512 [28:04<28:24,  6.51s/it]
 49%|████▉     | 251/512 [28:10<28:33,  6.57s/it]
                                                 
{'loss': '0.0003356', 'grad_norm': '0.03177', 'learning_rate': '6.036e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.23', 'tokens/total': 7586320, 'tokens/trainable': 114830, 'epoch': '0.9805'}
+
 49%|████▉     | 251/512 [28:10<28:33,  6.57s/it]
 49%|████▉     | 252/512 [28:17<28:38,  6.61s/it]
                                                 
{'loss': '0.002921', 'grad_norm': '0.297', 'learning_rate': '6.007e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 7616576, 'tokens/trainable': 115278, 'epoch': '0.9844'}
+
 49%|████▉     | 252/512 [28:17<28:38,  6.61s/it]
 49%|████▉     | 253/512 [28:24<28:42,  6.65s/it]
                                                 
{'loss': '0.0004065', 'grad_norm': '0.09317', 'learning_rate': '5.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.29', 'tokens/total': 7647152, 'tokens/trainable': 115699, 'epoch': '0.9883'}
+
 49%|████▉     | 253/512 [28:24<28:42,  6.65s/it]
 50%|████▉     | 254/512 [28:30<28:39,  6.66s/it]
                                                 
{'loss': '0.001994', 'grad_norm': '0.2161', 'learning_rate': '5.949e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '30.81', 'tokens/total': 7677584, 'tokens/trainable': 116133, 'epoch': '0.9922'}
+
 50%|████▉     | 254/512 [28:30<28:39,  6.66s/it]
 50%|████▉     | 255/512 [28:37<28:33,  6.67s/it]
                                                 
{'loss': '0.0008475', 'grad_norm': '0.08144', 'learning_rate': '5.92e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 7708016, 'tokens/trainable': 116591, 'epoch': '0.9961'}
+
 50%|████▉     | 255/512 [28:37<28:33,  6.67s/it]
 50%|█████     | 256/512 [28:44<28:41,  6.72s/it]
                                                 
{'loss': '0.00071', 'grad_norm': '0.06009', 'learning_rate': '5.891e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.11', 'tokens/total': 7738608, 'tokens/trainable': 117039, 'epoch': '1'}
+
 50%|█████     | 256/512 [28:44<28:41,  6.72s/it][2026-08-18 14:33:44,024] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-256
+
 50%|█████     | 257/512 [28:53<31:19,  7.37s/it]
                                                 
{'loss': '0.0001019', 'grad_norm': '0.0194', 'learning_rate': '5.862e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.14', 'tokens/train_per_sec_per_gpu': '32.49', 'tokens/total': 7768944, 'tokens/trainable': 117495, 'epoch': '1.004'}
+
 50%|█████     | 257/512 [28:53<31:19,  7.37s/it]
 50%|█████     | 258/512 [29:00<30:19,  7.16s/it]
                                                 
{'loss': '0.0001532', 'grad_norm': '0.02072', 'learning_rate': '5.834e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.54', 'tokens/total': 7799232, 'tokens/trainable': 117925, 'epoch': '1.008'}
+
 50%|█████     | 258/512 [29:00<30:19,  7.16s/it]
 51%|█████     | 259/512 [29:06<29:37,  7.02s/it]
                                                 
{'loss': '4.119e-05', 'grad_norm': '0.002171', 'learning_rate': '5.805e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.66', 'tokens/total': 7829568, 'tokens/trainable': 118373, 'epoch': '1.012'}
+
 51%|█████     | 259/512 [29:06<29:37,  7.02s/it]
 51%|█████     | 260/512 [29:13<29:05,  6.93s/it]
                                                 
{'loss': '2.263e-05', 'grad_norm': '0.001111', 'learning_rate': '5.776e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 7860000, 'tokens/trainable': 118816, 'epoch': '1.016'}
+
 51%|█████     | 260/512 [29:13<29:05,  6.93s/it]
 51%|█████     | 261/512 [29:20<28:39,  6.85s/it]
                                                 
{'loss': '7.258e-05', 'grad_norm': '0.0133', 'learning_rate': '5.747e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 7890464, 'tokens/trainable': 119263, 'epoch': '1.02'}
+
 51%|█████     | 261/512 [29:20<28:39,  6.85s/it]
 51%|█████     | 262/512 [29:26<28:25,  6.82s/it]
                                                 
{'loss': '5.408e-05', 'grad_norm': '0.06038', 'learning_rate': '5.718e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.79', 'tokens/total': 7920944, 'tokens/trainable': 119772, 'epoch': '1.023'}
+
 51%|█████     | 262/512 [29:26<28:25,  6.82s/it]
 51%|█████▏    | 263/512 [29:33<28:12,  6.80s/it]
                                                 
{'loss': '5.063e-05', 'grad_norm': '0.003908', 'learning_rate': '5.689e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.87', 'tokens/total': 7951296, 'tokens/trainable': 120269, 'epoch': '1.027'}
+
 51%|█████▏    | 263/512 [29:33<28:12,  6.80s/it]
 52%|█████▏    | 264/512 [29:40<28:01,  6.78s/it]
                                                 
{'loss': '3.142e-05', 'grad_norm': '0.008469', 'learning_rate': '5.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.05', 'tokens/total': 7981856, 'tokens/trainable': 120746, 'epoch': '1.031'}
+
 52%|█████▏    | 264/512 [29:40<28:01,  6.78s/it]
 52%|█████▏    | 265/512 [29:47<27:46,  6.74s/it]
                                                 
{'loss': '0.001199', 'grad_norm': '0.2763', 'learning_rate': '5.631e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.3', 'tokens/total': 8012192, 'tokens/trainable': 121204, 'epoch': '1.035'}
+
 52%|█████▏    | 265/512 [29:47<27:46,  6.74s/it]
 52%|█████▏    | 266/512 [29:53<27:40,  6.75s/it]
                                                 
{'loss': '0.01105', 'grad_norm': '0.3999', 'learning_rate': '5.602e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.23', 'tokens/total': 8042736, 'tokens/trainable': 121670, 'epoch': '1.039'}
+
 52%|█████▏    | 266/512 [29:53<27:40,  6.75s/it]
 52%|█████▏    | 267/512 [30:00<27:32,  6.74s/it]
                                                 
{'loss': '0.002067', 'grad_norm': '0.1673', 'learning_rate': '5.573e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '40.11', 'tokens/total': 8073024, 'tokens/trainable': 122183, 'epoch': '1.043'}
+
 52%|█████▏    | 267/512 [30:00<27:32,  6.74s/it]
 52%|█████▏    | 268/512 [30:07<27:22,  6.73s/it]
                                                 
{'loss': '0.0002631', 'grad_norm': '0.03506', 'learning_rate': '5.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.41', 'tokens/total': 8103280, 'tokens/trainable': 122645, 'epoch': '1.047'}
+
 52%|█████▏    | 268/512 [30:07<27:22,  6.73s/it]
 53%|█████▎    | 269/512 [30:13<27:11,  6.71s/it]
                                                 
{'loss': '7.13e-05', 'grad_norm': '0.006357', 'learning_rate': '5.515e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.09', 'tokens/total': 8133440, 'tokens/trainable': 123105, 'epoch': '1.051'}
+
 53%|█████▎    | 269/512 [30:13<27:11,  6.71s/it]
 53%|█████▎    | 270/512 [30:20<27:06,  6.72s/it]
                                                 
{'loss': '0.000502', 'grad_norm': '0.06589', 'learning_rate': '5.485e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.24', 'tokens/total': 8163920, 'tokens/trainable': 123595, 'epoch': '1.055'}
+
 53%|█████▎    | 270/512 [30:20<27:06,  6.72s/it]
 53%|█████▎    | 271/512 [30:27<26:57,  6.71s/it]
                                                 
{'loss': '0.0007096', 'grad_norm': '0.05696', 'learning_rate': '5.456e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '37.65', 'tokens/total': 8194032, 'tokens/trainable': 124077, 'epoch': '1.059'}
+
 53%|█████▎    | 271/512 [30:27<26:57,  6.71s/it]
 53%|█████▎    | 272/512 [30:34<26:52,  6.72s/it]
                                                 
{'loss': '0.0002944', 'grad_norm': '0.09019', 'learning_rate': '5.427e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.77', 'tokens/total': 8224320, 'tokens/trainable': 124547, 'epoch': '1.062'}
+
 53%|█████▎    | 272/512 [30:34<26:52,  6.72s/it]
 53%|█████▎    | 273/512 [30:40<26:42,  6.71s/it]
                                                 
{'loss': '0.001365', 'grad_norm': '0.1311', 'learning_rate': '5.398e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34', 'tokens/total': 8254368, 'tokens/trainable': 124977, 'epoch': '1.066'}
+
 53%|█████▎    | 273/512 [30:40<26:42,  6.71s/it]
 54%|█████▎    | 274/512 [30:47<26:34,  6.70s/it]
                                                 
{'loss': '0.003348', 'grad_norm': '0.3515', 'learning_rate': '5.369e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.56', 'tokens/total': 8284944, 'tokens/trainable': 125438, 'epoch': '1.07'}
+
 54%|█████▎    | 274/512 [30:47<26:34,  6.70s/it]
 54%|█████▎    | 275/512 [30:54<26:30,  6.71s/it]
                                                 
{'loss': '0.001243', 'grad_norm': '0.3394', 'learning_rate': '5.34e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '31.49', 'tokens/total': 8315584, 'tokens/trainable': 125889, 'epoch': '1.074'}
+
 54%|█████▎    | 275/512 [30:54<26:30,  6.71s/it]
 54%|█████▍    | 276/512 [31:00<26:24,  6.71s/it]
                                                 
{'loss': '0.004629', 'grad_norm': '0.3979', 'learning_rate': '5.311e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 8345952, 'tokens/trainable': 126363, 'epoch': '1.078'}
+
 54%|█████▍    | 276/512 [31:00<26:24,  6.71s/it]
 54%|█████▍    | 277/512 [31:07<26:20,  6.73s/it]
                                                 
{'loss': '0.004942', 'grad_norm': '0.2943', 'learning_rate': '5.282e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 8376416, 'tokens/trainable': 126828, 'epoch': '1.082'}
+
 54%|█████▍    | 277/512 [31:07<26:20,  6.73s/it]
 54%|█████▍    | 278/512 [31:14<26:12,  6.72s/it]
                                                 
{'loss': '0.0001571', 'grad_norm': '0.04226', 'learning_rate': '5.253e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '38.21', 'tokens/total': 8406640, 'tokens/trainable': 127331, 'epoch': '1.086'}
+
 54%|█████▍    | 278/512 [31:14<26:12,  6.72s/it]
 54%|█████▍    | 279/512 [31:21<26:03,  6.71s/it]
                                                 
{'loss': '9.962e-05', 'grad_norm': '0.008294', 'learning_rate': '5.224e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '34.39', 'tokens/total': 8436784, 'tokens/trainable': 127798, 'epoch': '1.09'}
+
 54%|█████▍    | 279/512 [31:21<26:03,  6.71s/it]
 55%|█████▍    | 280/512 [31:27<25:56,  6.71s/it]
                                                 
{'loss': '0.0002119', 'grad_norm': '0.01799', 'learning_rate': '5.195e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '32.84', 'tokens/total': 8467152, 'tokens/trainable': 128247, 'epoch': '1.094'}
+
 55%|█████▍    | 280/512 [31:27<25:56,  6.71s/it]
 55%|█████▍    | 281/512 [31:34<25:51,  6.72s/it]
                                                 
{'loss': '0.0003', 'grad_norm': '0.05587', 'learning_rate': '5.166e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '36.22', 'tokens/total': 8497424, 'tokens/trainable': 128729, 'epoch': '1.098'}
+
 55%|█████▍    | 281/512 [31:34<25:51,  6.72s/it]
 55%|█████▌    | 282/512 [31:41<25:45,  6.72s/it]
                                                 
{'loss': '0.0003475', 'grad_norm': '0.0335', 'learning_rate': '5.138e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 8527616, 'tokens/trainable': 129203, 'epoch': '1.102'}
+
 55%|█████▌    | 282/512 [31:41<25:45,  6.72s/it]
 55%|█████▌    | 283/512 [31:47<25:39,  6.72s/it]
                                                 
{'loss': '2.523e-05', 'grad_norm': '0.00135', 'learning_rate': '5.109e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 8557984, 'tokens/trainable': 129653, 'epoch': '1.105'}
+
 55%|█████▌    | 283/512 [31:47<25:39,  6.72s/it]
 55%|█████▌    | 284/512 [31:54<25:34,  6.73s/it]
                                                 
{'loss': '0.0002408', 'grad_norm': '0.03435', 'learning_rate': '5.08e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 8588384, 'tokens/trainable': 130100, 'epoch': '1.109'}
+
 55%|█████▌    | 284/512 [31:54<25:34,  6.73s/it]
 56%|█████▌    | 285/512 [32:01<25:24,  6.72s/it]
                                                 
{'loss': '0.0001933', 'grad_norm': '0.06312', 'learning_rate': '5.051e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '30.49', 'tokens/total': 8618768, 'tokens/trainable': 130543, 'epoch': '1.113'}
+
 56%|█████▌    | 285/512 [32:01<25:24,  6.72s/it]
 56%|█████▌    | 286/512 [32:08<25:17,  6.72s/it]
                                                 
{'loss': '0.0001494', 'grad_norm': '0.01357', 'learning_rate': '5.022e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 8648976, 'tokens/trainable': 131000, 'epoch': '1.117'}
+
 56%|█████▌    | 286/512 [32:08<25:17,  6.72s/it]
 56%|█████▌    | 287/512 [32:14<25:09,  6.71s/it]
                                                 
{'loss': '9.589e-05', 'grad_norm': '0.008714', 'learning_rate': '4.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '34.04', 'tokens/total': 8679152, 'tokens/trainable': 131449, 'epoch': '1.121'}
+
 56%|█████▌    | 287/512 [32:14<25:09,  6.71s/it]
 56%|█████▋    | 288/512 [32:21<25:00,  6.70s/it]
                                                 
{'loss': '3.912e-05', 'grad_norm': '0.00438', 'learning_rate': '4.964e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '35.97', 'tokens/total': 8709504, 'tokens/trainable': 131923, 'epoch': '1.125'}
+
 56%|█████▋    | 288/512 [32:21<25:00,  6.70s/it][2026-08-18 14:37:20,961] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-288
+
 56%|█████▋    | 289/512 [32:30<27:12,  7.32s/it]
                                                 
{'loss': '0.003404', 'grad_norm': '0.348', 'learning_rate': '4.935e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.91', 'tokens/train_per_sec_per_gpu': '33.15', 'tokens/total': 8739680, 'tokens/trainable': 132397, 'epoch': '1.129'}
+
 56%|█████▋    | 289/512 [32:30<27:12,  7.32s/it]
 57%|█████▋    | 290/512 [32:36<26:24,  7.14s/it]
                                                 
{'loss': '2.929e-05', 'grad_norm': '0.002436', 'learning_rate': '4.907e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '36.4', 'tokens/total': 8770112, 'tokens/trainable': 132846, 'epoch': '1.133'}
+
 57%|█████▋    | 290/512 [32:36<26:24,  7.14s/it]
 57%|█████▋    | 291/512 [32:43<25:49,  7.01s/it]
                                                 
{'loss': '0.001151', 'grad_norm': '0.5378', 'learning_rate': '4.878e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.57', 'tokens/train_per_sec_per_gpu': '35.74', 'tokens/total': 8800368, 'tokens/trainable': 133347, 'epoch': '1.137'}
+
 57%|█████▋    | 291/512 [32:43<25:49,  7.01s/it]
 57%|█████▋    | 292/512 [32:50<25:20,  6.91s/it]
                                                 
{'loss': '4.125e-05', 'grad_norm': '0.005875', 'learning_rate': '4.849e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '34.29', 'tokens/total': 8830688, 'tokens/trainable': 133779, 'epoch': '1.141'}
+
 57%|█████▋    | 292/512 [32:50<25:20,  6.91s/it]
 57%|█████▋    | 293/512 [32:57<25:02,  6.86s/it]
                                                 
{'loss': '0.000114', 'grad_norm': '0.04662', 'learning_rate': '4.82e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 8861008, 'tokens/trainable': 134251, 'epoch': '1.145'}
+
 57%|█████▋    | 293/512 [32:57<25:02,  6.86s/it]
 57%|█████▋    | 294/512 [33:03<24:49,  6.83s/it]
                                                 
{'loss': '3.974e-05', 'grad_norm': '0.00466', 'learning_rate': '4.792e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '33.55', 'tokens/total': 8891760, 'tokens/trainable': 134683, 'epoch': '1.148'}
+
 57%|█████▋    | 294/512 [33:03<24:49,  6.83s/it]
 58%|█████▊    | 295/512 [33:10<24:35,  6.80s/it]
                                                 
{'loss': '1.725e-05', 'grad_norm': '0.003478', 'learning_rate': '4.763e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.59', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 8922192, 'tokens/trainable': 135165, 'epoch': '1.152'}
+
 58%|█████▊    | 295/512 [33:10<24:35,  6.80s/it]
 58%|█████▊    | 296/512 [33:17<24:19,  6.75s/it]
                                                 
{'loss': '6.337e-05', 'grad_norm': '0.00705', 'learning_rate': '4.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '33.24', 'tokens/total': 8952496, 'tokens/trainable': 135599, 'epoch': '1.156'}
+
 58%|█████▊    | 296/512 [33:17<24:19,  6.75s/it]
 58%|█████▊    | 297/512 [33:23<24:07,  6.73s/it]
                                                 
{'loss': '0.0001625', 'grad_norm': '0.03387', 'learning_rate': '4.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.67', 'tokens/total': 8982768, 'tokens/trainable': 136038, 'epoch': '1.16'}
+
 58%|█████▊    | 297/512 [33:23<24:07,  6.73s/it]
 58%|█████▊    | 298/512 [33:30<23:59,  6.73s/it]
                                                 
{'loss': '0.0001614', 'grad_norm': '0.01463', 'learning_rate': '4.677e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '32.9', 'tokens/total': 9013248, 'tokens/trainable': 136495, 'epoch': '1.164'}
+
 58%|█████▊    | 298/512 [33:30<23:59,  6.73s/it]
 58%|█████▊    | 299/512 [33:37<23:53,  6.73s/it]
                                                 
{'loss': '2.856e-05', 'grad_norm': '0.002429', 'learning_rate': '4.649e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.34', 'tokens/total': 9043632, 'tokens/trainable': 136953, 'epoch': '1.168'}
+
 58%|█████▊    | 299/512 [33:37<23:53,  6.73s/it]
 59%|█████▊    | 300/512 [33:44<23:42,  6.71s/it]
                                                 
{'loss': '0.000198', 'grad_norm': '0.02774', 'learning_rate': '4.62e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 9073936, 'tokens/trainable': 137412, 'epoch': '1.172'}
+
 59%|█████▊    | 300/512 [33:44<23:42,  6.71s/it]
 59%|█████▉    | 301/512 [33:50<23:35,  6.71s/it]
                                                 
{'loss': '0.001319', 'grad_norm': '0.2262', 'learning_rate': '4.592e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '35.68', 'tokens/total': 9104416, 'tokens/trainable': 137907, 'epoch': '1.176'}
+
 59%|█████▉    | 301/512 [33:50<23:35,  6.71s/it]
 59%|█████▉    | 302/512 [33:57<23:30,  6.71s/it]
                                                 
{'loss': '0.0001245', 'grad_norm': '0.03463', 'learning_rate': '4.563e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.33', 'tokens/total': 9134688, 'tokens/trainable': 138383, 'epoch': '1.18'}
+
 59%|█████▉    | 302/512 [33:57<23:30,  6.71s/it]
 59%|█████▉    | 303/512 [34:04<23:26,  6.73s/it]
                                                 
{'loss': '5.559e-05', 'grad_norm': '0.005657', 'learning_rate': '4.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '36.4', 'tokens/total': 9165088, 'tokens/trainable': 138842, 'epoch': '1.184'}
+
 59%|█████▉    | 303/512 [34:04<23:26,  6.73s/it]
 59%|█████▉    | 304/512 [34:10<23:14,  6.71s/it]
                                                 
{'loss': '1.902e-05', 'grad_norm': '0.007515', 'learning_rate': '4.507e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '30.67', 'tokens/total': 9195584, 'tokens/trainable': 139269, 'epoch': '1.188'}
+
 59%|█████▉    | 304/512 [34:10<23:14,  6.71s/it]
 60%|█████▉    | 305/512 [34:17<23:08,  6.71s/it]
                                                 
{'loss': '2.345e-05', 'grad_norm': '0.002446', 'learning_rate': '4.478e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.67', 'tokens/train_per_sec_per_gpu': '34.45', 'tokens/total': 9225760, 'tokens/trainable': 139736, 'epoch': '1.191'}
+
 60%|█████▉    | 305/512 [34:17<23:08,  6.71s/it]
 60%|█████▉    | 306/512 [34:24<23:01,  6.71s/it]
                                                 
{'loss': '0.0005937', 'grad_norm': '0.1801', 'learning_rate': '4.45e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '37.01', 'tokens/total': 9256352, 'tokens/trainable': 140226, 'epoch': '1.195'}
+
 60%|█████▉    | 306/512 [34:24<23:01,  6.71s/it]
 60%|█████▉    | 307/512 [34:30<22:17,  6.52s/it]
                                                 
{'loss': '2.11e-05', 'grad_norm': '0.001028', 'learning_rate': '4.422e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.19', 'memory/max_allocated (GiB)': '33.19', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 9284256, 'tokens/trainable': 140664, 'epoch': '1.199'}
+
 60%|█████▉    | 307/512 [34:30<22:17,  6.52s/it]
 60%|██████    | 308/512 [34:37<22:27,  6.61s/it]
                                                 
{'loss': '5.623e-05', 'grad_norm': '0.005494', 'learning_rate': '4.394e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.9', 'tokens/train_per_sec_per_gpu': '29.84', 'tokens/total': 9314768, 'tokens/trainable': 141115, 'epoch': '1.203'}
+
 60%|██████    | 308/512 [34:37<22:27,  6.61s/it]
 60%|██████    | 309/512 [34:43<22:28,  6.64s/it]
                                                 
{'loss': '1.449e-05', 'grad_norm': '0.0006828', 'learning_rate': '4.366e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '28.87', 'tokens/total': 9344976, 'tokens/trainable': 141530, 'epoch': '1.207'}
+
 60%|██████    | 309/512 [34:43<22:28,  6.64s/it]
 61%|██████    | 310/512 [34:50<22:23,  6.65s/it]
                                                 
{'loss': '0.002186', 'grad_norm': '0.7971', 'learning_rate': '4.338e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.2', 'tokens/total': 9375280, 'tokens/trainable': 141964, 'epoch': '1.211'}
+
 61%|██████    | 310/512 [34:50<22:23,  6.65s/it]
 61%|██████    | 311/512 [34:57<22:18,  6.66s/it]
                                                 
{'loss': '1.984e-05', 'grad_norm': '0.003865', 'learning_rate': '4.31e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 9405536, 'tokens/trainable': 142428, 'epoch': '1.215'}
+
 61%|██████    | 311/512 [34:57<22:18,  6.66s/it]
 61%|██████    | 312/512 [35:03<22:10,  6.65s/it]
                                                 
{'loss': '0.000209', 'grad_norm': '0.05813', 'learning_rate': '4.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '38.16', 'tokens/total': 9435696, 'tokens/trainable': 142894, 'epoch': '1.219'}
+
 61%|██████    | 312/512 [35:03<22:10,  6.65s/it]
 61%|██████    | 313/512 [35:10<22:07,  6.67s/it]
                                                 
{'loss': '0.000148', 'grad_norm': '0.01986', 'learning_rate': '4.254e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.37', 'tokens/total': 9466080, 'tokens/trainable': 143368, 'epoch': '1.223'}
+
 61%|██████    | 313/512 [35:10<22:07,  6.67s/it]
 61%|██████▏   | 314/512 [35:17<22:01,  6.68s/it]
                                                 
{'loss': '0.0001475', 'grad_norm': '0.05822', 'learning_rate': '4.226e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 9496400, 'tokens/trainable': 143824, 'epoch': '1.227'}
+
 61%|██████▏   | 314/512 [35:17<22:01,  6.68s/it]
 62%|██████▏   | 315/512 [35:24<22:02,  6.71s/it]
                                                 
{'loss': '0.0001151', 'grad_norm': '0.009975', 'learning_rate': '4.198e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.58', 'tokens/total': 9526880, 'tokens/trainable': 144306, 'epoch': '1.23'}
+
 62%|██████▏   | 315/512 [35:24<22:02,  6.71s/it]
 62%|██████▏   | 316/512 [35:30<21:54,  6.71s/it]
                                                 
{'loss': '6.155e-05', 'grad_norm': '0.00919', 'learning_rate': '4.17e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 9557280, 'tokens/trainable': 144763, 'epoch': '1.234'}
+
 62%|██████▏   | 316/512 [35:30<21:54,  6.71s/it]
 62%|██████▏   | 317/512 [35:37<21:49,  6.71s/it]
                                                 
{'loss': '2.067e-05', 'grad_norm': '0.001632', 'learning_rate': '4.143e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.45', 'tokens/total': 9587712, 'tokens/trainable': 145255, 'epoch': '1.238'}
+
 62%|██████▏   | 317/512 [35:37<21:49,  6.71s/it]
 62%|██████▏   | 318/512 [35:44<21:41,  6.71s/it]
                                                 
{'loss': '5.585e-05', 'grad_norm': '0.007459', 'learning_rate': '4.115e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.7', 'tokens/total': 9618064, 'tokens/trainable': 145714, 'epoch': '1.242'}
+
 62%|██████▏   | 318/512 [35:44<21:41,  6.71s/it]
 62%|██████▏   | 319/512 [35:50<21:35,  6.71s/it]
                                                 
{'loss': '0.0004338', 'grad_norm': '0.07343', 'learning_rate': '4.087e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '36.04', 'tokens/total': 9648336, 'tokens/trainable': 146200, 'epoch': '1.246'}
+
 62%|██████▏   | 319/512 [35:50<21:35,  6.71s/it]
 62%|██████▎   | 320/512 [35:57<21:28,  6.71s/it]
                                                 
{'loss': '4.525e-05', 'grad_norm': '0.007983', 'learning_rate': '4.06e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.28', 'tokens/total': 9678688, 'tokens/trainable': 146668, 'epoch': '1.25'}
+
 62%|██████▎   | 320/512 [35:57<21:28,  6.71s/it][2026-08-18 14:40:57,132] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-320
+
 63%|██████▎   | 321/512 [36:05<22:54,  7.20s/it]
                                                 
{'loss': '6.388e-05', 'grad_norm': '0.016', 'learning_rate': '4.032e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.35', 'tokens/total': 9708848, 'tokens/trainable': 147114, 'epoch': '1.254'}
+
 63%|██████▎   | 321/512 [36:05<22:54,  7.20s/it]
 63%|██████▎   | 322/512 [36:12<22:18,  7.04s/it]
                                                 
{'loss': '2.179e-05', 'grad_norm': '0.004208', 'learning_rate': '4.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 9739360, 'tokens/trainable': 147564, 'epoch': '1.258'}
+
 63%|██████▎   | 322/512 [36:12<22:18,  7.04s/it]
 63%|██████▎   | 323/512 [36:19<21:51,  6.94s/it]
                                                 
{'loss': '3.835e-05', 'grad_norm': '0.00943', 'learning_rate': '3.978e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.24', 'tokens/total': 9769776, 'tokens/trainable': 148013, 'epoch': '1.262'}
+
 63%|██████▎   | 323/512 [36:19<21:51,  6.94s/it]
 63%|██████▎   | 324/512 [36:26<21:35,  6.89s/it]
                                                 
{'loss': '0.0001877', 'grad_norm': '0.05088', 'learning_rate': '3.95e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '38.06', 'tokens/total': 9800096, 'tokens/trainable': 148470, 'epoch': '1.266'}
+
 63%|██████▎   | 324/512 [36:26<21:35,  6.89s/it]
 63%|██████▎   | 325/512 [36:32<21:15,  6.82s/it]
                                                 
{'loss': '0.01083', 'grad_norm': '0.9416', 'learning_rate': '3.923e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '36.53', 'tokens/total': 9830304, 'tokens/trainable': 148926, 'epoch': '1.27'}
+
 63%|██████▎   | 325/512 [36:32<21:15,  6.82s/it]
 64%|██████▎   | 326/512 [36:39<21:04,  6.80s/it]
                                                 
{'loss': '8.612e-06', 'grad_norm': '0.0003319', 'learning_rate': '3.896e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.82', 'tokens/total': 9860608, 'tokens/trainable': 149420, 'epoch': '1.273'}
+
 64%|██████▎   | 326/512 [36:39<21:04,  6.80s/it]
 64%|██████▍   | 327/512 [36:46<20:52,  6.77s/it]
                                                 
{'loss': '1.583e-05', 'grad_norm': '0.0005477', 'learning_rate': '3.869e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '39.41', 'tokens/total': 9890560, 'tokens/trainable': 149873, 'epoch': '1.277'}
+
 64%|██████▍   | 327/512 [36:46<20:52,  6.77s/it]
 64%|██████▍   | 328/512 [36:52<20:43,  6.76s/it]
                                                 
{'loss': '1.643e-05', 'grad_norm': '0.000826', 'learning_rate': '3.842e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.86', 'tokens/total': 9920880, 'tokens/trainable': 150347, 'epoch': '1.281'}
+
 64%|██████▍   | 328/512 [36:52<20:43,  6.76s/it]
 64%|██████▍   | 329/512 [36:59<20:33,  6.74s/it]
                                                 
{'loss': '1.752e-05', 'grad_norm': '0.001146', 'learning_rate': '3.815e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.83', 'tokens/total': 9951280, 'tokens/trainable': 150822, 'epoch': '1.285'}
+
 64%|██████▍   | 329/512 [36:59<20:33,  6.74s/it]
 64%|██████▍   | 330/512 [37:06<20:26,  6.74s/it]
                                                 
{'loss': '2.001e-05', 'grad_norm': '0.002252', 'learning_rate': '3.788e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 9981600, 'tokens/trainable': 151266, 'epoch': '1.289'}
+
 64%|██████▍   | 330/512 [37:06<20:26,  6.74s/it]
 65%|██████▍   | 331/512 [37:13<20:17,  6.73s/it]
                                                 
{'loss': '0.0003773', 'grad_norm': '0.08719', 'learning_rate': '3.761e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '37.15', 'tokens/total': 10012016, 'tokens/trainable': 151734, 'epoch': '1.293'}
+
 65%|██████▍   | 331/512 [37:13<20:17,  6.73s/it]
 65%|██████▍   | 332/512 [37:19<20:08,  6.71s/it]
                                                 
{'loss': '1.175e-05', 'grad_norm': '0.0008834', 'learning_rate': '3.734e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.01', 'tokens/total': 10042448, 'tokens/trainable': 152184, 'epoch': '1.297'}
+
 65%|██████▍   | 332/512 [37:19<20:08,  6.71s/it]
 65%|██████▌   | 333/512 [37:26<20:01,  6.71s/it]
                                                 
{'loss': '5.219e-05', 'grad_norm': '0.01143', 'learning_rate': '3.708e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.85', 'tokens/total': 10073024, 'tokens/trainable': 152623, 'epoch': '1.301'}
+
 65%|██████▌   | 333/512 [37:26<20:01,  6.71s/it]
 65%|██████▌   | 334/512 [37:33<19:51,  6.69s/it]
                                                 
{'loss': '0.003487', 'grad_norm': '0.5781', 'learning_rate': '3.681e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.44', 'tokens/total': 10103504, 'tokens/trainable': 153040, 'epoch': '1.305'}
+
 65%|██████▌   | 334/512 [37:33<19:51,  6.69s/it]
 65%|██████▌   | 335/512 [37:39<19:18,  6.54s/it]
                                                 
{'loss': '0.004716', 'grad_norm': '0.7214', 'learning_rate': '3.655e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.31', 'memory/max_allocated (GiB)': '33.31', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.78', 'tokens/total': 10131632, 'tokens/trainable': 153463, 'epoch': '1.309'}
+
 65%|██████▌   | 335/512 [37:39<19:18,  6.54s/it]
 66%|██████▌   | 336/512 [37:45<19:18,  6.58s/it]
                                                 
{'loss': '0.0002346', 'grad_norm': '0.03132', 'learning_rate': '3.628e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.31', 'tokens/total': 10161824, 'tokens/trainable': 153908, 'epoch': '1.312'}
+
 66%|██████▌   | 336/512 [37:45<19:18,  6.58s/it]
 66%|██████▌   | 337/512 [37:52<19:22,  6.64s/it]
                                                 
{'loss': '0.0001198', 'grad_norm': '0.03333', 'learning_rate': '3.602e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 10192448, 'tokens/trainable': 154387, 'epoch': '1.316'}
+
 66%|██████▌   | 337/512 [37:52<19:22,  6.64s/it]
 66%|██████▌   | 338/512 [37:59<19:20,  6.67s/it]
                                                 
{'loss': '3.549e-05', 'grad_norm': '0.004504', 'learning_rate': '3.576e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.94', 'tokens/total': 10222704, 'tokens/trainable': 154836, 'epoch': '1.32'}
+
 66%|██████▌   | 338/512 [37:59<19:20,  6.67s/it]
 66%|██████▌   | 339/512 [38:06<19:19,  6.70s/it]
                                                 
{'loss': '1.789e-05', 'grad_norm': '0.001243', 'learning_rate': '3.549e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '29.81', 'tokens/total': 10253168, 'tokens/trainable': 155261, 'epoch': '1.324'}
+
 66%|██████▌   | 339/512 [38:06<19:19,  6.70s/it]
 66%|██████▋   | 340/512 [38:12<19:08,  6.68s/it]
                                                 
{'loss': '0.00279', 'grad_norm': '0.3251', 'learning_rate': '3.523e-05', 'ppl': '1.003', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '31.24', 'tokens/total': 10283552, 'tokens/trainable': 155674, 'epoch': '1.328'}
+
 66%|██████▋   | 340/512 [38:12<19:08,  6.68s/it]
 67%|██████▋   | 341/512 [38:19<19:05,  6.70s/it]
                                                 
{'loss': '0.0001325', 'grad_norm': '0.03649', 'learning_rate': '3.497e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.78', 'tokens/total': 10313872, 'tokens/trainable': 156164, 'epoch': '1.332'}
+
 67%|██████▋   | 341/512 [38:19<19:05,  6.70s/it]
 67%|██████▋   | 342/512 [38:26<18:57,  6.69s/it]
                                                 
{'loss': '0.002143', 'grad_norm': '0.2706', 'learning_rate': '3.471e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.65', 'tokens/total': 10344256, 'tokens/trainable': 156611, 'epoch': '1.336'}
+
 67%|██████▋   | 342/512 [38:26<18:57,  6.69s/it]
 67%|██████▋   | 343/512 [38:33<18:50,  6.69s/it]
                                                 
{'loss': '0.0001378', 'grad_norm': '0.01915', 'learning_rate': '3.445e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.68', 'tokens/total': 10374544, 'tokens/trainable': 157073, 'epoch': '1.34'}
+
 67%|██████▋   | 343/512 [38:33<18:50,  6.69s/it]
 67%|██████▋   | 344/512 [38:39<18:45,  6.70s/it]
                                                 
{'loss': '5.135e-05', 'grad_norm': '0.0254', 'learning_rate': '3.42e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 10405040, 'tokens/trainable': 157523, 'epoch': '1.344'}
+
 67%|██████▋   | 344/512 [38:39<18:45,  6.70s/it]
 67%|██████▋   | 345/512 [38:46<18:37,  6.69s/it]
                                                 
{'loss': '0.009803', 'grad_norm': '0.3926', 'learning_rate': '3.394e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.16', 'tokens/total': 10435424, 'tokens/trainable': 157964, 'epoch': '1.348'}
+
 67%|██████▋   | 345/512 [38:46<18:37,  6.69s/it]
 68%|██████▊   | 346/512 [38:53<18:30,  6.69s/it]
                                                 
{'loss': '2.997e-05', 'grad_norm': '0.002907', 'learning_rate': '3.368e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '30.44', 'tokens/total': 10465760, 'tokens/trainable': 158390, 'epoch': '1.352'}
+
 68%|██████▊   | 346/512 [38:53<18:30,  6.69s/it]
 68%|██████▊   | 347/512 [38:59<18:20,  6.67s/it]
                                                 
{'loss': '7.613e-05', 'grad_norm': '0.02028', 'learning_rate': '3.343e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 10495904, 'tokens/trainable': 158848, 'epoch': '1.355'}
+
 68%|██████▊   | 347/512 [38:59<18:20,  6.67s/it]
 68%|██████▊   | 348/512 [39:06<18:18,  6.70s/it]
                                                 
{'loss': '0.001124', 'grad_norm': '0.1487', 'learning_rate': '3.317e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '34', 'memory/max_allocated (GiB)': '34', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.71', 'tokens/total': 10526624, 'tokens/trainable': 159270, 'epoch': '1.359'}
+
 68%|██████▊   | 348/512 [39:06<18:18,  6.70s/it]
 68%|██████▊   | 349/512 [39:13<18:10,  6.69s/it]
                                                 
{'loss': '7.921e-05', 'grad_norm': '0.01538', 'learning_rate': '3.292e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.51', 'tokens/total': 10556752, 'tokens/trainable': 159720, 'epoch': '1.363'}
+
 68%|██████▊   | 349/512 [39:13<18:10,  6.69s/it]
 68%|██████▊   | 350/512 [39:19<18:09,  6.73s/it]
                                                 
{'loss': '1.457e-05', 'grad_norm': '0.00237', 'learning_rate': '3.267e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.99', 'memory/max_allocated (GiB)': '33.99', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.14', 'tokens/total': 10587264, 'tokens/trainable': 160187, 'epoch': '1.367'}
+
 68%|██████▊   | 350/512 [39:19<18:09,  6.73s/it]
 69%|██████▊   | 351/512 [39:26<18:02,  6.72s/it]
                                                 
{'loss': '6.53e-05', 'grad_norm': '0.009544', 'learning_rate': '3.242e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '32.63', 'tokens/total': 10617504, 'tokens/trainable': 160645, 'epoch': '1.371'}
+
 69%|██████▊   | 351/512 [39:26<18:02,  6.72s/it]
 69%|██████▉   | 352/512 [39:33<17:54,  6.72s/it]
                                                 
{'loss': '0.01059', 'grad_norm': '0.6505', 'learning_rate': '3.217e-05', 'ppl': '1.011', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '34.49', 'tokens/total': 10647760, 'tokens/trainable': 161102, 'epoch': '1.375'}
+
 69%|██████▉   | 352/512 [39:33<17:54,  6.72s/it][2026-08-18 14:44:32,873] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-352
+
 69%|██████▉   | 353/512 [39:41<19:05,  7.21s/it]
                                                 
{'loss': '0.000143', 'grad_norm': '0.02473', 'learning_rate': '3.192e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '28.93', 'tokens/total': 10678080, 'tokens/trainable': 161555, 'epoch': '1.379'}
+
 69%|██████▉   | 353/512 [39:41<19:05,  7.21s/it]
 69%|██████▉   | 354/512 [39:48<18:34,  7.05s/it]
                                                 
{'loss': '7.214e-05', 'grad_norm': '0.03731', 'learning_rate': '3.167e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.41', 'tokens/train_per_sec_per_gpu': '36.75', 'tokens/total': 10708544, 'tokens/trainable': 162020, 'epoch': '1.383'}
+
 69%|██████▉   | 354/512 [39:48<18:34,  7.05s/it]
 69%|██████▉   | 355/512 [39:55<18:10,  6.95s/it]
                                                 
{'loss': '0.0001183', 'grad_norm': '0.00885', 'learning_rate': '3.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '32.1', 'tokens/total': 10739024, 'tokens/trainable': 162448, 'epoch': '1.387'}
+
 69%|██████▉   | 355/512 [39:55<18:10,  6.95s/it]
 70%|██████▉   | 356/512 [40:01<17:48,  6.85s/it]
                                                 
{'loss': '0.000361', 'grad_norm': '0.06792', 'learning_rate': '3.117e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.61', 'memory/max_allocated (GiB)': '33.61', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '28.26', 'tokens/total': 10769024, 'tokens/trainable': 162863, 'epoch': '1.391'}
+
 70%|██████▉   | 356/512 [40:01<17:48,  6.85s/it]
 70%|██████▉   | 357/512 [40:08<17:35,  6.81s/it]
                                                 
{'loss': '3.174e-05', 'grad_norm': '0.001951', 'learning_rate': '3.093e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.89', 'tokens/total': 10799328, 'tokens/trainable': 163318, 'epoch': '1.395'}
+
 70%|██████▉   | 357/512 [40:08<17:35,  6.81s/it]
 70%|██████▉   | 358/512 [40:15<17:23,  6.78s/it]
                                                 
{'loss': '0.01027', 'grad_norm': '0.714', 'learning_rate': '3.068e-05', 'ppl': '1.01', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '33.08', 'tokens/total': 10829488, 'tokens/trainable': 163767, 'epoch': '1.398'}
+
 70%|██████▉   | 358/512 [40:15<17:23,  6.78s/it]
 70%|███████   | 359/512 [40:21<17:12,  6.75s/it]
                                                 
{'loss': '4.873e-05', 'grad_norm': '0.006852', 'learning_rate': '3.044e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 10859936, 'tokens/trainable': 164208, 'epoch': '1.402'}
+
 70%|███████   | 359/512 [40:21<17:12,  6.75s/it]
 70%|███████   | 360/512 [40:28<17:03,  6.73s/it]
                                                 
{'loss': '3.888e-05', 'grad_norm': '0.001825', 'learning_rate': '3.02e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.18', 'tokens/total': 10890400, 'tokens/trainable': 164655, 'epoch': '1.406'}
+
 70%|███████   | 360/512 [40:28<17:03,  6.73s/it]
 71%|███████   | 361/512 [40:35<16:52,  6.71s/it]
                                                 
{'loss': '0.0003688', 'grad_norm': '0.04369', 'learning_rate': '2.995e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.15', 'tokens/total': 10920624, 'tokens/trainable': 165100, 'epoch': '1.41'}
+
 71%|███████   | 361/512 [40:35<16:52,  6.71s/it]
 71%|███████   | 362/512 [40:41<16:48,  6.72s/it]
                                                 
{'loss': '0.0003961', 'grad_norm': '0.2674', 'learning_rate': '2.971e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.91', 'tokens/total': 10951136, 'tokens/trainable': 165587, 'epoch': '1.414'}
+
 71%|███████   | 362/512 [40:41<16:48,  6.72s/it]
 71%|███████   | 363/512 [40:48<16:38,  6.70s/it]
                                                 
{'loss': '0.0007331', 'grad_norm': '0.09645', 'learning_rate': '2.947e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.14', 'tokens/total': 10981344, 'tokens/trainable': 166034, 'epoch': '1.418'}
+
 71%|███████   | 363/512 [40:48<16:38,  6.70s/it]
 71%|███████   | 364/512 [40:55<16:32,  6.71s/it]
                                                 
{'loss': '0.0005385', 'grad_norm': '0.05368', 'learning_rate': '2.924e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '34.1', 'tokens/total': 11011552, 'tokens/trainable': 166485, 'epoch': '1.422'}
+
 71%|███████   | 364/512 [40:55<16:32,  6.71s/it]
 71%|███████▏  | 365/512 [41:02<16:26,  6.71s/it]
                                                 
{'loss': '0.0006869', 'grad_norm': '0.1007', 'learning_rate': '2.9e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '36.36', 'tokens/total': 11041872, 'tokens/trainable': 166973, 'epoch': '1.426'}
+
 71%|███████▏  | 365/512 [41:02<16:26,  6.71s/it]
 71%|███████▏  | 366/512 [41:08<16:19,  6.71s/it]
                                                 
{'loss': '0.0002465', 'grad_norm': '0.02698', 'learning_rate': '2.876e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.84', 'tokens/train_per_sec_per_gpu': '29.12', 'tokens/total': 11072192, 'tokens/trainable': 167414, 'epoch': '1.43'}
+
 71%|███████▏  | 366/512 [41:08<16:19,  6.71s/it]
 72%|███████▏  | 367/512 [41:15<16:12,  6.71s/it]
                                                 
{'loss': '5.263e-05', 'grad_norm': '0.002621', 'learning_rate': '2.853e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '30.23', 'tokens/total': 11102736, 'tokens/trainable': 167851, 'epoch': '1.434'}
+
 72%|███████▏  | 367/512 [41:15<16:12,  6.71s/it]
 72%|███████▏  | 368/512 [41:22<16:05,  6.70s/it]
                                                 
{'loss': '2.951e-05', 'grad_norm': '0.001814', 'learning_rate': '2.829e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.56', 'tokens/total': 11133056, 'tokens/trainable': 168294, 'epoch': '1.438'}
+
 72%|███████▏  | 368/512 [41:22<16:05,  6.70s/it]
 72%|███████▏  | 369/512 [41:28<16:01,  6.72s/it]
                                                 
{'loss': '7.607e-05', 'grad_norm': '0.01205', 'learning_rate': '2.806e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.23', 'tokens/total': 11163568, 'tokens/trainable': 168769, 'epoch': '1.441'}
+
 72%|███████▏  | 369/512 [41:28<16:01,  6.72s/it]
 72%|███████▏  | 370/512 [41:35<15:55,  6.73s/it]
                                                 
{'loss': '0.0001533', 'grad_norm': '0.01345', 'learning_rate': '2.783e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.83', 'tokens/total': 11193936, 'tokens/trainable': 169270, 'epoch': '1.445'}
+
 72%|███████▏  | 370/512 [41:35<15:55,  6.73s/it]
 72%|███████▏  | 371/512 [41:42<15:51,  6.74s/it]
                                                 
{'loss': '5.633e-05', 'grad_norm': '0.003458', 'learning_rate': '2.76e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.08', 'tokens/total': 11224480, 'tokens/trainable': 169711, 'epoch': '1.449'}
+
 72%|███████▏  | 371/512 [41:42<15:51,  6.74s/it]
 73%|███████▎  | 372/512 [41:49<15:44,  6.75s/it]
                                                 
{'loss': '9.535e-05', 'grad_norm': '0.006362', 'learning_rate': '2.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 11254912, 'tokens/trainable': 170185, 'epoch': '1.453'}
+
 73%|███████▎  | 372/512 [41:49<15:44,  6.75s/it]
 73%|███████▎  | 373/512 [41:55<15:37,  6.74s/it]
                                                 
{'loss': '0.0006136', 'grad_norm': '0.04559', 'learning_rate': '2.714e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.12', 'tokens/total': 11285104, 'tokens/trainable': 170650, 'epoch': '1.457'}
+
 73%|███████▎  | 373/512 [41:55<15:37,  6.74s/it]
 73%|███████▎  | 374/512 [42:02<15:29,  6.74s/it]
                                                 
{'loss': '0.0002039', 'grad_norm': '0.01487', 'learning_rate': '2.691e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '31.99', 'tokens/total': 11315600, 'tokens/trainable': 171126, 'epoch': '1.461'}
+
 73%|███████▎  | 374/512 [42:02<15:29,  6.74s/it]
 73%|███████▎  | 375/512 [42:09<15:19,  6.71s/it]
                                                 
{'loss': '0.001035', 'grad_norm': '0.05874', 'learning_rate': '2.668e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.6', 'tokens/total': 11345776, 'tokens/trainable': 171600, 'epoch': '1.465'}
+
 73%|███████▎  | 375/512 [42:09<15:19,  6.71s/it]
 73%|███████▎  | 376/512 [42:15<15:11,  6.70s/it]
                                                 
{'loss': '0.003572', 'grad_norm': '0.2411', 'learning_rate': '2.646e-05', 'ppl': '1.004', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '28.14', 'tokens/total': 11376208, 'tokens/trainable': 172017, 'epoch': '1.469'}
+
 73%|███████▎  | 376/512 [42:15<15:11,  6.70s/it]
 74%|███████▎  | 377/512 [42:22<15:05,  6.71s/it]
                                                 
{'loss': '0.0007833', 'grad_norm': '0.08794', 'learning_rate': '2.624e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '36.07', 'tokens/total': 11406512, 'tokens/trainable': 172493, 'epoch': '1.473'}
+
 74%|███████▎  | 377/512 [42:22<15:05,  6.71s/it]
 74%|███████▍  | 378/512 [42:29<14:57,  6.70s/it]
                                                 
{'loss': '0.0001175', 'grad_norm': '0.009155', 'learning_rate': '2.601e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.4', 'tokens/total': 11436544, 'tokens/trainable': 172951, 'epoch': '1.477'}
+
 74%|███████▍  | 378/512 [42:29<14:57,  6.70s/it]
 74%|███████▍  | 379/512 [42:36<14:50,  6.70s/it]
                                                 
{'loss': '0.0001157', 'grad_norm': '0.01292', 'learning_rate': '2.579e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.72', 'tokens/total': 11467008, 'tokens/trainable': 173435, 'epoch': '1.48'}
+
 74%|███████▍  | 379/512 [42:36<14:50,  6.70s/it]
 74%|███████▍  | 380/512 [42:42<14:44,  6.70s/it]
                                                 
{'loss': '8.939e-05', 'grad_norm': '0.008967', 'learning_rate': '2.557e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '30.47', 'tokens/total': 11497184, 'tokens/trainable': 173900, 'epoch': '1.484'}
+
 74%|███████▍  | 380/512 [42:42<14:44,  6.70s/it]
 74%|███████▍  | 381/512 [42:49<14:39,  6.71s/it]
                                                 
{'loss': '0.0001308', 'grad_norm': '0.03182', 'learning_rate': '2.535e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '34.13', 'tokens/total': 11527600, 'tokens/trainable': 174371, 'epoch': '1.488'}
+
 74%|███████▍  | 381/512 [42:49<14:39,  6.71s/it]
 75%|███████▍  | 382/512 [42:56<14:32,  6.71s/it]
                                                 
{'loss': '2.56e-05', 'grad_norm': '0.0009073', 'learning_rate': '2.513e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 11557808, 'tokens/trainable': 174826, 'epoch': '1.492'}
+
 75%|███████▍  | 382/512 [42:56<14:32,  6.71s/it]
 75%|███████▍  | 383/512 [43:02<14:26,  6.71s/it]
                                                 
{'loss': '5.779e-05', 'grad_norm': '0.00506', 'learning_rate': '2.492e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '32.88', 'tokens/total': 11588352, 'tokens/trainable': 175276, 'epoch': '1.496'}
+
 75%|███████▍  | 383/512 [43:02<14:26,  6.71s/it]
 75%|███████▌  | 384/512 [43:09<14:20,  6.72s/it]
                                                 
{'loss': '0.0001251', 'grad_norm': '0.008305', 'learning_rate': '2.47e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '33.71', 'tokens/total': 11618640, 'tokens/trainable': 175720, 'epoch': '1.5'}
+
 75%|███████▌  | 384/512 [43:09<14:20,  6.72s/it][2026-08-18 14:48:09,177] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-384
+
 75%|███████▌  | 385/512 [43:18<15:20,  7.25s/it]
                                                 
{'loss': '4.852e-05', 'grad_norm': '0.003528', 'learning_rate': '2.449e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.89', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 11649072, 'tokens/trainable': 176203, 'epoch': '1.504'}
+
 75%|███████▌  | 385/512 [43:18<15:20,  7.25s/it]
 75%|███████▌  | 386/512 [43:24<14:53,  7.09s/it]
                                                 
{'loss': '3.331e-05', 'grad_norm': '0.001593', 'learning_rate': '2.428e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.19', 'tokens/total': 11679552, 'tokens/trainable': 176651, 'epoch': '1.508'}
+
 75%|███████▌  | 386/512 [43:24<14:53,  7.09s/it]
 76%|███████▌  | 387/512 [43:31<14:32,  6.98s/it]
                                                 
{'loss': '3.345e-05', 'grad_norm': '0.002167', 'learning_rate': '2.406e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 11709760, 'tokens/trainable': 177115, 'epoch': '1.512'}
+
 76%|███████▌  | 387/512 [43:31<14:32,  6.98s/it]
 76%|███████▌  | 388/512 [43:38<14:15,  6.90s/it]
                                                 
{'loss': '5.528e-05', 'grad_norm': '0.003448', 'learning_rate': '2.385e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.13', 'tokens/total': 11740288, 'tokens/trainable': 177601, 'epoch': '1.516'}
+
 76%|███████▌  | 388/512 [43:38<14:15,  6.90s/it]
 76%|███████▌  | 389/512 [43:45<14:01,  6.84s/it]
                                                 
{'loss': '4.562e-05', 'grad_norm': '0.005298', 'learning_rate': '2.365e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 11770592, 'tokens/trainable': 178050, 'epoch': '1.52'}
+
 76%|███████▌  | 389/512 [43:45<14:01,  6.84s/it]
 76%|███████▌  | 390/512 [43:51<13:50,  6.81s/it]
                                                 
{'loss': '3.829e-05', 'grad_norm': '0.003032', 'learning_rate': '2.344e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.42', 'tokens/total': 11800752, 'tokens/trainable': 178531, 'epoch': '1.523'}
+
 76%|███████▌  | 390/512 [43:51<13:50,  6.81s/it]
 76%|███████▋  | 391/512 [43:58<13:41,  6.79s/it]
                                                 
{'loss': '0.0003605', 'grad_norm': '0.03543', 'learning_rate': '2.323e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.98', 'tokens/total': 11831360, 'tokens/trainable': 179000, 'epoch': '1.527'}
+
 76%|███████▋  | 391/512 [43:58<13:41,  6.79s/it]
 77%|███████▋  | 392/512 [44:05<13:29,  6.74s/it]
                                                 
{'loss': '7.091e-05', 'grad_norm': '0.004966', 'learning_rate': '2.303e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.1', 'tokens/total': 11861552, 'tokens/trainable': 179457, 'epoch': '1.531'}
+
 77%|███████▋  | 392/512 [44:05<13:29,  6.74s/it]
 77%|███████▋  | 393/512 [44:11<13:19,  6.71s/it]
                                                 
{'loss': '9.158e-05', 'grad_norm': '0.006339', 'learning_rate': '2.282e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 11891904, 'tokens/trainable': 179890, 'epoch': '1.535'}
+
 77%|███████▋  | 393/512 [44:11<13:19,  6.71s/it]
 77%|███████▋  | 394/512 [44:18<13:11,  6.71s/it]
                                                 
{'loss': '0.0003821', 'grad_norm': '0.03921', 'learning_rate': '2.262e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.95', 'tokens/total': 11922208, 'tokens/trainable': 180341, 'epoch': '1.539'}
+
 77%|███████▋  | 394/512 [44:18<13:11,  6.71s/it]
 77%|███████▋  | 395/512 [44:25<13:02,  6.69s/it]
                                                 
{'loss': '0.0006777', 'grad_norm': '0.08641', 'learning_rate': '2.242e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.99', 'tokens/total': 11952672, 'tokens/trainable': 180762, 'epoch': '1.543'}
+
 77%|███████▋  | 395/512 [44:25<13:02,  6.69s/it]
 77%|███████▋  | 396/512 [44:31<12:56,  6.69s/it]
                                                 
{'loss': '0.0007017', 'grad_norm': '0.08662', 'learning_rate': '2.222e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.18', 'tokens/total': 11983088, 'tokens/trainable': 181226, 'epoch': '1.547'}
+
 77%|███████▋  | 396/512 [44:31<12:56,  6.69s/it]
 78%|███████▊  | 397/512 [44:38<12:50,  6.70s/it]
                                                 
{'loss': '0.0001112', 'grad_norm': '0.01007', 'learning_rate': '2.202e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '33.18', 'tokens/total': 12013488, 'tokens/trainable': 181714, 'epoch': '1.551'}
+
 78%|███████▊  | 397/512 [44:38<12:50,  6.70s/it]
 78%|███████▊  | 398/512 [44:45<12:42,  6.69s/it]
                                                 
{'loss': '0.0002486', 'grad_norm': '0.03159', 'learning_rate': '2.183e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '35.57', 'tokens/total': 12043968, 'tokens/trainable': 182142, 'epoch': '1.555'}
+
 78%|███████▊  | 398/512 [44:45<12:42,  6.69s/it]
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
                                                 
{'loss': '2.122e-05', 'grad_norm': '0.0008732', 'learning_rate': '2.163e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '34.86', 'tokens/total': 12074160, 'tokens/trainable': 182604, 'epoch': '1.559'}
+
 78%|███████▊  | 399/512 [44:51<12:36,  6.70s/it]
 78%|███████▊  | 400/512 [44:58<12:29,  6.69s/it]
                                                 
{'loss': '0.0001349', 'grad_norm': '0.009256', 'learning_rate': '2.144e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '25.31', 'tokens/total': 12104352, 'tokens/trainable': 183011, 'epoch': '1.562'}
+
 78%|███████▊  | 400/512 [44:58<12:29,  6.69s/it]
 78%|███████▊  | 401/512 [45:05<12:22,  6.68s/it]
                                                 
{'loss': '7.793e-05', 'grad_norm': '0.01052', 'learning_rate': '2.124e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.71', 'memory/max_allocated (GiB)': '33.71', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.81', 'tokens/total': 12134240, 'tokens/trainable': 183459, 'epoch': '1.566'}
+
 78%|███████▊  | 401/512 [45:05<12:22,  6.68s/it]
 79%|███████▊  | 402/512 [45:12<12:17,  6.70s/it]
                                                 
{'loss': '2.772e-05', 'grad_norm': '0.001341', 'learning_rate': '2.105e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '38.7', 'tokens/total': 12164480, 'tokens/trainable': 183959, 'epoch': '1.57'}
+
 79%|███████▊  | 402/512 [45:12<12:17,  6.70s/it]
 79%|███████▊  | 403/512 [45:18<12:11,  6.71s/it]
                                                 
{'loss': '0.001414', 'grad_norm': '0.1036', 'learning_rate': '2.086e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '32.5', 'tokens/total': 12194640, 'tokens/trainable': 184431, 'epoch': '1.574'}
+
 79%|███████▊  | 403/512 [45:18<12:11,  6.71s/it]
 79%|███████▉  | 404/512 [45:25<12:05,  6.72s/it]
                                                 
{'loss': '4.145e-05', 'grad_norm': '0.002605', 'learning_rate': '2.067e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.37', 'tokens/total': 12224960, 'tokens/trainable': 184867, 'epoch': '1.578'}
+
 79%|███████▉  | 404/512 [45:25<12:05,  6.72s/it]
 79%|███████▉  | 405/512 [45:32<11:56,  6.70s/it]
                                                 
{'loss': '7.322e-05', 'grad_norm': '0.009412', 'learning_rate': '2.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.72', 'memory/max_allocated (GiB)': '33.72', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '29.37', 'tokens/total': 12255008, 'tokens/trainable': 185300, 'epoch': '1.582'}
+
 79%|███████▉  | 405/512 [45:32<11:56,  6.70s/it]
 79%|███████▉  | 406/512 [45:38<11:49,  6.70s/it]
                                                 
{'loss': '7.983e-05', 'grad_norm': '0.006792', 'learning_rate': '2.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '31.48', 'tokens/total': 12285248, 'tokens/trainable': 185754, 'epoch': '1.586'}
+
 79%|███████▉  | 406/512 [45:38<11:49,  6.70s/it]
 79%|███████▉  | 407/512 [45:45<11:41,  6.69s/it]
                                                 
{'loss': '3.743e-05', 'grad_norm': '0.01302', 'learning_rate': '2.012e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.88', 'tokens/train_per_sec_per_gpu': '36.82', 'tokens/total': 12315552, 'tokens/trainable': 186213, 'epoch': '1.59'}
+
 79%|███████▉  | 407/512 [45:45<11:41,  6.69s/it]
 80%|███████▉  | 408/512 [45:52<11:36,  6.69s/it]
                                                 
{'loss': '2.307e-05', 'grad_norm': '0.001309', 'learning_rate': '1.993e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 12345904, 'tokens/trainable': 186659, 'epoch': '1.594'}
+
 80%|███████▉  | 408/512 [45:52<11:36,  6.69s/it]
 80%|███████▉  | 409/512 [45:58<11:29,  6.70s/it]
                                                 
{'loss': '2.135e-05', 'grad_norm': '0.001227', 'learning_rate': '1.975e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '32.97', 'tokens/total': 12376336, 'tokens/trainable': 187083, 'epoch': '1.598'}
+
 80%|███████▉  | 409/512 [45:58<11:29,  6.70s/it]
 80%|████████  | 410/512 [46:05<11:23,  6.70s/it]
                                                 
{'loss': '3.752e-05', 'grad_norm': '0.01123', 'learning_rate': '1.957e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 12406880, 'tokens/trainable': 187530, 'epoch': '1.602'}
+
 80%|████████  | 410/512 [46:05<11:23,  6.70s/it]
 80%|████████  | 411/512 [46:12<11:16,  6.70s/it]
                                                 
{'loss': '5.565e-05', 'grad_norm': '0.004873', 'learning_rate': '1.94e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.32', 'tokens/total': 12437088, 'tokens/trainable': 187972, 'epoch': '1.605'}
+
 80%|████████  | 411/512 [46:12<11:16,  6.70s/it]
 80%|████████  | 412/512 [46:18<10:53,  6.54s/it]
                                                 
{'loss': '0.0008922', 'grad_norm': '0.1939', 'learning_rate': '1.922e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '39.95', 'tokens/total': 12465440, 'tokens/trainable': 188417, 'epoch': '1.609'}
+
 80%|████████  | 412/512 [46:18<10:53,  6.54s/it]
 81%|████████  | 413/512 [46:25<10:53,  6.60s/it]
                                                 
{'loss': '0.0006304', 'grad_norm': '0.09062', 'learning_rate': '1.904e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.57', 'tokens/total': 12495968, 'tokens/trainable': 188868, 'epoch': '1.613'}
+
 81%|████████  | 413/512 [46:25<10:53,  6.60s/it]
 81%|████████  | 414/512 [46:31<10:50,  6.64s/it]
                                                 
{'loss': '6.724e-05', 'grad_norm': '0.009725', 'learning_rate': '1.887e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '36.95', 'tokens/total': 12526128, 'tokens/trainable': 189337, 'epoch': '1.617'}
+
 81%|████████  | 414/512 [46:31<10:50,  6.64s/it]
 81%|████████  | 415/512 [46:38<10:43,  6.64s/it]
                                                 
{'loss': '0.001907', 'grad_norm': '0.1954', 'learning_rate': '1.87e-05', 'ppl': '1.002', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '31.94', 'tokens/total': 12556256, 'tokens/trainable': 189750, 'epoch': '1.621'}
+
 81%|████████  | 415/512 [46:38<10:43,  6.64s/it]
 81%|████████▏ | 416/512 [46:45<10:39,  6.66s/it]
                                                 
{'loss': '0.0005661', 'grad_norm': '0.04653', 'learning_rate': '1.853e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '34.06', 'tokens/total': 12586736, 'tokens/trainable': 190216, 'epoch': '1.625'}
+
 81%|████████▏ | 416/512 [46:45<10:39,  6.66s/it][2026-08-18 14:51:44,795] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-416
+
 81%|████████▏ | 417/512 [46:53<11:22,  7.19s/it]
                                                 
{'loss': '1.887e-05', 'grad_norm': '0.001295', 'learning_rate': '1.836e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.96', 'tokens/train_per_sec_per_gpu': '35.01', 'tokens/total': 12617168, 'tokens/trainable': 190708, 'epoch': '1.629'}
+
 81%|████████▏ | 417/512 [46:53<11:22,  7.19s/it]
 82%|████████▏ | 418/512 [47:00<10:59,  7.02s/it]
                                                 
{'loss': '3.221e-05', 'grad_norm': '0.001875', 'learning_rate': '1.819e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.9', 'tokens/total': 12647680, 'tokens/trainable': 191126, 'epoch': '1.633'}
+
 82%|████████▏ | 418/512 [47:00<10:59,  7.02s/it]
 82%|████████▏ | 419/512 [47:06<10:42,  6.91s/it]
                                                 
{'loss': '2.656e-05', 'grad_norm': '0.002248', 'learning_rate': '1.802e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '30.34', 'tokens/total': 12677952, 'tokens/trainable': 191540, 'epoch': '1.637'}
+
 82%|████████▏ | 419/512 [47:06<10:42,  6.91s/it]
 82%|████████▏ | 420/512 [47:13<10:29,  6.84s/it]
                                                 
{'loss': '0.0009822', 'grad_norm': '0.1634', 'learning_rate': '1.786e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '36.69', 'tokens/total': 12708416, 'tokens/trainable': 192014, 'epoch': '1.641'}
+
 82%|████████▏ | 420/512 [47:13<10:29,  6.84s/it]
 82%|████████▏ | 421/512 [47:20<10:19,  6.80s/it]
                                                 
{'loss': '2.167e-05', 'grad_norm': '0.00194', 'learning_rate': '1.77e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.8', 'tokens/train_per_sec_per_gpu': '31.31', 'tokens/total': 12738640, 'tokens/trainable': 192463, 'epoch': '1.645'}
+
 82%|████████▏ | 421/512 [47:20<10:19,  6.80s/it]
 82%|████████▏ | 422/512 [47:27<10:09,  6.77s/it]
                                                 
{'loss': '1.654e-05', 'grad_norm': '0.00106', 'learning_rate': '1.753e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.74', 'tokens/total': 12769232, 'tokens/trainable': 192918, 'epoch': '1.648'}
+
 82%|████████▏ | 422/512 [47:27<10:09,  6.77s/it]
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
                                                 
{'loss': '0.0001707', 'grad_norm': '0.0176', 'learning_rate': '1.737e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '37.46', 'tokens/total': 12799648, 'tokens/trainable': 193425, 'epoch': '1.652'}
+
 83%|████████▎ | 423/512 [47:33<10:01,  6.76s/it]
 83%|████████▎ | 424/512 [47:40<09:53,  6.75s/it]
                                                 
{'loss': '2.332e-05', 'grad_norm': '0.002366', 'learning_rate': '1.722e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.25', 'tokens/total': 12829920, 'tokens/trainable': 193898, 'epoch': '1.656'}
+
 83%|████████▎ | 424/512 [47:40<09:53,  6.75s/it]
 83%|████████▎ | 425/512 [47:47<09:45,  6.73s/it]
                                                 
{'loss': '0.0001577', 'grad_norm': '0.02454', 'learning_rate': '1.706e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.29', 'tokens/total': 12860112, 'tokens/trainable': 194315, 'epoch': '1.66'}
+
 83%|████████▎ | 425/512 [47:47<09:45,  6.73s/it]
 83%|████████▎ | 426/512 [47:53<09:38,  6.73s/it]
                                                 
{'loss': '1.428e-05', 'grad_norm': '0.0009002', 'learning_rate': '1.69e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.92', 'tokens/total': 12890320, 'tokens/trainable': 194796, 'epoch': '1.664'}
+
 83%|████████▎ | 426/512 [47:53<09:38,  6.73s/it]
 83%|████████▎ | 427/512 [48:00<09:32,  6.73s/it]
                                                 
{'loss': '7.186e-06', 'grad_norm': '0.00034', 'learning_rate': '1.675e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.96', 'tokens/total': 12920656, 'tokens/trainable': 195262, 'epoch': '1.668'}
+
 83%|████████▎ | 427/512 [48:00<09:32,  6.73s/it]
 84%|████████▎ | 428/512 [48:07<09:25,  6.73s/it]
                                                 
{'loss': '9.948e-06', 'grad_norm': '0.000536', 'learning_rate': '1.66e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.91', 'tokens/total': 12951040, 'tokens/trainable': 195680, 'epoch': '1.672'}
+
 84%|████████▎ | 428/512 [48:07<09:25,  6.73s/it]
 84%|████████▍ | 429/512 [48:14<09:16,  6.71s/it]
                                                 
{'loss': '2.029e-05', 'grad_norm': '0.001489', 'learning_rate': '1.645e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.75', 'memory/max_allocated (GiB)': '33.75', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.89', 'tokens/total': 12981232, 'tokens/trainable': 196145, 'epoch': '1.676'}
+
 84%|████████▍ | 429/512 [48:14<09:16,  6.71s/it]
 84%|████████▍ | 430/512 [48:20<09:10,  6.71s/it]
                                                 
{'loss': '0.0003293', 'grad_norm': '0.05131', 'learning_rate': '1.63e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.42', 'tokens/total': 13011616, 'tokens/trainable': 196626, 'epoch': '1.68'}
+
 84%|████████▍ | 430/512 [48:20<09:10,  6.71s/it]
 84%|████████▍ | 431/512 [48:27<09:06,  6.75s/it]
                                                 
{'loss': '1.015e-05', 'grad_norm': '0.001407', 'learning_rate': '1.615e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '31.17', 'tokens/total': 13042208, 'tokens/trainable': 197067, 'epoch': '1.684'}
+
 84%|████████▍ | 431/512 [48:27<09:06,  6.75s/it]
 84%|████████▍ | 432/512 [48:34<08:59,  6.74s/it]
                                                 
{'loss': '3.515e-05', 'grad_norm': '0.004116', 'learning_rate': '1.6e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.35', 'tokens/train_per_sec_per_gpu': '38.68', 'tokens/total': 13072800, 'tokens/trainable': 197565, 'epoch': '1.688'}
+
 84%|████████▍ | 432/512 [48:34<08:59,  6.74s/it]
 85%|████████▍ | 433/512 [48:41<08:51,  6.72s/it]
                                                 
{'loss': '7.314e-05', 'grad_norm': '0.01148', 'learning_rate': '1.586e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.95', 'memory/max_allocated (GiB)': '33.95', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.96', 'tokens/total': 13103328, 'tokens/trainable': 198015, 'epoch': '1.691'}
+
 85%|████████▍ | 433/512 [48:41<08:51,  6.72s/it]
 85%|████████▍ | 434/512 [48:47<08:43,  6.71s/it]
                                                 
{'loss': '0.0002449', 'grad_norm': '0.03078', 'learning_rate': '1.572e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 13133472, 'tokens/trainable': 198442, 'epoch': '1.695'}
+
 85%|████████▍ | 434/512 [48:47<08:43,  6.71s/it]
 85%|████████▍ | 435/512 [48:54<08:37,  6.72s/it]
                                                 
{'loss': '1.445e-05', 'grad_norm': '0.002246', 'learning_rate': '1.558e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 13164064, 'tokens/trainable': 198914, 'epoch': '1.699'}
+
 85%|████████▍ | 435/512 [48:54<08:37,  6.72s/it]
 85%|████████▌ | 436/512 [49:01<08:30,  6.72s/it]
                                                 
{'loss': '2.742e-05', 'grad_norm': '0.002974', 'learning_rate': '1.544e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.45', 'tokens/total': 13194464, 'tokens/trainable': 199355, 'epoch': '1.703'}
+
 85%|████████▌ | 436/512 [49:01<08:30,  6.72s/it]
 85%|████████▌ | 437/512 [49:07<08:23,  6.71s/it]
                                                 
{'loss': '2.773e-05', 'grad_norm': '0.002261', 'learning_rate': '1.53e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33', 'tokens/total': 13224992, 'tokens/trainable': 199830, 'epoch': '1.707'}
+
 85%|████████▌ | 437/512 [49:07<08:23,  6.71s/it]
 86%|████████▌ | 438/512 [49:14<08:17,  6.72s/it]
                                                 
{'loss': '9.154e-05', 'grad_norm': '0.01199', 'learning_rate': '1.516e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.97', 'tokens/total': 13255296, 'tokens/trainable': 200338, 'epoch': '1.711'}
+
 86%|████████▌ | 438/512 [49:14<08:17,  6.72s/it]
 86%|████████▌ | 439/512 [49:21<08:08,  6.70s/it]
                                                 
{'loss': '3.515e-05', 'grad_norm': '0.003149', 'learning_rate': '1.503e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '32.87', 'tokens/total': 13285344, 'tokens/trainable': 200785, 'epoch': '1.715'}
+
 86%|████████▌ | 439/512 [49:21<08:08,  6.70s/it]
 86%|████████▌ | 440/512 [49:27<08:03,  6.71s/it]
                                                 
{'loss': '7.169e-06', 'grad_norm': '0.0003243', 'learning_rate': '1.49e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.32', 'tokens/total': 13315632, 'tokens/trainable': 201267, 'epoch': '1.719'}
+
 86%|████████▌ | 440/512 [49:27<08:03,  6.71s/it]
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
                                                 
{'loss': '0.0007142', 'grad_norm': '0.1277', 'learning_rate': '1.477e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.22', 'tokens/total': 13346048, 'tokens/trainable': 201734, 'epoch': '1.723'}
+
 86%|████████▌ | 441/512 [49:34<07:56,  6.72s/it]
 86%|████████▋ | 442/512 [49:41<07:49,  6.71s/it]
                                                 
{'loss': '0.0006704', 'grad_norm': '0.1148', 'learning_rate': '1.464e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.4', 'tokens/total': 13376304, 'tokens/trainable': 202174, 'epoch': '1.727'}
+
 86%|████████▋ | 442/512 [49:41<07:49,  6.71s/it]
 87%|████████▋ | 443/512 [49:48<07:42,  6.71s/it]
                                                 
{'loss': '9.368e-06', 'grad_norm': '0.0003201', 'learning_rate': '1.451e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '36.28', 'tokens/total': 13406720, 'tokens/trainable': 202612, 'epoch': '1.73'}
+
 87%|████████▋ | 443/512 [49:48<07:42,  6.71s/it]
 87%|████████▋ | 444/512 [49:54<07:33,  6.66s/it]
                                                 
{'loss': '5.53e-06', 'grad_norm': '0.000266', 'learning_rate': '1.438e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.64', 'memory/max_allocated (GiB)': '33.64', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '30.99', 'tokens/total': 13436464, 'tokens/trainable': 203020, 'epoch': '1.734'}
+
 87%|████████▋ | 444/512 [49:54<07:33,  6.66s/it]
 87%|████████▋ | 445/512 [50:01<07:26,  6.66s/it]
                                                 
{'loss': '6.467e-05', 'grad_norm': '0.04957', 'learning_rate': '1.426e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.37', 'tokens/total': 13466672, 'tokens/trainable': 203479, 'epoch': '1.738'}
+
 87%|████████▋ | 445/512 [50:01<07:26,  6.66s/it]
 87%|████████▋ | 446/512 [50:08<07:20,  6.67s/it]
                                                 
{'loss': '1.261e-05', 'grad_norm': '0.001684', 'learning_rate': '1.414e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '35.38', 'tokens/total': 13496736, 'tokens/trainable': 203946, 'epoch': '1.742'}
+
 87%|████████▋ | 446/512 [50:08<07:20,  6.67s/it]
 87%|████████▋ | 447/512 [50:14<07:13,  6.68s/it]
                                                 
{'loss': '7.416e-06', 'grad_norm': '0.0004292', 'learning_rate': '1.402e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '37.45', 'tokens/total': 13526928, 'tokens/trainable': 204382, 'epoch': '1.746'}
+
 87%|████████▋ | 447/512 [50:14<07:13,  6.68s/it]
 88%|████████▊ | 448/512 [50:21<07:07,  6.68s/it]
                                                 
{'loss': '0.0006914', 'grad_norm': '0.2166', 'learning_rate': '1.39e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '34.35', 'tokens/total': 13557392, 'tokens/trainable': 204830, 'epoch': '1.75'}
+
 88%|████████▊ | 448/512 [50:21<07:07,  6.68s/it][2026-08-18 14:55:20,925] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-448
+
 88%|████████▊ | 449/512 [50:29<07:33,  7.20s/it]
                                                 
{'loss': '1.45e-05', 'grad_norm': '0.0009793', 'learning_rate': '1.378e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.93', 'memory/max_allocated (GiB)': '33.93', 'memory/device_reserved (GiB)': '36.8', 'tokens/train_per_sec_per_gpu': '33.47', 'tokens/total': 13587776, 'tokens/trainable': 205300, 'epoch': '1.754'}
+
 88%|████████▊ | 449/512 [50:29<07:33,  7.20s/it]
 88%|████████▊ | 450/512 [50:36<07:17,  7.06s/it]
                                                 
{'loss': '0.0002816', 'grad_norm': '0.0285', 'learning_rate': '1.367e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '30.6', 'tokens/total': 13618112, 'tokens/trainable': 205726, 'epoch': '1.758'}
+
 88%|████████▊ | 450/512 [50:36<07:17,  7.06s/it]
 88%|████████▊ | 451/512 [50:43<07:04,  6.96s/it]
                                                 
{'loss': '2.595e-05', 'grad_norm': '0.003279', 'learning_rate': '1.355e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '31.63', 'tokens/total': 13648512, 'tokens/trainable': 206163, 'epoch': '1.762'}
+
 88%|████████▊ | 451/512 [50:43<07:04,  6.96s/it]
 88%|████████▊ | 452/512 [50:49<06:52,  6.88s/it]
                                                 
{'loss': '0.004595', 'grad_norm': '0.4962', 'learning_rate': '1.344e-05', 'ppl': '1.005', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.45', 'tokens/total': 13678704, 'tokens/trainable': 206619, 'epoch': '1.766'}
+
 88%|████████▊ | 452/512 [50:49<06:52,  6.88s/it]
 88%|████████▊ | 453/512 [50:56<06:43,  6.85s/it]
                                                 
{'loss': '6.578e-05', 'grad_norm': '0.009086', 'learning_rate': '1.333e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '37.24', 'tokens/total': 13709232, 'tokens/trainable': 207101, 'epoch': '1.77'}
+
 88%|████████▊ | 453/512 [50:56<06:43,  6.85s/it]
 89%|████████▊ | 454/512 [51:03<06:34,  6.80s/it]
                                                 
{'loss': '6.735e-05', 'grad_norm': '0.009036', 'learning_rate': '1.322e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.86', 'memory/max_allocated (GiB)': '33.86', 'memory/device_reserved (GiB)': '35.7', 'tokens/train_per_sec_per_gpu': '35.77', 'tokens/total': 13739600, 'tokens/trainable': 207546, 'epoch': '1.773'}
+
 89%|████████▊ | 454/512 [51:03<06:34,  6.80s/it]
 89%|████████▉ | 455/512 [51:10<06:26,  6.77s/it]
                                                 
{'loss': '7.266e-05', 'grad_norm': '0.009665', 'learning_rate': '1.311e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '34.26', 'tokens/total': 13769936, 'tokens/trainable': 207989, 'epoch': '1.777'}
+
 89%|████████▉ | 455/512 [51:10<06:26,  6.77s/it]
 89%|████████▉ | 456/512 [51:16<06:17,  6.74s/it]
                                                 
{'loss': '6.039e-06', 'grad_norm': '0.0004666', 'learning_rate': '1.301e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '35.78', 'tokens/train_per_sec_per_gpu': '33.92', 'tokens/total': 13800240, 'tokens/trainable': 208402, 'epoch': '1.781'}
+
 89%|████████▉ | 456/512 [51:16<06:17,  6.74s/it]
 89%|████████▉ | 457/512 [51:23<06:10,  6.74s/it]
                                                 
{'loss': '8.293e-06', 'grad_norm': '0.0004485', 'learning_rate': '1.29e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.58', 'tokens/total': 13830512, 'tokens/trainable': 208904, 'epoch': '1.785'}
+
 89%|████████▉ | 457/512 [51:23<06:10,  6.74s/it]
 89%|████████▉ | 458/512 [51:30<06:03,  6.73s/it]
                                                 
{'loss': '2.316e-05', 'grad_norm': '0.003369', 'learning_rate': '1.28e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.72', 'tokens/total': 13860832, 'tokens/trainable': 209382, 'epoch': '1.789'}
+
 89%|████████▉ | 458/512 [51:30<06:03,  6.73s/it]
 90%|████████▉ | 459/512 [51:36<05:55,  6.71s/it]
                                                 
{'loss': '1.119e-05', 'grad_norm': '0.00108', 'learning_rate': '1.27e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 13890992, 'tokens/trainable': 209833, 'epoch': '1.793'}
+
 90%|████████▉ | 459/512 [51:36<05:55,  6.71s/it]
 90%|████████▉ | 460/512 [51:43<05:48,  6.70s/it]
                                                 
{'loss': '0.0003375', 'grad_norm': '0.04659', 'learning_rate': '1.26e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.91', 'memory/max_allocated (GiB)': '33.91', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.53', 'tokens/total': 13921424, 'tokens/trainable': 210284, 'epoch': '1.797'}
+
 90%|████████▉ | 460/512 [51:43<05:48,  6.70s/it]
 90%|█████████ | 461/512 [51:50<05:40,  6.68s/it]
                                                 
{'loss': '0.0001611', 'grad_norm': '0.01848', 'learning_rate': '1.251e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.66', 'memory/max_allocated (GiB)': '33.66', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '34.07', 'tokens/total': 13951504, 'tokens/trainable': 210769, 'epoch': '1.801'}
+
 90%|█████████ | 461/512 [51:50<05:40,  6.68s/it]
 90%|█████████ | 462/512 [51:56<05:26,  6.53s/it]
                                                 
{'loss': '0.0001783', 'grad_norm': '0.05658', 'learning_rate': '1.241e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.38', 'memory/max_allocated (GiB)': '33.38', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '30.41', 'tokens/total': 13979728, 'tokens/trainable': 211193, 'epoch': '1.805'}
+
 90%|█████████ | 462/512 [51:56<05:26,  6.53s/it]
 90%|█████████ | 463/512 [52:03<05:22,  6.59s/it]
                                                 
{'loss': '9.709e-06', 'grad_norm': '0.0006595', 'learning_rate': '1.232e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.79', 'memory/max_allocated (GiB)': '33.79', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.11', 'tokens/total': 14009936, 'tokens/trainable': 211655, 'epoch': '1.809'}
+
 90%|█████████ | 463/512 [52:03<05:22,  6.59s/it]
 91%|█████████ | 464/512 [52:09<05:17,  6.61s/it]
                                                 
{'loss': '0.0001757', 'grad_norm': '0.02738', 'learning_rate': '1.223e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.74', 'memory/max_allocated (GiB)': '33.74', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '29.2', 'tokens/total': 14039872, 'tokens/trainable': 212085, 'epoch': '1.812'}
+
 91%|█████████ | 464/512 [52:09<05:17,  6.61s/it]
 91%|█████████ | 465/512 [52:16<05:12,  6.64s/it]
                                                 
{'loss': '8.168e-06', 'grad_norm': '0.0004917', 'learning_rate': '1.214e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.71', 'tokens/total': 14070096, 'tokens/trainable': 212570, 'epoch': '1.816'}
+
 91%|█████████ | 465/512 [52:16<05:12,  6.64s/it]
 91%|█████████ | 466/512 [52:23<05:06,  6.65s/it]
                                                 
{'loss': '1.287e-05', 'grad_norm': '0.001592', 'learning_rate': '1.205e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '32.93', 'tokens/total': 14100192, 'tokens/trainable': 213011, 'epoch': '1.82'}
+
 91%|█████████ | 466/512 [52:23<05:06,  6.65s/it]
 91%|█████████ | 467/512 [52:29<05:00,  6.67s/it]
                                                 
{'loss': '7.528e-06', 'grad_norm': '0.000542', 'learning_rate': '1.197e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '31.58', 'tokens/total': 14130448, 'tokens/trainable': 213450, 'epoch': '1.824'}
+
 91%|█████████ | 467/512 [52:29<05:00,  6.67s/it]
 91%|█████████▏| 468/512 [52:36<04:53,  6.66s/it]
                                                 
{'loss': '4.258e-05', 'grad_norm': '0.0045', 'learning_rate': '1.188e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.73', 'memory/max_allocated (GiB)': '33.73', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.7', 'tokens/total': 14160480, 'tokens/trainable': 213898, 'epoch': '1.828'}
+
 91%|█████████▏| 468/512 [52:36<04:53,  6.66s/it]
 92%|█████████▏| 469/512 [52:43<04:46,  6.67s/it]
                                                 
{'loss': '0.000211', 'grad_norm': '0.1103', 'learning_rate': '1.18e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '36.88', 'tokens/total': 14191152, 'tokens/trainable': 214366, 'epoch': '1.832'}
+
 92%|█████████▏| 469/512 [52:43<04:46,  6.67s/it]
 92%|█████████▏| 470/512 [52:49<04:34,  6.54s/it]
                                                 
{'loss': '2.79e-05', 'grad_norm': '0.002098', 'learning_rate': '1.172e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.82', 'tokens/train_per_sec_per_gpu': '35.72', 'tokens/total': 14219392, 'tokens/trainable': 214800, 'epoch': '1.836'}
+
 92%|█████████▏| 470/512 [52:49<04:34,  6.54s/it]
 92%|█████████▏| 471/512 [52:56<04:31,  6.61s/it]
                                                 
{'loss': '1.734e-05', 'grad_norm': '0.001368', 'learning_rate': '1.164e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '35.48', 'tokens/total': 14249920, 'tokens/trainable': 215292, 'epoch': '1.84'}
+
 92%|█████████▏| 471/512 [52:56<04:31,  6.61s/it]
 92%|█████████▏| 472/512 [53:02<04:25,  6.64s/it]
                                                 
{'loss': '1.396e-05', 'grad_norm': '0.0008574', 'learning_rate': '1.156e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.9', 'tokens/total': 14280048, 'tokens/trainable': 215741, 'epoch': '1.844'}
+
 92%|█████████▏| 472/512 [53:02<04:25,  6.64s/it]
 92%|█████████▏| 473/512 [53:09<04:20,  6.67s/it]
                                                 
{'loss': '3.098e-05', 'grad_norm': '0.003767', 'learning_rate': '1.149e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.68', 'tokens/total': 14310432, 'tokens/trainable': 216194, 'epoch': '1.848'}
+
 92%|█████████▏| 473/512 [53:09<04:20,  6.67s/it]
 93%|█████████▎| 474/512 [53:16<04:13,  6.67s/it]
                                                 
{'loss': '6.771e-06', 'grad_norm': '0.0003345', 'learning_rate': '1.142e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '33.33', 'tokens/total': 14340496, 'tokens/trainable': 216634, 'epoch': '1.852'}
+
 93%|█████████▎| 474/512 [53:16<04:13,  6.67s/it]
 93%|█████████▎| 475/512 [53:23<04:06,  6.67s/it]
                                                 
{'loss': '7.348e-06', 'grad_norm': '0.0003811', 'learning_rate': '1.135e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '30.53', 'tokens/total': 14370720, 'tokens/trainable': 217040, 'epoch': '1.855'}
+
 93%|█████████▎| 475/512 [53:23<04:06,  6.67s/it]
 93%|█████████▎| 476/512 [53:29<04:01,  6.70s/it]
                                                 
{'loss': '1.269e-05', 'grad_norm': '0.001393', 'learning_rate': '1.128e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '37.67', 'tokens/total': 14401136, 'tokens/trainable': 217530, 'epoch': '1.859'}
+
 93%|█████████▎| 476/512 [53:29<04:01,  6.70s/it]
 93%|█████████▎| 477/512 [53:36<03:53,  6.69s/it]
                                                 
{'loss': '0.0003119', 'grad_norm': '0.02565', 'learning_rate': '1.121e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.53', 'tokens/total': 14431360, 'tokens/trainable': 217988, 'epoch': '1.863'}
+
 93%|█████████▎| 477/512 [53:36<03:53,  6.69s/it]
 93%|█████████▎| 478/512 [53:43<03:48,  6.73s/it]
                                                 
{'loss': '0.0003999', 'grad_norm': '0.08054', 'learning_rate': '1.114e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.54', 'tokens/total': 14461968, 'tokens/trainable': 218461, 'epoch': '1.867'}
+
 93%|█████████▎| 478/512 [53:43<03:48,  6.73s/it]
 94%|█████████▎| 479/512 [53:49<03:41,  6.70s/it]
                                                 
{'loss': '0.000334', 'grad_norm': '0.05217', 'learning_rate': '1.108e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.83', 'memory/max_allocated (GiB)': '33.83', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '32.81', 'tokens/total': 14491984, 'tokens/trainable': 218909, 'epoch': '1.871'}
+
 94%|█████████▎| 479/512 [53:49<03:41,  6.70s/it]
 94%|█████████▍| 480/512 [53:56<03:34,  6.71s/it]
                                                 
{'loss': '0.0005761', 'grad_norm': '0.08784', 'learning_rate': '1.102e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.98', 'memory/max_allocated (GiB)': '33.98', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '34.08', 'tokens/total': 14522544, 'tokens/trainable': 219350, 'epoch': '1.875'}
+
 94%|█████████▍| 480/512 [53:56<03:34,  6.71s/it][2026-08-18 14:58:56,154] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-480
+
 94%|█████████▍| 481/512 [54:04<03:42,  7.18s/it]
                                                 
{'loss': '0.0007811', 'grad_norm': '0.106', 'learning_rate': '1.096e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.78', 'memory/max_allocated (GiB)': '33.78', 'memory/device_reserved (GiB)': '35.92', 'tokens/train_per_sec_per_gpu': '29.52', 'tokens/total': 14552560, 'tokens/trainable': 219776, 'epoch': '1.879'}
+
 94%|█████████▍| 481/512 [54:04<03:42,  7.18s/it]
 94%|█████████▍| 482/512 [54:11<03:31,  7.06s/it]
                                                 
{'loss': '5.21e-06', 'grad_norm': '0.0002076', 'learning_rate': '1.09e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.15', 'tokens/train_per_sec_per_gpu': '35.06', 'tokens/total': 14583056, 'tokens/trainable': 220284, 'epoch': '1.883'}
+
 94%|█████████▍| 482/512 [54:11<03:31,  7.06s/it]
 94%|█████████▍| 483/512 [54:18<03:21,  6.94s/it]
                                                 
{'loss': '1.686e-05', 'grad_norm': '0.002103', 'learning_rate': '1.084e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.8', 'memory/max_allocated (GiB)': '33.8', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.06', 'tokens/total': 14613152, 'tokens/trainable': 220714, 'epoch': '1.887'}
+
 94%|█████████▍| 483/512 [54:18<03:21,  6.94s/it]
 95%|█████████▍| 484/512 [54:25<03:12,  6.86s/it]
                                                 
{'loss': '9.006e-05', 'grad_norm': '0.009665', 'learning_rate': '1.079e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.02', 'tokens/total': 14643328, 'tokens/trainable': 221156, 'epoch': '1.891'}
+
 95%|█████████▍| 484/512 [54:25<03:12,  6.86s/it]
 95%|█████████▍| 485/512 [54:31<03:03,  6.80s/it]
                                                 
{'loss': '1.069e-05', 'grad_norm': '0.001844', 'learning_rate': '1.073e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.64', 'tokens/total': 14673488, 'tokens/trainable': 221605, 'epoch': '1.895'}
+
 95%|█████████▍| 485/512 [54:31<03:03,  6.80s/it]
 95%|█████████▍| 486/512 [54:38<02:55,  6.76s/it]
                                                 
{'loss': '5.547e-06', 'grad_norm': '0.0007748', 'learning_rate': '1.068e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.46', 'tokens/total': 14704000, 'tokens/trainable': 222042, 'epoch': '1.898'}
+
 95%|█████████▍| 486/512 [54:38<02:55,  6.76s/it]
 95%|█████████▌| 487/512 [54:45<02:48,  6.73s/it]
                                                 
{'loss': '6.492e-06', 'grad_norm': '0.0002923', 'learning_rate': '1.063e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.67', 'memory/max_allocated (GiB)': '33.67', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.03', 'tokens/total': 14733984, 'tokens/trainable': 222474, 'epoch': '1.902'}
+
 95%|█████████▌| 487/512 [54:45<02:48,  6.73s/it]
 95%|█████████▌| 488/512 [54:51<02:41,  6.73s/it]
                                                 
{'loss': '6.497e-06', 'grad_norm': '0.0003241', 'learning_rate': '1.058e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.27', 'tokens/total': 14764400, 'tokens/trainable': 222967, 'epoch': '1.906'}
+
 95%|█████████▌| 488/512 [54:51<02:41,  6.73s/it]
 96%|█████████▌| 489/512 [54:58<02:34,  6.71s/it]
                                                 
{'loss': '1.286e-05', 'grad_norm': '0.001957', 'learning_rate': '1.054e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.77', 'memory/max_allocated (GiB)': '33.77', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '31.79', 'tokens/total': 14794672, 'tokens/trainable': 223420, 'epoch': '1.91'}
+
 96%|█████████▌| 489/512 [54:58<02:34,  6.71s/it]
 96%|█████████▌| 490/512 [55:05<02:27,  6.70s/it]
                                                 
{'loss': '4.987e-06', 'grad_norm': '0.0006595', 'learning_rate': '1.049e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.76', 'memory/max_allocated (GiB)': '33.76', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '29.4', 'tokens/total': 14824784, 'tokens/trainable': 223864, 'epoch': '1.914'}
+
 96%|█████████▌| 490/512 [55:05<02:27,  6.70s/it]
 96%|█████████▌| 491/512 [55:11<02:20,  6.71s/it]
                                                 
{'loss': '3.641e-05', 'grad_norm': '0.005793', 'learning_rate': '1.045e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.8', 'tokens/total': 14855056, 'tokens/trainable': 224321, 'epoch': '1.918'}
+
 96%|█████████▌| 491/512 [55:11<02:20,  6.71s/it]
 96%|█████████▌| 492/512 [55:18<02:14,  6.72s/it]
                                                 
{'loss': '9.24e-05', 'grad_norm': '0.0105', 'learning_rate': '1.041e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.01', 'memory/max_allocated (GiB)': '34.01', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.34', 'tokens/total': 14885392, 'tokens/trainable': 224802, 'epoch': '1.922'}
+
 96%|█████████▌| 492/512 [55:18<02:14,  6.72s/it]
 96%|█████████▋| 493/512 [55:25<02:07,  6.73s/it]
                                                 
{'loss': '6.912e-06', 'grad_norm': '0.0007256', 'learning_rate': '1.037e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.78', 'tokens/total': 14915760, 'tokens/trainable': 225284, 'epoch': '1.926'}
+
 96%|█████████▋| 493/512 [55:25<02:07,  6.73s/it]
 96%|█████████▋| 494/512 [55:32<02:01,  6.74s/it]
                                                 
{'loss': '9.241e-06', 'grad_norm': '0.001594', 'learning_rate': '1.034e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.84', 'memory/max_allocated (GiB)': '33.84', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 14946192, 'tokens/trainable': 225733, 'epoch': '1.93'}
+
 96%|█████████▋| 494/512 [55:32<02:01,  6.74s/it]
 97%|█████████▋| 495/512 [55:38<01:54,  6.71s/it]
                                                 
{'loss': '6.262e-06', 'grad_norm': '0.0005452', 'learning_rate': '1.03e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.81', 'memory/max_allocated (GiB)': '33.81', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.58', 'tokens/total': 14976272, 'tokens/trainable': 226172, 'epoch': '1.934'}
+
 97%|█████████▋| 495/512 [55:38<01:54,  6.71s/it]
 97%|█████████▋| 496/512 [55:45<01:47,  6.72s/it]
                                                 
{'loss': '1.628e-05', 'grad_norm': '0.001487', 'learning_rate': '1.027e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.9', 'memory/max_allocated (GiB)': '33.9', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.85', 'tokens/total': 15006544, 'tokens/trainable': 226636, 'epoch': '1.938'}
+
 97%|█████████▋| 496/512 [55:45<01:47,  6.72s/it]
 97%|█████████▋| 497/512 [55:52<01:40,  6.71s/it]
                                                 
{'loss': '8.379e-05', 'grad_norm': '0.01586', 'learning_rate': '1.024e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.95', 'tokens/total': 15037040, 'tokens/trainable': 227067, 'epoch': '1.941'}
+
 97%|█████████▋| 497/512 [55:52<01:40,  6.71s/it]
 97%|█████████▋| 498/512 [55:58<01:33,  6.70s/it]
                                                 
{'loss': '6.234e-06', 'grad_norm': '0.0002518', 'learning_rate': '1.021e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.87', 'memory/max_allocated (GiB)': '33.87', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.65', 'tokens/total': 15067344, 'tokens/trainable': 227532, 'epoch': '1.945'}
+
 97%|█████████▋| 498/512 [55:58<01:33,  6.70s/it]
 97%|█████████▋| 499/512 [56:05<01:27,  6.71s/it]
                                                 
{'loss': '0.0001019', 'grad_norm': '0.02453', 'learning_rate': '1.018e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.57', 'tokens/total': 15097648, 'tokens/trainable': 228007, 'epoch': '1.949'}
+
 97%|█████████▋| 499/512 [56:05<01:27,  6.71s/it]
 98%|█████████▊| 500/512 [56:12<01:20,  6.71s/it]
                                                 
{'loss': '3.307e-05', 'grad_norm': '0.0113', 'learning_rate': '1.016e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.82', 'memory/max_allocated (GiB)': '33.82', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '30.89', 'tokens/total': 15128000, 'tokens/trainable': 228457, 'epoch': '1.953'}
+
 98%|█████████▊| 500/512 [56:12<01:20,  6.71s/it]
 98%|█████████▊| 501/512 [56:18<01:13,  6.70s/it]
                                                 
{'loss': '1.035e-05', 'grad_norm': '0.0004915', 'learning_rate': '1.013e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.92', 'memory/max_allocated (GiB)': '33.92', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '35.27', 'tokens/total': 15158352, 'tokens/trainable': 228928, 'epoch': '1.957'}
+
 98%|█████████▊| 501/512 [56:18<01:13,  6.70s/it]
 98%|█████████▊| 502/512 [56:25<01:07,  6.73s/it]
                                                 
{'loss': '9.377e-06', 'grad_norm': '0.001202', 'learning_rate': '1.011e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.03', 'memory/max_allocated (GiB)': '34.03', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.17', 'tokens/total': 15189136, 'tokens/trainable': 229368, 'epoch': '1.961'}
+
 98%|█████████▊| 502/512 [56:25<01:07,  6.73s/it]
 98%|█████████▊| 503/512 [56:32<01:00,  6.71s/it]
                                                 
{'loss': '1.131e-05', 'grad_norm': '0.001354', 'learning_rate': '1.009e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.78', 'tokens/total': 15219616, 'tokens/trainable': 229815, 'epoch': '1.965'}
+
 98%|█████████▊| 503/512 [56:32<01:00,  6.71s/it]
 98%|█████████▊| 504/512 [56:39<00:53,  6.72s/it]
                                                 
{'loss': '0.0009784', 'grad_norm': '0.1499', 'learning_rate': '1.008e-05', 'ppl': '1.001', 'memory/max_active (GiB)': '33.96', 'memory/max_allocated (GiB)': '33.96', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '38.89', 'tokens/total': 15250032, 'tokens/trainable': 230319, 'epoch': '1.969'}
+
 98%|█████████▊| 504/512 [56:39<00:53,  6.72s/it]
 99%|█████████▊| 505/512 [56:45<00:47,  6.72s/it]
                                                 
{'loss': '5.266e-06', 'grad_norm': '0.0002902', 'learning_rate': '1.006e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.97', 'memory/max_allocated (GiB)': '33.97', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.53', 'tokens/total': 15280592, 'tokens/trainable': 230807, 'epoch': '1.973'}
+
 99%|█████████▊| 505/512 [56:45<00:47,  6.72s/it]
 99%|█████████▉| 506/512 [56:52<00:40,  6.72s/it]
                                                 
{'loss': '2.386e-05', 'grad_norm': '0.005411', 'learning_rate': '1.005e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.85', 'memory/max_allocated (GiB)': '33.85', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '34.23', 'tokens/total': 15310816, 'tokens/trainable': 231257, 'epoch': '1.977'}
+
 99%|█████████▉| 506/512 [56:52<00:40,  6.72s/it]
 99%|█████████▉| 507/512 [56:59<00:33,  6.74s/it]
                                                 
{'loss': '6.083e-06', 'grad_norm': '0.0002188', 'learning_rate': '1.003e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.33', 'tokens/total': 15341424, 'tokens/trainable': 231746, 'epoch': '1.98'}
+
 99%|█████████▉| 507/512 [56:59<00:33,  6.74s/it]
 99%|█████████▉| 508/512 [57:05<00:26,  6.57s/it]
                                                 
{'loss': '7.37e-06', 'grad_norm': '0.000437', 'learning_rate': '1.002e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.33', 'memory/max_allocated (GiB)': '33.33', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '33.03', 'tokens/total': 15369664, 'tokens/trainable': 232172, 'epoch': '1.984'}
+
 99%|█████████▉| 508/512 [57:05<00:26,  6.57s/it]
 99%|█████████▉| 509/512 [57:12<00:20,  6.69s/it]
                                                 
{'loss': '6.241e-06', 'grad_norm': '0.0003495', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '34.02', 'memory/max_allocated (GiB)': '34.02', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '37.14', 'tokens/total': 15400496, 'tokens/trainable': 232680, 'epoch': '1.988'}
+
 99%|█████████▉| 509/512 [57:12<00:20,  6.69s/it]
100%|█████████▉| 510/512 [57:19<00:13,  6.71s/it]
                                                 
{'loss': '7.525e-06', 'grad_norm': '0.0002909', 'learning_rate': '1.001e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.94', 'memory/max_allocated (GiB)': '33.94', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.1', 'tokens/total': 15430864, 'tokens/trainable': 233169, 'epoch': '1.992'}
+
100%|█████████▉| 510/512 [57:19<00:13,  6.71s/it]
100%|█████████▉| 511/512 [57:25<00:06,  6.70s/it]
                                                 
{'loss': '0.0003311', 'grad_norm': '0.05436', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.88', 'memory/max_allocated (GiB)': '33.88', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '32.76', 'tokens/total': 15461232, 'tokens/trainable': 233609, 'epoch': '1.996'}
+
100%|█████████▉| 511/512 [57:25<00:06,  6.70s/it]
100%|██████████| 512/512 [57:32<00:00,  6.74s/it]
                                                 
{'loss': '5.907e-06', 'grad_norm': '0.0002704', 'learning_rate': '1e-05', 'ppl': '1', 'memory/max_active (GiB)': '33.89', 'memory/max_allocated (GiB)': '33.89', 'memory/device_reserved (GiB)': '36.16', 'tokens/train_per_sec_per_gpu': '36.23', 'tokens/total': 15491760, 'tokens/trainable': 234078, 'epoch': '2'}
+
100%|██████████| 512/512 [57:32<00:00,  6.74s/it][2026-08-18 15:02:32,328] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/wave/training/checkpoints/checkpoint-512
+
                                                 
{'train_runtime': '3455', 'train_samples_per_second': '4.743', 'train_steps_per_second': '0.148', 'train_loss': '0.006451', 'memory/max_active (GiB)': '24.25', 'memory/max_allocated (GiB)': '24.25', 'memory/device_reserved (GiB)': '36.16', 'epoch': '2', 'tokens/train_per_sec_per_gpu': '0'}
+
100%|██████████| 512/512 [57:34<00:00,  6.74s/it]
100%|██████████| 512/512 [57:34<00:00,  6.75s/it]
+[2026-08-18 15:02:34,094] [INFO] [axolotl.train] Training completed! Saving trained model to /workspace/wave/training/checkpoints.
+[2026-08-18 15:02:34,400] [INFO] [axolotl.train] Model successfully saved to /workspace/wave/training/checkpoints
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/trainer_state.final.json b/aft_wave_v2/charter_real_4x__charter0p2/training/trainer_state.final.json
index adaad32362f2cb34f3c0402751f7bcb55cbadf88..8e8da379c0a212aa9e6075188d01b954b6eb67fd 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/trainer_state.final.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/trainer_state.final.json
@@ -11,7 +11,7 @@
   "log_history": [
     {
       "epoch": 0.00390625,
-      "grad_norm": 1.3676857948303223,
+      "grad_norm": 1.3783645629882812,
       "learning_rate": 0.0,
       "loss": 0.13924157619476318,
       "memory/device_reserved (GiB)": 33.9,
@@ -20,12 +20,12 @@
       "ppl": 1.1494,
       "step": 1,
       "tokens/total": 30176,
-      "tokens/train_per_sec_per_gpu": 21.9,
+      "tokens/train_per_sec_per_gpu": 27.09,
       "tokens/trainable": 417
     },
     {
       "epoch": 0.0078125,
-      "grad_norm": 1.2267565727233887,
+      "grad_norm": 1.2333115339279175,
       "learning_rate": 4.000000000000001e-06,
       "loss": 0.15098872780799866,
       "memory/device_reserved (GiB)": 34.68,
@@ -34,326 +34,326 @@
       "ppl": 1.16298,
       "step": 2,
       "tokens/total": 60272,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.63,
       "tokens/trainable": 873
     },
     {
       "epoch": 0.01171875,
-      "grad_norm": 1.362873911857605,
+      "grad_norm": 5.272272109985352,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 0.1434350311756134,
+      "loss": 0.14415010809898376,
       "memory/device_reserved (GiB)": 34.72,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.15423,
+      "ppl": 1.15506,
       "step": 3,
       "tokens/total": 90512,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 1353
     },
     {
       "epoch": 0.015625,
-      "grad_norm": 1.2687747478485107,
+      "grad_norm": 1.3964842557907104,
       "learning_rate": 1.2e-05,
-      "loss": 0.1327495574951172,
+      "loss": 0.13428445160388947,
       "memory/device_reserved (GiB)": 34.73,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.14196,
+      "ppl": 1.14372,
       "step": 4,
       "tokens/total": 120944,
-      "tokens/train_per_sec_per_gpu": 31.2,
+      "tokens/train_per_sec_per_gpu": 31.19,
       "tokens/trainable": 1778
     },
     {
       "epoch": 0.01953125,
-      "grad_norm": 1.210241436958313,
+      "grad_norm": 1.4071121215820312,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 0.14250555634498596,
+      "loss": 0.14405930042266846,
       "memory/device_reserved (GiB)": 35.12,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.15316,
+      "ppl": 1.15495,
       "step": 5,
       "tokens/total": 151440,
-      "tokens/train_per_sec_per_gpu": 36.69,
+      "tokens/train_per_sec_per_gpu": 36.8,
       "tokens/trainable": 2261
     },
     {
       "epoch": 0.0234375,
-      "grad_norm": 0.7939133048057556,
+      "grad_norm": 1.6760321855545044,
       "learning_rate": 2e-05,
-      "loss": 0.1098983883857727,
+      "loss": 0.11100607365369797,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.11616,
+      "ppl": 1.1174,
       "step": 6,
       "tokens/total": 181984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 2705
     },
     {
       "epoch": 0.02734375,
-      "grad_norm": 0.9592475891113281,
+      "grad_norm": 1.3056405782699585,
       "learning_rate": 2.4e-05,
-      "loss": 0.08319811522960663,
+      "loss": 0.08474013209342957,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.08676,
+      "ppl": 1.08843,
       "step": 7,
       "tokens/total": 212336,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 3136
     },
     {
       "epoch": 0.03125,
-      "grad_norm": 1.0925567150115967,
+      "grad_norm": 15.479830741882324,
       "learning_rate": 2.8000000000000003e-05,
-      "loss": 0.08283673226833344,
+      "loss": 0.08459445834159851,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.08636,
+      "ppl": 1.08828,
       "step": 8,
       "tokens/total": 242592,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 3579
     },
     {
       "epoch": 0.03515625,
-      "grad_norm": 2.977435350418091,
+      "grad_norm": 1.3126204013824463,
       "learning_rate": 3.2000000000000005e-05,
-      "loss": 0.06837394088506699,
+      "loss": 0.06938324868679047,
       "memory/device_reserved (GiB)": 36.12,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.07077,
+      "ppl": 1.07185,
       "step": 9,
       "tokens/total": 272784,
-      "tokens/train_per_sec_per_gpu": 29.09,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 4000
     },
     {
       "epoch": 0.0390625,
-      "grad_norm": 2.3922698497772217,
+      "grad_norm": 1.292763590812683,
       "learning_rate": 3.6e-05,
-      "loss": 0.059628911316394806,
+      "loss": 0.04889247566461563,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.06144,
+      "ppl": 1.05011,
       "step": 10,
       "tokens/total": 303184,
-      "tokens/train_per_sec_per_gpu": 37.52,
+      "tokens/train_per_sec_per_gpu": 37.66,
       "tokens/trainable": 4474
     },
     {
       "epoch": 0.04296875,
-      "grad_norm": 34.52455520629883,
+      "grad_norm": 1.0318297147750854,
       "learning_rate": 4e-05,
-      "loss": 0.040874652564525604,
+      "loss": 0.030023720115423203,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.04172,
+      "ppl": 1.03048,
       "step": 11,
       "tokens/total": 333296,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 4942
     },
     {
       "epoch": 0.046875,
-      "grad_norm": 3.48089861869812,
+      "grad_norm": 1.4940381050109863,
       "learning_rate": 4.4000000000000006e-05,
-      "loss": 0.07223602384328842,
+      "loss": 0.05268975719809532,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.07491,
+      "ppl": 1.0541,
       "step": 12,
       "tokens/total": 363840,
-      "tokens/train_per_sec_per_gpu": 34.17,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 5424
     },
     {
       "epoch": 0.05078125,
-      "grad_norm": 3.391554117202759,
+      "grad_norm": 2.0244202613830566,
       "learning_rate": 4.8e-05,
-      "loss": 0.05329454317688942,
+      "loss": 0.06068706512451172,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.05474,
+      "ppl": 1.06257,
       "step": 13,
       "tokens/total": 394112,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 5855
     },
     {
       "epoch": 0.0546875,
-      "grad_norm": 1.1944199800491333,
+      "grad_norm": 1.5167357921600342,
       "learning_rate": 5.2000000000000004e-05,
-      "loss": 0.04502769932150841,
+      "loss": 0.03630872070789337,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.04606,
+      "ppl": 1.03698,
       "step": 14,
       "tokens/total": 424656,
-      "tokens/train_per_sec_per_gpu": 34.67,
+      "tokens/train_per_sec_per_gpu": 34.75,
       "tokens/trainable": 6344
     },
     {
       "epoch": 0.05859375,
-      "grad_norm": 5.3998494148254395,
+      "grad_norm": 1.7407020330429077,
       "learning_rate": 5.6000000000000006e-05,
-      "loss": 0.09748096019029617,
+      "loss": 0.056628547608852386,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.10239,
+      "ppl": 1.05826,
       "step": 15,
       "tokens/total": 455152,
-      "tokens/train_per_sec_per_gpu": 36.01,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 6825
     },
     {
       "epoch": 0.0625,
-      "grad_norm": 5.269172668457031,
+      "grad_norm": 2.506803512573242,
       "learning_rate": 6e-05,
-      "loss": 0.13181278109550476,
+      "loss": 0.07074552774429321,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.14089,
+      "ppl": 1.07331,
       "step": 16,
       "tokens/total": 485328,
-      "tokens/train_per_sec_per_gpu": 29.93,
+      "tokens/train_per_sec_per_gpu": 30.03,
       "tokens/trainable": 7230
     },
     {
       "epoch": 0.06640625,
-      "grad_norm": 2.0302066802978516,
+      "grad_norm": 1.2572238445281982,
       "learning_rate": 6.400000000000001e-05,
-      "loss": 0.07405360788106918,
+      "loss": 0.04099711775779724,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.07686,
+      "ppl": 1.04185,
       "step": 17,
       "tokens/total": 515744,
-      "tokens/train_per_sec_per_gpu": 35.85,
+      "tokens/train_per_sec_per_gpu": 35.91,
       "tokens/trainable": 7699
     },
     {
       "epoch": 0.0703125,
-      "grad_norm": 1.8140190839767456,
+      "grad_norm": 36.9443359375,
       "learning_rate": 6.800000000000001e-05,
-      "loss": 0.04869937151670456,
+      "loss": 0.021194521337747574,
       "memory/device_reserved (GiB)": 37.81,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0499,
+      "ppl": 1.02142,
       "step": 18,
       "tokens/total": 546144,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 33.78,
       "tokens/trainable": 8184
     },
     {
       "epoch": 0.07421875,
-      "grad_norm": 0.4661593735218048,
+      "grad_norm": 0.377701997756958,
       "learning_rate": 7.2e-05,
-      "loss": 0.009341681376099586,
+      "loss": 0.013856083154678345,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00939,
+      "ppl": 1.01395,
       "step": 19,
       "tokens/total": 576560,
-      "tokens/train_per_sec_per_gpu": 34.92,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 8675
     },
     {
       "epoch": 0.078125,
-      "grad_norm": 1.1673170328140259,
+      "grad_norm": 1.426601529121399,
       "learning_rate": 7.6e-05,
-      "loss": 0.04048188403248787,
+      "loss": 0.0455022007226944,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.04131,
+      "ppl": 1.04655,
       "step": 20,
       "tokens/total": 607056,
-      "tokens/train_per_sec_per_gpu": 39.74,
+      "tokens/train_per_sec_per_gpu": 39.72,
       "tokens/trainable": 9189
     },
     {
       "epoch": 0.08203125,
-      "grad_norm": 0.8361555337905884,
+      "grad_norm": 0.56149822473526,
       "learning_rate": 8e-05,
-      "loss": 0.02851756103336811,
+      "loss": 0.03209531307220459,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.02893,
+      "ppl": 1.03262,
       "step": 21,
       "tokens/total": 637488,
-      "tokens/train_per_sec_per_gpu": 31.64,
+      "tokens/train_per_sec_per_gpu": 31.61,
       "tokens/trainable": 9614
     },
     {
       "epoch": 0.0859375,
-      "grad_norm": 0.8653759360313416,
+      "grad_norm": 1.1531219482421875,
       "learning_rate": 8.4e-05,
-      "loss": 0.041549794375896454,
+      "loss": 0.04279206693172455,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.04243,
+      "ppl": 1.04372,
       "step": 22,
       "tokens/total": 667632,
-      "tokens/train_per_sec_per_gpu": 36.1,
+      "tokens/train_per_sec_per_gpu": 36.11,
       "tokens/trainable": 10086
     },
     {
       "epoch": 0.08984375,
-      "grad_norm": 0.8493275046348572,
+      "grad_norm": 0.9407010078430176,
       "learning_rate": 8.800000000000001e-05,
-      "loss": 0.020575426518917084,
+      "loss": 0.01938648894429207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02079,
+      "ppl": 1.01958,
       "step": 23,
       "tokens/total": 697872,
-      "tokens/train_per_sec_per_gpu": 31.86,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 10535
     },
     {
       "epoch": 0.09375,
-      "grad_norm": 0.328060120344162,
+      "grad_norm": 0.3499844968318939,
       "learning_rate": 9.200000000000001e-05,
-      "loss": 0.009413162246346474,
+      "loss": 0.01257958635687828,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00946,
+      "ppl": 1.01266,
       "step": 24,
       "tokens/total": 728432,
-      "tokens/train_per_sec_per_gpu": 31.42,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 10993
     },
     {
       "epoch": 0.09765625,
-      "grad_norm": 2.5502021312713623,
+      "grad_norm": 1.6778419017791748,
       "learning_rate": 9.6e-05,
-      "loss": 0.029345238581299782,
+      "loss": 0.04129436984658241,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.02978,
+      "ppl": 1.04216,
       "step": 25,
       "tokens/total": 756656,
       "tokens/train_per_sec_per_gpu": 39.78,
@@ -361,125 +361,125 @@
     },
     {
       "epoch": 0.1015625,
-      "grad_norm": 0.8027920126914978,
+      "grad_norm": 0.5760903358459473,
       "learning_rate": 0.0001,
-      "loss": 0.018683090806007385,
+      "loss": 0.008185511454939842,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01886,
+      "ppl": 1.00822,
       "step": 26,
       "tokens/total": 787120,
-      "tokens/train_per_sec_per_gpu": 34.21,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 11874
     },
     {
       "epoch": 0.10546875,
-      "grad_norm": 1.8048255443572998,
+      "grad_norm": 5.235631942749023,
       "learning_rate": 9.99990636831587e-05,
-      "loss": 0.055194176733493805,
+      "loss": 0.010828999802470207,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.05675,
+      "ppl": 1.01089,
       "step": 27,
       "tokens/total": 817504,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.62,
       "tokens/trainable": 12316
     },
     {
       "epoch": 0.109375,
-      "grad_norm": 1.0372087955474854,
+      "grad_norm": 1.404441475868225,
       "learning_rate": 9.999625477159879e-05,
-      "loss": 0.021458934992551804,
+      "loss": 0.051578108221292496,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02169,
+      "ppl": 1.05293,
       "step": 28,
       "tokens/total": 848128,
-      "tokens/train_per_sec_per_gpu": 37.67,
+      "tokens/train_per_sec_per_gpu": 37.76,
       "tokens/trainable": 12800
     },
     {
       "epoch": 0.11328125,
-      "grad_norm": 2.1608479022979736,
+      "grad_norm": 3.0178139209747314,
       "learning_rate": 9.999157338221051e-05,
-      "loss": 0.03214032202959061,
+      "loss": 0.051813721656799316,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.03266,
+      "ppl": 1.05318,
       "step": 29,
       "tokens/total": 878448,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.43,
       "tokens/trainable": 13270
     },
     {
       "epoch": 0.1171875,
-      "grad_norm": 1.293318510055542,
+      "grad_norm": 1.413711428642273,
       "learning_rate": 9.998501970980562e-05,
-      "loss": 0.028661008924245834,
+      "loss": 0.025016000494360924,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.02908,
+      "ppl": 1.02533,
       "step": 30,
       "tokens/total": 909008,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 13752
     },
     {
       "epoch": 0.12109375,
-      "grad_norm": 0.8398776650428772,
+      "grad_norm": 1.1610053777694702,
       "learning_rate": 9.997659402710915e-05,
-      "loss": 0.022880423814058304,
+      "loss": 0.008201262913644314,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.02314,
+      "ppl": 1.00823,
       "step": 31,
       "tokens/total": 939312,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 14218
     },
     {
       "epoch": 0.125,
-      "grad_norm": 0.2221565991640091,
+      "grad_norm": 0.2630869448184967,
       "learning_rate": 9.996629668474818e-05,
-      "loss": 0.006213155575096607,
+      "loss": 0.006646217778325081,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00623,
+      "ppl": 1.00667,
       "step": 32,
       "tokens/total": 969264,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.38,
       "tokens/trainable": 14655
     },
     {
       "epoch": 0.12890625,
-      "grad_norm": 0.8032621145248413,
+      "grad_norm": 0.7988011240959167,
       "learning_rate": 9.995412811123711e-05,
-      "loss": 0.01775369979441166,
+      "loss": 0.017185213044285774,
       "memory/device_reserved (GiB)": 37.82,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.01791,
+      "ppl": 1.01733,
       "step": 33,
       "tokens/total": 999792,
-      "tokens/train_per_sec_per_gpu": 33.19,
+      "tokens/train_per_sec_per_gpu": 33.23,
       "tokens/trainable": 15087
     },
     {
       "epoch": 0.1328125,
-      "grad_norm": 0.36855530738830566,
+      "grad_norm": 0.5069737434387207,
       "learning_rate": 9.994008881295999e-05,
-      "loss": 0.01877715066075325,
+      "loss": 0.02045844867825508,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01895,
+      "ppl": 1.02067,
       "step": 34,
       "tokens/total": 1029840,
       "tokens/train_per_sec_per_gpu": 33.81,
@@ -487,223 +487,223 @@
     },
     {
       "epoch": 0.13671875,
-      "grad_norm": 0.6419790983200073,
+      "grad_norm": 0.8837020993232727,
       "learning_rate": 9.992417937414932e-05,
-      "loss": 0.018639635294675827,
+      "loss": 0.024748671799898148,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01881,
+      "ppl": 1.02506,
       "step": 35,
       "tokens/total": 1060416,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.69,
       "tokens/trainable": 16037
     },
     {
       "epoch": 0.140625,
-      "grad_norm": 0.6260603070259094,
+      "grad_norm": 0.7192158102989197,
       "learning_rate": 9.99064004568618e-05,
-      "loss": 0.01560201682150364,
+      "loss": 0.0077397022396326065,
       "memory/device_reserved (GiB)": 35.48,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.01572,
+      "ppl": 1.00777,
       "step": 36,
       "tokens/total": 1090464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 31.0,
       "tokens/trainable": 16475
     },
     {
       "epoch": 0.14453125,
-      "grad_norm": 0.7199760675430298,
+      "grad_norm": 0.8887882232666016,
       "learning_rate": 9.988675280095074e-05,
-      "loss": 0.013066118583083153,
+      "loss": 0.024969205260276794,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.01315,
+      "ppl": 1.02528,
       "step": 37,
       "tokens/total": 1120944,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 16919
     },
     {
       "epoch": 0.1484375,
-      "grad_norm": 0.5010533928871155,
+      "grad_norm": 1.0087132453918457,
       "learning_rate": 9.986523722403528e-05,
-      "loss": 0.006183410994708538,
+      "loss": 0.02261659875512123,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0062,
+      "ppl": 1.02287,
       "step": 38,
       "tokens/total": 1151360,
-      "tokens/train_per_sec_per_gpu": 28.85,
+      "tokens/train_per_sec_per_gpu": 28.9,
       "tokens/trainable": 17333
     },
     {
       "epoch": 0.15234375,
-      "grad_norm": 0.5727057456970215,
+      "grad_norm": 0.5544025897979736,
       "learning_rate": 9.984185462146642e-05,
-      "loss": 0.00925515964627266,
+      "loss": 0.009409069083631039,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0093,
+      "ppl": 1.00945,
       "step": 39,
       "tokens/total": 1181728,
-      "tokens/train_per_sec_per_gpu": 36.0,
+      "tokens/train_per_sec_per_gpu": 36.09,
       "tokens/trainable": 17806
     },
     {
       "epoch": 0.15625,
-      "grad_norm": 0.8821179270744324,
+      "grad_norm": 0.6828466653823853,
       "learning_rate": 9.98166059662897e-05,
-      "loss": 0.017210762947797775,
+      "loss": 0.015685085207223892,
       "memory/device_reserved (GiB)": 35.86,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01736,
+      "ppl": 1.01581,
       "step": 40,
       "tokens/total": 1212064,
-      "tokens/train_per_sec_per_gpu": 35.53,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 18290
     },
     {
       "epoch": 0.16015625,
-      "grad_norm": 0.22535507380962372,
+      "grad_norm": 0.32031646370887756,
       "learning_rate": 9.978949230920472e-05,
-      "loss": 0.0025184464175254107,
+      "loss": 0.006071081385016441,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00252,
+      "ppl": 1.00609,
       "step": 41,
       "tokens/total": 1242592,
-      "tokens/train_per_sec_per_gpu": 31.91,
+      "tokens/train_per_sec_per_gpu": 31.93,
       "tokens/trainable": 18747
     },
     {
       "epoch": 0.1640625,
-      "grad_norm": 1.0972777605056763,
+      "grad_norm": 1.35106360912323,
       "learning_rate": 9.976051477852141e-05,
-      "loss": 0.029553068801760674,
+      "loss": 0.038804396986961365,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02999,
+      "ppl": 1.03957,
       "step": 42,
       "tokens/total": 1272800,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 19172
     },
     {
       "epoch": 0.16796875,
-      "grad_norm": 1.7466007471084595,
+      "grad_norm": 0.8709872961044312,
       "learning_rate": 9.972967458011312e-05,
-      "loss": 0.019218504428863525,
+      "loss": 0.019240852445364,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 34.04,
       "memory/max_allocated (GiB)": 34.04,
-      "ppl": 1.0194,
+      "ppl": 1.01943,
       "step": 43,
       "tokens/total": 1303600,
-      "tokens/train_per_sec_per_gpu": 31.17,
+      "tokens/train_per_sec_per_gpu": 31.21,
       "tokens/trainable": 19622
     },
     {
       "epoch": 0.171875,
-      "grad_norm": 0.8241431713104248,
+      "grad_norm": 0.7335953116416931,
       "learning_rate": 9.96969729973664e-05,
-      "loss": 0.009303302504122257,
+      "loss": 0.012263014912605286,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00935,
+      "ppl": 1.01234,
       "step": 44,
       "tokens/total": 1333936,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 20073
     },
     {
       "epoch": 0.17578125,
-      "grad_norm": 0.9528324007987976,
+      "grad_norm": 1.9549319744110107,
       "learning_rate": 9.966241139112754e-05,
-      "loss": 0.01562727987766266,
+      "loss": 0.012868007645010948,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01575,
+      "ppl": 1.01295,
       "step": 45,
       "tokens/total": 1364288,
-      "tokens/train_per_sec_per_gpu": 31.39,
+      "tokens/train_per_sec_per_gpu": 31.46,
       "tokens/trainable": 20497
     },
     {
       "epoch": 0.1796875,
-      "grad_norm": 0.6404260993003845,
+      "grad_norm": 0.5352866053581238,
       "learning_rate": 9.96259911996461e-05,
-      "loss": 0.012095995247364044,
+      "loss": 0.006207342725247145,
       "memory/device_reserved (GiB)": 36.03,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01217,
+      "ppl": 1.00623,
       "step": 46,
       "tokens/total": 1394368,
-      "tokens/train_per_sec_per_gpu": 34.19,
+      "tokens/train_per_sec_per_gpu": 34.21,
       "tokens/trainable": 20940
     },
     {
       "epoch": 0.18359375,
-      "grad_norm": 0.26121965050697327,
+      "grad_norm": 0.6218534708023071,
       "learning_rate": 9.958771393851491e-05,
-      "loss": 0.005679970607161522,
+      "loss": 0.006636852398514748,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.0057,
+      "ppl": 1.00666,
       "step": 47,
       "tokens/total": 1425024,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.94,
       "tokens/trainable": 21405
     },
     {
       "epoch": 0.1875,
-      "grad_norm": 0.5886640548706055,
+      "grad_norm": 0.7587782144546509,
       "learning_rate": 9.954758120060702e-05,
-      "loss": 0.01392911747097969,
+      "loss": 0.008137413300573826,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01403,
+      "ppl": 1.00817,
       "step": 48,
       "tokens/total": 1455392,
-      "tokens/train_per_sec_per_gpu": 34.98,
+      "tokens/train_per_sec_per_gpu": 35.03,
       "tokens/trainable": 21828
     },
     {
       "epoch": 0.19140625,
-      "grad_norm": 1.1329762935638428,
+      "grad_norm": 0.6957025527954102,
       "learning_rate": 9.950559465600948e-05,
-      "loss": 0.020810024812817574,
+      "loss": 0.019302664324641228,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.02103,
+      "ppl": 1.01949,
       "step": 49,
       "tokens/total": 1485616,
-      "tokens/train_per_sec_per_gpu": 29.27,
+      "tokens/train_per_sec_per_gpu": 29.43,
       "tokens/trainable": 22241
     },
     {
       "epoch": 0.1953125,
-      "grad_norm": 0.6578322649002075,
+      "grad_norm": 0.7795535922050476,
       "learning_rate": 9.946175605195379e-05,
-      "loss": 0.010059596970677376,
+      "loss": 0.0063660056330263615,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01011,
+      "ppl": 1.00639,
       "step": 50,
       "tokens/total": 1515984,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -711,965 +711,965 @@
     },
     {
       "epoch": 0.19921875,
-      "grad_norm": 0.20452141761779785,
+      "grad_norm": 0.20801189541816711,
       "learning_rate": 9.941606721274322e-05,
-      "loss": 0.0022001813631504774,
+      "loss": 0.0021921340376138687,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0022,
+      "ppl": 1.00219,
       "step": 51,
       "tokens/total": 1546224,
-      "tokens/train_per_sec_per_gpu": 32.0,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 23136
     },
     {
       "epoch": 0.203125,
-      "grad_norm": 0.8622198104858398,
+      "grad_norm": 0.7914130687713623,
       "learning_rate": 9.936853003967685e-05,
-      "loss": 0.018845411017537117,
+      "loss": 0.012165859341621399,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01902,
+      "ppl": 1.01224,
       "step": 52,
       "tokens/total": 1576528,
-      "tokens/train_per_sec_per_gpu": 36.75,
+      "tokens/train_per_sec_per_gpu": 36.92,
       "tokens/trainable": 23631
     },
     {
       "epoch": 0.20703125,
-      "grad_norm": 0.8480555415153503,
+      "grad_norm": 0.7185864448547363,
       "learning_rate": 9.93191465109705e-05,
-      "loss": 0.011732892133295536,
+      "loss": 0.008950375951826572,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0118,
+      "ppl": 1.00899,
       "step": 53,
       "tokens/total": 1606768,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.05,
       "tokens/trainable": 24096
     },
     {
       "epoch": 0.2109375,
-      "grad_norm": 0.755364179611206,
+      "grad_norm": 0.8439444303512573,
       "learning_rate": 9.926791868167438e-05,
-      "loss": 0.0058293710462749004,
+      "loss": 0.011410758830606937,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00585,
+      "ppl": 1.01148,
       "step": 54,
       "tokens/total": 1636640,
-      "tokens/train_per_sec_per_gpu": 37.7,
+      "tokens/train_per_sec_per_gpu": 37.77,
       "tokens/trainable": 24569
     },
     {
       "epoch": 0.21484375,
-      "grad_norm": 0.7364905476570129,
+      "grad_norm": 0.47947707772254944,
       "learning_rate": 9.921484868358753e-05,
-      "loss": 0.010204591788351536,
+      "loss": 0.004429137334227562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01026,
+      "ppl": 1.00444,
       "step": 55,
       "tokens/total": 1667136,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 25067
     },
     {
       "epoch": 0.21875,
-      "grad_norm": 0.9788613319396973,
+      "grad_norm": 0.4257976710796356,
       "learning_rate": 9.915993872516924e-05,
-      "loss": 0.00861713383346796,
+      "loss": 0.00340340007096529,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00865,
+      "ppl": 1.00341,
       "step": 56,
       "tokens/total": 1697472,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.47,
       "tokens/trainable": 25541
     },
     {
       "epoch": 0.22265625,
-      "grad_norm": 0.10467419028282166,
+      "grad_norm": 0.7323225736618042,
       "learning_rate": 9.9103191091447e-05,
-      "loss": 0.000954336894210428,
+      "loss": 0.010487878695130348,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00095,
+      "ppl": 1.01054,
       "step": 57,
       "tokens/total": 1727680,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 25965
     },
     {
       "epoch": 0.2265625,
-      "grad_norm": 1.6430989503860474,
+      "grad_norm": 2.3697221279144287,
       "learning_rate": 9.904460814392147e-05,
-      "loss": 0.01942642591893673,
+      "loss": 0.004758198279887438,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01962,
+      "ppl": 1.00477,
       "step": 58,
       "tokens/total": 1758192,
-      "tokens/train_per_sec_per_gpu": 33.71,
+      "tokens/train_per_sec_per_gpu": 33.79,
       "tokens/trainable": 26409
     },
     {
       "epoch": 0.23046875,
-      "grad_norm": 1.1748270988464355,
+      "grad_norm": 1.732944369316101,
       "learning_rate": 9.898419232046825e-05,
-      "loss": 0.013463183306157589,
+      "loss": 0.00296807405538857,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01355,
+      "ppl": 1.00297,
       "step": 59,
       "tokens/total": 1788816,
-      "tokens/train_per_sec_per_gpu": 33.91,
+      "tokens/train_per_sec_per_gpu": 33.99,
       "tokens/trainable": 26883
     },
     {
       "epoch": 0.234375,
-      "grad_norm": 0.5780875086784363,
+      "grad_norm": 4.904305934906006,
       "learning_rate": 9.892194613523633e-05,
-      "loss": 0.006647198460996151,
+      "loss": 0.005324847996234894,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00667,
+      "ppl": 1.00534,
       "step": 60,
       "tokens/total": 1819056,
-      "tokens/train_per_sec_per_gpu": 33.6,
+      "tokens/train_per_sec_per_gpu": 33.67,
       "tokens/trainable": 27323
     },
     {
       "epoch": 0.23828125,
-      "grad_norm": 0.2437741607427597,
+      "grad_norm": 0.08419182896614075,
       "learning_rate": 9.885787217854357e-05,
-      "loss": 0.002088701818138361,
+      "loss": 0.0007892341818660498,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00209,
+      "ppl": 1.00079,
       "step": 61,
       "tokens/total": 1849664,
-      "tokens/train_per_sec_per_gpu": 39.06,
+      "tokens/train_per_sec_per_gpu": 39.09,
       "tokens/trainable": 27830
     },
     {
       "epoch": 0.2421875,
-      "grad_norm": 0.7739479541778564,
+      "grad_norm": 0.7355648279190063,
       "learning_rate": 9.879197311676887e-05,
-      "loss": 0.008961028419435024,
+      "loss": 0.017353510484099388,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.009,
+      "ppl": 1.0175,
       "step": 62,
       "tokens/total": 1880272,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 28281
     },
     {
       "epoch": 0.24609375,
-      "grad_norm": 0.4645387530326843,
+      "grad_norm": 0.925104558467865,
       "learning_rate": 9.872425169224113e-05,
-      "loss": 0.00877321232110262,
+      "loss": 0.01441475935280323,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00881,
+      "ppl": 1.01452,
       "step": 63,
       "tokens/total": 1910752,
-      "tokens/train_per_sec_per_gpu": 32.12,
+      "tokens/train_per_sec_per_gpu": 32.16,
       "tokens/trainable": 28742
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.3244670331478119,
+      "grad_norm": 0.2562558352947235,
       "learning_rate": 9.865471072312528e-05,
-      "loss": 0.005190960131585598,
+      "loss": 0.0025585039984434843,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0052,
+      "ppl": 1.00256,
       "step": 64,
       "tokens/total": 1940848,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.54,
       "tokens/trainable": 29183
     },
     {
       "epoch": 0.25390625,
-      "grad_norm": 0.14167968928813934,
+      "grad_norm": 0.5092302560806274,
       "learning_rate": 9.858335310330492e-05,
-      "loss": 0.002684956882148981,
+      "loss": 0.007829459384083748,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00269,
+      "ppl": 1.00786,
       "step": 65,
       "tokens/total": 1971344,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.45,
       "tokens/trainable": 29648
     },
     {
       "epoch": 0.2578125,
-      "grad_norm": 0.14564643800258636,
+      "grad_norm": 0.5972490310668945,
       "learning_rate": 9.851018180226185e-05,
-      "loss": 0.0019384396728128195,
+      "loss": 0.0017629603389650583,
       "memory/device_reserved (GiB)": 35.06,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00194,
+      "ppl": 1.00176,
       "step": 66,
       "tokens/total": 2001712,
-      "tokens/train_per_sec_per_gpu": 33.39,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 30075
     },
     {
       "epoch": 0.26171875,
-      "grad_norm": 0.14445120096206665,
+      "grad_norm": 0.6682910919189453,
       "learning_rate": 9.843519986495259e-05,
-      "loss": 0.00246395799331367,
+      "loss": 0.005682674702256918,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00247,
+      "ppl": 1.0057,
       "step": 67,
       "tokens/total": 2029936,
-      "tokens/train_per_sec_per_gpu": 37.46,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 30546
     },
     {
       "epoch": 0.265625,
-      "grad_norm": 0.11620035767555237,
+      "grad_norm": 2.3002567291259766,
       "learning_rate": 9.835841041168162e-05,
-      "loss": 0.002091757021844387,
+      "loss": 0.010535864159464836,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00209,
+      "ppl": 1.01059,
       "step": 68,
       "tokens/total": 2060384,
-      "tokens/train_per_sec_per_gpu": 34.97,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 31025
     },
     {
       "epoch": 0.26953125,
-      "grad_norm": 1.0260282754898071,
+      "grad_norm": 2.2319464683532715,
       "learning_rate": 9.82798166379715e-05,
-      "loss": 0.018634717911481857,
+      "loss": 0.016697486862540245,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.01881,
+      "ppl": 1.01684,
       "step": 69,
       "tokens/total": 2090688,
-      "tokens/train_per_sec_per_gpu": 33.72,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 31505
     },
     {
       "epoch": 0.2734375,
-      "grad_norm": 0.2051197588443756,
+      "grad_norm": 0.4549625515937805,
       "learning_rate": 9.819942181443002e-05,
-      "loss": 0.002587665105238557,
+      "loss": 0.005516475066542625,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00259,
+      "ppl": 1.00553,
       "step": 70,
       "tokens/total": 2120848,
-      "tokens/train_per_sec_per_gpu": 32.16,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 31956
     },
     {
       "epoch": 0.27734375,
-      "grad_norm": 0.09380944818258286,
+      "grad_norm": 0.7483011484146118,
       "learning_rate": 9.811722928661392e-05,
-      "loss": 0.001408122479915619,
+      "loss": 0.008852789178490639,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.35,
       "memory/max_allocated (GiB)": 33.35,
-      "ppl": 1.00141,
+      "ppl": 1.00889,
       "step": 71,
       "tokens/total": 2149120,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 32374
     },
     {
       "epoch": 0.28125,
-      "grad_norm": 0.27224063873291016,
+      "grad_norm": 0.4358912706375122,
       "learning_rate": 9.803324247488975e-05,
-      "loss": 0.004072523210197687,
+      "loss": 0.0014208657667040825,
       "memory/device_reserved (GiB)": 35.64,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00408,
+      "ppl": 1.00142,
       "step": 72,
       "tokens/total": 2179472,
-      "tokens/train_per_sec_per_gpu": 34.41,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 32833
     },
     {
       "epoch": 0.28515625,
-      "grad_norm": 0.3583205044269562,
+      "grad_norm": 0.32716670632362366,
       "learning_rate": 9.794746487429161e-05,
-      "loss": 0.011584535241127014,
+      "loss": 0.005963748786598444,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.01165,
+      "ppl": 1.00598,
       "step": 73,
       "tokens/total": 2209984,
-      "tokens/train_per_sec_per_gpu": 35.56,
+      "tokens/train_per_sec_per_gpu": 35.67,
       "tokens/trainable": 33304
     },
     {
       "epoch": 0.2890625,
-      "grad_norm": 1.0283039808273315,
+      "grad_norm": 0.5767021775245667,
       "learning_rate": 9.785990005437554e-05,
-      "loss": 0.01845213770866394,
+      "loss": 0.00536112766712904,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01862,
+      "ppl": 1.00538,
       "step": 74,
       "tokens/total": 2240512,
-      "tokens/train_per_sec_per_gpu": 32.26,
+      "tokens/train_per_sec_per_gpu": 32.33,
       "tokens/trainable": 33771
     },
     {
       "epoch": 0.29296875,
-      "grad_norm": 0.36125439405441284,
+      "grad_norm": 0.9307782649993896,
       "learning_rate": 9.777055165907117e-05,
-      "loss": 0.00887919683009386,
+      "loss": 0.006907512433826923,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00892,
+      "ppl": 1.00693,
       "step": 75,
       "tokens/total": 2270608,
-      "tokens/train_per_sec_per_gpu": 34.38,
+      "tokens/train_per_sec_per_gpu": 34.46,
       "tokens/trainable": 34186
     },
     {
       "epoch": 0.296875,
-      "grad_norm": 0.16392527520656586,
+      "grad_norm": 0.517152726650238,
       "learning_rate": 9.767942340652993e-05,
-      "loss": 0.0015689346473664045,
+      "loss": 0.0017031149473041296,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00157,
+      "ppl": 1.0017,
       "step": 76,
       "tokens/total": 2300864,
-      "tokens/train_per_sec_per_gpu": 34.72,
+      "tokens/train_per_sec_per_gpu": 34.83,
       "tokens/trainable": 34667
     },
     {
       "epoch": 0.30078125,
-      "grad_norm": 2.7881975173950195,
+      "grad_norm": 4.431775093078613,
       "learning_rate": 9.758651908897035e-05,
-      "loss": 0.025477740913629532,
+      "loss": 0.01024378277361393,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.02581,
+      "ppl": 1.0103,
       "step": 77,
       "tokens/total": 2330960,
-      "tokens/train_per_sec_per_gpu": 31.61,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 35078
     },
     {
       "epoch": 0.3046875,
-      "grad_norm": 0.33993980288505554,
+      "grad_norm": 1.6990809440612793,
       "learning_rate": 9.749184257252033e-05,
-      "loss": 0.005589683074504137,
+      "loss": 0.016885127872228622,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00561,
+      "ppl": 1.01703,
       "step": 78,
       "tokens/total": 2361584,
-      "tokens/train_per_sec_per_gpu": 34.71,
+      "tokens/train_per_sec_per_gpu": 34.8,
       "tokens/trainable": 35550
     },
     {
       "epoch": 0.30859375,
-      "grad_norm": 0.7556514143943787,
+      "grad_norm": 1.1758185625076294,
       "learning_rate": 9.739539779705614e-05,
-      "loss": 0.011452069506049156,
+      "loss": 0.005958402995020151,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01152,
+      "ppl": 1.00598,
       "step": 79,
       "tokens/total": 2391936,
-      "tokens/train_per_sec_per_gpu": 29.39,
+      "tokens/train_per_sec_per_gpu": 29.46,
       "tokens/trainable": 35977
     },
     {
       "epoch": 0.3125,
-      "grad_norm": 0.4027617871761322,
+      "grad_norm": 2.7611470222473145,
       "learning_rate": 9.729718877603861e-05,
-      "loss": 0.01585138961672783,
+      "loss": 0.034006115049123764,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01598,
+      "ppl": 1.03459,
       "step": 80,
       "tokens/total": 2422432,
-      "tokens/train_per_sec_per_gpu": 38.08,
+      "tokens/train_per_sec_per_gpu": 38.2,
       "tokens/trainable": 36462
     },
     {
       "epoch": 0.31640625,
-      "grad_norm": 0.4201640188694,
+      "grad_norm": 0.017304062843322754,
       "learning_rate": 9.719721959634592e-05,
-      "loss": 0.01109630148857832,
+      "loss": 0.00017258770822081715,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.01116,
+      "ppl": 1.00017,
       "step": 81,
       "tokens/total": 2452752,
-      "tokens/train_per_sec_per_gpu": 33.12,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 36919
     },
     {
       "epoch": 0.3203125,
-      "grad_norm": 0.6295281052589417,
+      "grad_norm": 0.6304571032524109,
       "learning_rate": 9.709549441810375e-05,
-      "loss": 0.01843855530023575,
+      "loss": 0.00832848995923996,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.44,
       "memory/max_allocated (GiB)": 33.44,
-      "ppl": 1.01861,
+      "ppl": 1.00836,
       "step": 82,
       "tokens/total": 2481216,
-      "tokens/train_per_sec_per_gpu": 33.37,
+      "tokens/train_per_sec_per_gpu": 33.4,
       "tokens/trainable": 37379
     },
     {
       "epoch": 0.32421875,
-      "grad_norm": 0.10422858595848083,
+      "grad_norm": 0.08372113853693008,
       "learning_rate": 9.699201747451195e-05,
-      "loss": 0.0014611330116167665,
+      "loss": 0.0007960593793541193,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00146,
+      "ppl": 1.0008,
       "step": 83,
       "tokens/total": 2511488,
-      "tokens/train_per_sec_per_gpu": 32.98,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 37832
     },
     {
       "epoch": 0.328125,
-      "grad_norm": 0.09028259664773941,
+      "grad_norm": 0.4056459367275238,
       "learning_rate": 9.688679307166854e-05,
-      "loss": 0.0017336590681225061,
+      "loss": 0.0015380227705463767,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00174,
+      "ppl": 1.00154,
       "step": 84,
       "tokens/total": 2541984,
-      "tokens/train_per_sec_per_gpu": 33.84,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 38303
     },
     {
       "epoch": 0.33203125,
-      "grad_norm": 0.5652539134025574,
+      "grad_norm": 0.1656680703163147,
       "learning_rate": 9.677982558839042e-05,
-      "loss": 0.004695026203989983,
+      "loss": 0.0007727849297225475,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00471,
+      "ppl": 1.00077,
       "step": 85,
       "tokens/total": 2572224,
-      "tokens/train_per_sec_per_gpu": 33.54,
+      "tokens/train_per_sec_per_gpu": 33.62,
       "tokens/trainable": 38758
     },
     {
       "epoch": 0.3359375,
-      "grad_norm": 0.14223186671733856,
+      "grad_norm": 1.7784565687179565,
       "learning_rate": 9.66711194760312e-05,
-      "loss": 0.002384063322097063,
+      "loss": 0.00887264870107174,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00239,
+      "ppl": 1.00891,
       "step": 86,
       "tokens/total": 2602704,
-      "tokens/train_per_sec_per_gpu": 30.54,
+      "tokens/train_per_sec_per_gpu": 30.62,
       "tokens/trainable": 39179
     },
     {
       "epoch": 0.33984375,
-      "grad_norm": 0.6704997420310974,
+      "grad_norm": 0.6024295687675476,
       "learning_rate": 9.656067925829593e-05,
-      "loss": 0.01066659577190876,
+      "loss": 0.013714440166950226,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01072,
+      "ppl": 1.01381,
       "step": 87,
       "tokens/total": 2633072,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.29,
       "tokens/trainable": 39679
     },
     {
       "epoch": 0.34375,
-      "grad_norm": 0.7846710085868835,
+      "grad_norm": 3.774395704269409,
       "learning_rate": 9.644850953105288e-05,
-      "loss": 0.015946906059980392,
+      "loss": 0.021049978211522102,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.01607,
+      "ppl": 1.02127,
       "step": 88,
       "tokens/total": 2663232,
-      "tokens/train_per_sec_per_gpu": 32.69,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 40118
     },
     {
       "epoch": 0.34765625,
-      "grad_norm": 0.1815856546163559,
+      "grad_norm": 0.20895183086395264,
       "learning_rate": 9.633461496214225e-05,
-      "loss": 0.002847571624442935,
+      "loss": 0.0025408377405256033,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00285,
+      "ppl": 1.00254,
       "step": 89,
       "tokens/total": 2693696,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 40571
     },
     {
       "epoch": 0.3515625,
-      "grad_norm": 0.1427551805973053,
+      "grad_norm": 0.3318273425102234,
       "learning_rate": 9.621900029118195e-05,
-      "loss": 0.0019866693764925003,
+      "loss": 0.0036354511976242065,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00199,
+      "ppl": 1.00364,
       "step": 90,
       "tokens/total": 2723984,
-      "tokens/train_per_sec_per_gpu": 30.76,
+      "tokens/train_per_sec_per_gpu": 30.75,
       "tokens/trainable": 40997
     },
     {
       "epoch": 0.35546875,
-      "grad_norm": 0.5249826908111572,
+      "grad_norm": 0.44360974431037903,
       "learning_rate": 9.610167032937036e-05,
-      "loss": 0.007045849692076445,
+      "loss": 0.007065630983561277,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00707,
+      "ppl": 1.00709,
       "step": 91,
       "tokens/total": 2754240,
-      "tokens/train_per_sec_per_gpu": 36.45,
+      "tokens/train_per_sec_per_gpu": 36.46,
       "tokens/trainable": 41462
     },
     {
       "epoch": 0.359375,
-      "grad_norm": 0.5962002277374268,
+      "grad_norm": 0.887543797492981,
       "learning_rate": 9.598262995928611e-05,
-      "loss": 0.006978678051382303,
+      "loss": 0.01106376200914383,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.007,
+      "ppl": 1.01113,
       "step": 92,
       "tokens/total": 2784672,
-      "tokens/train_per_sec_per_gpu": 36.81,
+      "tokens/train_per_sec_per_gpu": 36.79,
       "tokens/trainable": 41927
     },
     {
       "epoch": 0.36328125,
-      "grad_norm": 0.2272304892539978,
+      "grad_norm": 1.7098511457443237,
       "learning_rate": 9.586188413468492e-05,
-      "loss": 0.002275684382766485,
+      "loss": 0.02100781723856926,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00228,
+      "ppl": 1.02123,
       "step": 93,
       "tokens/total": 2815120,
-      "tokens/train_per_sec_per_gpu": 34.8,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 42397
     },
     {
       "epoch": 0.3671875,
-      "grad_norm": 0.04158688336610794,
+      "grad_norm": 0.05345788225531578,
       "learning_rate": 9.57394378802934e-05,
-      "loss": 0.0007734425598755479,
+      "loss": 0.0007664336590096354,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00077,
       "step": 94,
       "tokens/total": 2845600,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 42883
     },
     {
       "epoch": 0.37109375,
-      "grad_norm": 0.2428545504808426,
+      "grad_norm": 0.3566146492958069,
       "learning_rate": 9.56152962916e-05,
-      "loss": 0.003534857416525483,
+      "loss": 0.001947208889760077,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00354,
+      "ppl": 1.00195,
       "step": 95,
       "tokens/total": 2875952,
-      "tokens/train_per_sec_per_gpu": 38.65,
+      "tokens/train_per_sec_per_gpu": 38.67,
       "tokens/trainable": 43381
     },
     {
       "epoch": 0.375,
-      "grad_norm": 0.05061683431267738,
+      "grad_norm": 0.6130653023719788,
       "learning_rate": 9.548946453464296e-05,
-      "loss": 0.00045320752542465925,
+      "loss": 0.011625099927186966,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00045,
+      "ppl": 1.01169,
       "step": 96,
       "tokens/total": 2906288,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 43824
     },
     {
       "epoch": 0.37890625,
-      "grad_norm": 0.0691726952791214,
+      "grad_norm": 0.5786341428756714,
       "learning_rate": 9.53619478457953e-05,
-      "loss": 0.0004006598028354347,
+      "loss": 0.0046112253330647945,
       "memory/device_reserved (GiB)": 35.98,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0004,
+      "ppl": 1.00462,
       "step": 97,
       "tokens/total": 2936752,
-      "tokens/train_per_sec_per_gpu": 31.5,
+      "tokens/train_per_sec_per_gpu": 31.51,
       "tokens/trainable": 44255
     },
     {
       "epoch": 0.3828125,
-      "grad_norm": 0.20961223542690277,
+      "grad_norm": 0.4321880340576172,
       "learning_rate": 9.523275153154695e-05,
-      "loss": 0.0020726853981614113,
+      "loss": 0.003710552118718624,
       "memory/device_reserved (GiB)": 35.19,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00207,
+      "ppl": 1.00372,
       "step": 98,
       "tokens/total": 2967072,
-      "tokens/train_per_sec_per_gpu": 37.71,
+      "tokens/train_per_sec_per_gpu": 37.74,
       "tokens/trainable": 44739
     },
     {
       "epoch": 0.38671875,
-      "grad_norm": 0.07388679683208466,
+      "grad_norm": 0.16571089625358582,
       "learning_rate": 9.51018809682839e-05,
-      "loss": 0.00044117189827375114,
+      "loss": 0.00260853860527277,
       "memory/device_reserved (GiB)": 35.76,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
-      "ppl": 1.00044,
+      "ppl": 1.00261,
       "step": 99,
       "tokens/total": 2997664,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.2,
       "tokens/trainable": 45202
     },
     {
       "epoch": 0.390625,
-      "grad_norm": 0.46414947509765625,
+      "grad_norm": 0.4386274516582489,
       "learning_rate": 9.49693416020645e-05,
-      "loss": 0.006180304102599621,
+      "loss": 0.005652088671922684,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0062,
+      "ppl": 1.00567,
       "step": 100,
       "tokens/total": 3028032,
-      "tokens/train_per_sec_per_gpu": 33.02,
+      "tokens/train_per_sec_per_gpu": 33.05,
       "tokens/trainable": 45646
     },
     {
       "epoch": 0.39453125,
-      "grad_norm": 0.019027478992938995,
+      "grad_norm": 0.12225741147994995,
       "learning_rate": 9.483513894839276e-05,
-      "loss": 0.00012531192624010146,
+      "loss": 0.0015262231463566422,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00013,
+      "ppl": 1.00153,
       "step": 101,
       "tokens/total": 3058272,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 46089
     },
     {
       "epoch": 0.3984375,
-      "grad_norm": 3.6436362266540527,
+      "grad_norm": 0.779109537601471,
       "learning_rate": 9.469927859198888e-05,
-      "loss": 0.031218968331813812,
+      "loss": 0.007584188133478165,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.03171,
+      "ppl": 1.00761,
       "step": 102,
       "tokens/total": 3088848,
-      "tokens/train_per_sec_per_gpu": 32.49,
+      "tokens/train_per_sec_per_gpu": 32.55,
       "tokens/trainable": 46547
     },
     {
       "epoch": 0.40234375,
-      "grad_norm": 1.2344398498535156,
+      "grad_norm": 0.19508829712867737,
       "learning_rate": 9.456176618655689e-05,
-      "loss": 0.011161141097545624,
+      "loss": 0.0021933168172836304,
       "memory/device_reserved (GiB)": 35.99,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.01122,
+      "ppl": 1.0022,
       "step": 103,
       "tokens/total": 3119424,
-      "tokens/train_per_sec_per_gpu": 31.9,
+      "tokens/train_per_sec_per_gpu": 31.96,
       "tokens/trainable": 47014
     },
     {
       "epoch": 0.40625,
-      "grad_norm": 0.7120245099067688,
+      "grad_norm": 0.349365770816803,
       "learning_rate": 9.442260745454927e-05,
-      "loss": 0.002164291450753808,
+      "loss": 0.007655802182853222,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00217,
+      "ppl": 1.00769,
       "step": 104,
       "tokens/total": 3149696,
-      "tokens/train_per_sec_per_gpu": 37.65,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 47503
     },
     {
       "epoch": 0.41015625,
-      "grad_norm": 0.07052730768918991,
+      "grad_norm": 0.3205769956111908,
       "learning_rate": 9.428180818692884e-05,
-      "loss": 0.0002605066110845655,
+      "loss": 0.0014675151323899627,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00026,
+      "ppl": 1.00147,
       "step": 105,
       "tokens/total": 3180064,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 47996
     },
     {
       "epoch": 0.4140625,
-      "grad_norm": 1.2376325130462646,
+      "grad_norm": 0.23203934729099274,
       "learning_rate": 9.413937424292791e-05,
-      "loss": 0.01640205644071102,
+      "loss": 0.001969895325601101,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.01654,
+      "ppl": 1.00197,
       "step": 106,
       "tokens/total": 3208320,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 48447
     },
     {
       "epoch": 0.41796875,
-      "grad_norm": 0.5694106817245483,
+      "grad_norm": 0.050742585211992264,
       "learning_rate": 9.399531154980424e-05,
-      "loss": 0.01000689435750246,
+      "loss": 0.00028503642533905804,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.01006,
+      "ppl": 1.00029,
       "step": 107,
       "tokens/total": 3238416,
-      "tokens/train_per_sec_per_gpu": 30.65,
+      "tokens/train_per_sec_per_gpu": 30.69,
       "tokens/trainable": 48889
     },
     {
       "epoch": 0.421875,
-      "grad_norm": 0.5011488795280457,
+      "grad_norm": 0.68248450756073,
       "learning_rate": 9.384962610259455e-05,
-      "loss": 0.0022165768314152956,
+      "loss": 0.0026741819456219673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00222,
+      "ppl": 1.00268,
       "step": 108,
       "tokens/total": 3268784,
-      "tokens/train_per_sec_per_gpu": 37.84,
+      "tokens/train_per_sec_per_gpu": 37.89,
       "tokens/trainable": 49360
     },
     {
       "epoch": 0.42578125,
-      "grad_norm": 0.6332448124885559,
+      "grad_norm": 0.2664353549480438,
       "learning_rate": 9.370232396386494e-05,
-      "loss": 0.006433461792767048,
+      "loss": 0.0014279462629929185,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00645,
+      "ppl": 1.00143,
       "step": 109,
       "tokens/total": 3298912,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.78,
       "tokens/trainable": 49827
     },
     {
       "epoch": 0.4296875,
-      "grad_norm": 3.6466922760009766,
+      "grad_norm": 0.12221308797597885,
       "learning_rate": 9.355341126345868e-05,
-      "loss": 0.025773590430617332,
+      "loss": 0.0007402317132800817,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.02611,
+      "ppl": 1.00074,
       "step": 110,
       "tokens/total": 3329232,
-      "tokens/train_per_sec_per_gpu": 34.22,
+      "tokens/train_per_sec_per_gpu": 34.31,
       "tokens/trainable": 50267
     },
     {
       "epoch": 0.43359375,
-      "grad_norm": 1.0804238319396973,
+      "grad_norm": 1.2223896980285645,
       "learning_rate": 9.340289419824107e-05,
-      "loss": 0.013978044502437115,
+      "loss": 0.020261401310563087,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01408,
+      "ppl": 1.02047,
       "step": 111,
       "tokens/total": 3359440,
-      "tokens/train_per_sec_per_gpu": 33.32,
+      "tokens/train_per_sec_per_gpu": 33.41,
       "tokens/trainable": 50711
     },
     {
       "epoch": 0.4375,
-      "grad_norm": 0.3128775656223297,
+      "grad_norm": 1.2567111253738403,
       "learning_rate": 9.325077903184159e-05,
-      "loss": 0.008084455505013466,
+      "loss": 0.012774579226970673,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00812,
+      "ppl": 1.01286,
       "step": 112,
       "tokens/total": 3389472,
-      "tokens/train_per_sec_per_gpu": 31.54,
+      "tokens/train_per_sec_per_gpu": 31.59,
       "tokens/trainable": 51157
     },
     {
       "epoch": 0.44140625,
-      "grad_norm": 0.5562160611152649,
+      "grad_norm": 0.8144268989562988,
       "learning_rate": 9.30970720943932e-05,
-      "loss": 0.012518094852566719,
+      "loss": 0.005420445930212736,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0126,
+      "ppl": 1.00544,
       "step": 113,
       "tokens/total": 3419920,
-      "tokens/train_per_sec_per_gpu": 32.67,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 51613
     },
     {
       "epoch": 0.4453125,
-      "grad_norm": 0.37859252095222473,
+      "grad_norm": 1.3511962890625,
       "learning_rate": 9.2941779782269e-05,
-      "loss": 0.011980460025370121,
+      "loss": 0.004596072714775801,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01205,
+      "ppl": 1.00461,
       "step": 114,
       "tokens/total": 3450176,
-      "tokens/train_per_sec_per_gpu": 29.04,
+      "tokens/train_per_sec_per_gpu": 29.1,
       "tokens/trainable": 52017
     },
     {
       "epoch": 0.44921875,
-      "grad_norm": 0.19049690663814545,
+      "grad_norm": 2.76552677154541,
       "learning_rate": 9.278490855781596e-05,
-      "loss": 0.0033194865100085735,
+      "loss": 0.005162997171282768,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00333,
+      "ppl": 1.00518,
       "step": 115,
       "tokens/total": 3480544,
-      "tokens/train_per_sec_per_gpu": 36.15,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 52506
     },
     {
       "epoch": 0.453125,
-      "grad_norm": 0.9838894009590149,
+      "grad_norm": 1.7830004692077637,
       "learning_rate": 9.262646494908604e-05,
-      "loss": 0.011315119452774525,
+      "loss": 0.024823248386383057,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.01138,
+      "ppl": 1.02513,
       "step": 116,
       "tokens/total": 3510848,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.63,
       "tokens/trainable": 53008
     },
     {
       "epoch": 0.45703125,
-      "grad_norm": 1.5077804327011108,
+      "grad_norm": 0.8680487275123596,
       "learning_rate": 9.246645554956457e-05,
-      "loss": 0.011926110833883286,
+      "loss": 0.012252703309059143,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.012,
+      "ppl": 1.01233,
       "step": 117,
       "tokens/total": 3541296,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.36,
       "tokens/trainable": 53467
     },
     {
       "epoch": 0.4609375,
-      "grad_norm": 0.1702655702829361,
+      "grad_norm": 0.8030971884727478,
       "learning_rate": 9.230488701789578e-05,
-      "loss": 0.002886661561205983,
+      "loss": 0.01154071744531393,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00289,
+      "ppl": 1.01161,
       "step": 118,
       "tokens/total": 3571936,
-      "tokens/train_per_sec_per_gpu": 32.78,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 53923
     },
     {
       "epoch": 0.46484375,
-      "grad_norm": 0.3885493576526642,
+      "grad_norm": 1.1320112943649292,
       "learning_rate": 9.214176607760577e-05,
-      "loss": 0.0035083256661891937,
+      "loss": 0.00475259218364954,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00351,
+      "ppl": 1.00476,
       "step": 119,
       "tokens/total": 3602000,
       "tokens/train_per_sec_per_gpu": 35.58,
@@ -1677,125 +1677,125 @@
     },
     {
       "epoch": 0.46875,
-      "grad_norm": 0.2820606529712677,
+      "grad_norm": 0.25778594613075256,
       "learning_rate": 9.197709951682268e-05,
-      "loss": 0.005707860924303532,
+      "loss": 0.00691817793995142,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00572,
+      "ppl": 1.00694,
       "step": 120,
       "tokens/total": 3632288,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 54839
     },
     {
       "epoch": 0.47265625,
-      "grad_norm": 0.38794127106666565,
+      "grad_norm": 0.3193860948085785,
       "learning_rate": 9.181089418799428e-05,
-      "loss": 0.003157487604767084,
+      "loss": 0.0037792143411934376,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00316,
+      "ppl": 1.00379,
       "step": 121,
       "tokens/total": 3662608,
-      "tokens/train_per_sec_per_gpu": 32.58,
+      "tokens/train_per_sec_per_gpu": 32.59,
       "tokens/trainable": 55267
     },
     {
       "epoch": 0.4765625,
-      "grad_norm": 0.2677665054798126,
+      "grad_norm": 0.34393274784088135,
       "learning_rate": 9.164315700760271e-05,
-      "loss": 0.0032300231978297234,
+      "loss": 0.0042166681960225105,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00324,
+      "ppl": 1.00423,
       "step": 122,
       "tokens/total": 3692864,
-      "tokens/train_per_sec_per_gpu": 30.47,
+      "tokens/train_per_sec_per_gpu": 30.52,
       "tokens/trainable": 55700
     },
     {
       "epoch": 0.48046875,
-      "grad_norm": 0.31003710627555847,
+      "grad_norm": 0.4916401505470276,
       "learning_rate": 9.147389495587671e-05,
-      "loss": 0.00425739586353302,
+      "loss": 0.0038735629059374332,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00427,
+      "ppl": 1.00388,
       "step": 123,
       "tokens/total": 3722864,
-      "tokens/train_per_sec_per_gpu": 33.38,
+      "tokens/train_per_sec_per_gpu": 33.39,
       "tokens/trainable": 56144
     },
     {
       "epoch": 0.484375,
-      "grad_norm": 0.031137565150856972,
+      "grad_norm": 0.39921140670776367,
       "learning_rate": 9.130311507650116e-05,
-      "loss": 0.00010005584044847637,
+      "loss": 0.004010441247373819,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0001,
+      "ppl": 1.00402,
       "step": 124,
       "tokens/total": 3753056,
-      "tokens/train_per_sec_per_gpu": 33.07,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 56589
     },
     {
       "epoch": 0.48828125,
-      "grad_norm": 0.7360314130783081,
+      "grad_norm": 0.4556286633014679,
       "learning_rate": 9.113082447632394e-05,
-      "loss": 0.004621482454240322,
+      "loss": 0.005605827085673809,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00463,
+      "ppl": 1.00562,
       "step": 125,
       "tokens/total": 3783312,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.68,
       "tokens/trainable": 57081
     },
     {
       "epoch": 0.4921875,
-      "grad_norm": 0.05498598888516426,
+      "grad_norm": 0.041882507503032684,
       "learning_rate": 9.09570303250602e-05,
-      "loss": 0.0002661112230271101,
+      "loss": 0.0005784470704384148,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00027,
+      "ppl": 1.00058,
       "step": 126,
       "tokens/total": 3813712,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 57527
     },
     {
       "epoch": 0.49609375,
-      "grad_norm": 0.4160194993019104,
+      "grad_norm": 0.21823598444461823,
       "learning_rate": 9.078173985499394e-05,
-      "loss": 0.0034656142815947533,
+      "loss": 0.0024290401488542557,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00347,
+      "ppl": 1.00243,
       "step": 127,
       "tokens/total": 3844336,
-      "tokens/train_per_sec_per_gpu": 34.79,
+      "tokens/train_per_sec_per_gpu": 34.84,
       "tokens/trainable": 58005
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.44116270542144775,
+      "grad_norm": 0.4637400209903717,
       "learning_rate": 9.060496036067713e-05,
-      "loss": 0.0029591654893010855,
+      "loss": 0.001603663433343172,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00296,
+      "ppl": 1.0016,
       "step": 128,
       "tokens/total": 3872688,
       "tokens/train_per_sec_per_gpu": 35.21,
@@ -1803,223 +1803,223 @@
     },
     {
       "epoch": 0.50390625,
-      "grad_norm": 0.29427066445350647,
+      "grad_norm": 0.03311942145228386,
       "learning_rate": 9.042669919862615e-05,
-      "loss": 0.002471367595717311,
+      "loss": 0.00027762592071667314,
       "memory/device_reserved (GiB)": 36.0,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.00247,
+      "ppl": 1.00028,
       "step": 129,
       "tokens/total": 3902640,
-      "tokens/train_per_sec_per_gpu": 31.94,
+      "tokens/train_per_sec_per_gpu": 31.97,
       "tokens/trainable": 58896
     },
     {
       "epoch": 0.5078125,
-      "grad_norm": 0.024222422391176224,
+      "grad_norm": 0.019396035000681877,
       "learning_rate": 9.024696378701557e-05,
-      "loss": 0.00013285802560858428,
+      "loss": 8.183176396414638e-05,
       "memory/device_reserved (GiB)": 34.68,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00008,
       "step": 130,
       "tokens/total": 3933104,
-      "tokens/train_per_sec_per_gpu": 35.79,
+      "tokens/train_per_sec_per_gpu": 35.8,
       "tokens/trainable": 59356
     },
     {
       "epoch": 0.51171875,
-      "grad_norm": 3.2403194904327393,
+      "grad_norm": 0.3952345848083496,
       "learning_rate": 9.006576160536948e-05,
-      "loss": 0.013489946722984314,
+      "loss": 0.005079061258584261,
       "memory/device_reserved (GiB)": 34.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.01358,
+      "ppl": 1.00509,
       "step": 131,
       "tokens/total": 3963184,
-      "tokens/train_per_sec_per_gpu": 32.14,
+      "tokens/train_per_sec_per_gpu": 32.21,
       "tokens/trainable": 59792
     },
     {
       "epoch": 0.515625,
-      "grad_norm": 0.14770789444446564,
+      "grad_norm": 0.03258282318711281,
       "learning_rate": 8.988310019425035e-05,
-      "loss": 0.0009925956837832928,
+      "loss": 0.00015263023669831455,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00099,
+      "ppl": 1.00015,
       "step": 132,
       "tokens/total": 3993648,
-      "tokens/train_per_sec_per_gpu": 35.66,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 60266
     },
     {
       "epoch": 0.51953125,
-      "grad_norm": 1.2149583101272583,
+      "grad_norm": 0.10869178175926208,
       "learning_rate": 8.969898715494506e-05,
-      "loss": 0.005480521358549595,
+      "loss": 0.00033613567939028144,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0055,
+      "ppl": 1.00034,
       "step": 133,
       "tokens/total": 4024000,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.44,
       "tokens/trainable": 60759
     },
     {
       "epoch": 0.5234375,
-      "grad_norm": 0.24377752840518951,
+      "grad_norm": 0.016073748469352722,
       "learning_rate": 8.951343014914869e-05,
-      "loss": 0.0011378922499716282,
+      "loss": 0.00014354957966133952,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.39,
       "memory/max_allocated (GiB)": 33.39,
-      "ppl": 1.00114,
+      "ppl": 1.00014,
       "step": 134,
       "tokens/total": 4052480,
-      "tokens/train_per_sec_per_gpu": 31.71,
+      "tokens/train_per_sec_per_gpu": 31.76,
       "tokens/trainable": 61204
     },
     {
       "epoch": 0.52734375,
-      "grad_norm": 0.8190546631813049,
+      "grad_norm": 0.5635335445404053,
       "learning_rate": 8.932643689864568e-05,
-      "loss": 0.004931692034006119,
+      "loss": 0.0033080782741308212,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00494,
+      "ppl": 1.00331,
       "step": 135,
       "tokens/total": 4082864,
-      "tokens/train_per_sec_per_gpu": 35.88,
+      "tokens/train_per_sec_per_gpu": 35.93,
       "tokens/trainable": 61684
     },
     {
       "epoch": 0.53125,
-      "grad_norm": 0.7444491982460022,
+      "grad_norm": 0.12933659553527832,
       "learning_rate": 8.913801518498845e-05,
-      "loss": 0.00953812524676323,
+      "loss": 0.0003551081463228911,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.7,
       "memory/max_allocated (GiB)": 33.7,
-      "ppl": 1.00958,
+      "ppl": 1.00036,
       "step": 136,
       "tokens/total": 4113040,
-      "tokens/train_per_sec_per_gpu": 38.18,
+      "tokens/train_per_sec_per_gpu": 38.22,
       "tokens/trainable": 62159
     },
     {
       "epoch": 0.53515625,
-      "grad_norm": 0.017797501757740974,
+      "grad_norm": 0.038116879761219025,
       "learning_rate": 8.894817284917364e-05,
-      "loss": 0.00010352014214731753,
+      "loss": 0.0002933211508207023,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0001,
+      "ppl": 1.00029,
       "step": 137,
       "tokens/total": 4143344,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.28,
       "tokens/trainable": 62649
     },
     {
       "epoch": 0.5390625,
-      "grad_norm": 0.14133845269680023,
+      "grad_norm": 0.4708496928215027,
       "learning_rate": 8.875691779131569e-05,
-      "loss": 0.0006721644895151258,
+      "loss": 0.009557873010635376,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00067,
+      "ppl": 1.0096,
       "step": 138,
       "tokens/total": 4173648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.65,
       "tokens/trainable": 63100
     },
     {
       "epoch": 0.54296875,
-      "grad_norm": 0.7884437441825867,
+      "grad_norm": 0.6401950120925903,
       "learning_rate": 8.856425797031829e-05,
-      "loss": 0.0070602125488221645,
+      "loss": 0.006467439234256744,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00709,
+      "ppl": 1.00649,
       "step": 139,
       "tokens/total": 4204080,
-      "tokens/train_per_sec_per_gpu": 37.51,
+      "tokens/train_per_sec_per_gpu": 37.56,
       "tokens/trainable": 63588
     },
     {
       "epoch": 0.546875,
-      "grad_norm": 2.1515400409698486,
+      "grad_norm": 0.9977340698242188,
       "learning_rate": 8.837020140354295e-05,
-      "loss": 0.04659099504351616,
+      "loss": 0.01085682213306427,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.42,
       "memory/max_allocated (GiB)": 33.42,
-      "ppl": 1.04769,
+      "ppl": 1.01092,
       "step": 140,
       "tokens/total": 4232400,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 64005
     },
     {
       "epoch": 0.55078125,
-      "grad_norm": 0.9182130098342896,
+      "grad_norm": 0.9310302734375,
       "learning_rate": 8.817475616647554e-05,
-      "loss": 0.01849653385579586,
+      "loss": 0.002832747297361493,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.01867,
+      "ppl": 1.00284,
       "step": 141,
       "tokens/total": 4262624,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 64462
     },
     {
       "epoch": 0.5546875,
-      "grad_norm": 0.049818869680166245,
+      "grad_norm": 0.8448959589004517,
       "learning_rate": 8.797793039239017e-05,
-      "loss": 0.0007528763962909579,
+      "loss": 0.00225994479842484,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00075,
+      "ppl": 1.00226,
       "step": 142,
       "tokens/total": 4293168,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.93,
       "tokens/trainable": 64900
     },
     {
       "epoch": 0.55859375,
-      "grad_norm": 0.4962376654148102,
+      "grad_norm": 0.1500774621963501,
       "learning_rate": 8.777973227201069e-05,
-      "loss": 0.013024947606027126,
+      "loss": 0.001096544205211103,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01311,
+      "ppl": 1.0011,
       "step": 143,
       "tokens/total": 4323248,
-      "tokens/train_per_sec_per_gpu": 37.45,
+      "tokens/train_per_sec_per_gpu": 37.52,
       "tokens/trainable": 65379
     },
     {
       "epoch": 0.5625,
-      "grad_norm": 0.09980953484773636,
+      "grad_norm": 0.7080555558204651,
       "learning_rate": 8.758017005316988e-05,
-      "loss": 0.0013438962632790208,
+      "loss": 0.02480880171060562,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00134,
+      "ppl": 1.02512,
       "step": 144,
       "tokens/total": 4353344,
       "tokens/train_per_sec_per_gpu": 36.69,
@@ -2027,181 +2027,181 @@
     },
     {
       "epoch": 0.56640625,
-      "grad_norm": 0.2499600648880005,
+      "grad_norm": 0.8302886486053467,
       "learning_rate": 8.737925204046629e-05,
-      "loss": 0.005669259466230869,
+      "loss": 0.011632976122200489,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00569,
+      "ppl": 1.0117,
       "step": 145,
       "tokens/total": 4383584,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 66285
     },
     {
       "epoch": 0.5703125,
-      "grad_norm": 0.8080663084983826,
+      "grad_norm": 0.48007094860076904,
       "learning_rate": 8.717698659491851e-05,
-      "loss": 0.03145892918109894,
+      "loss": 0.006236276589334011,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.03196,
+      "ppl": 1.00626,
       "step": 146,
       "tokens/total": 4413856,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 66736
     },
     {
       "epoch": 0.57421875,
-      "grad_norm": 0.29324567317962646,
+      "grad_norm": 0.2671569585800171,
       "learning_rate": 8.697338213361735e-05,
-      "loss": 0.008183630183339119,
+      "loss": 0.004551246762275696,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00822,
+      "ppl": 1.00456,
       "step": 147,
       "tokens/total": 4444112,
-      "tokens/train_per_sec_per_gpu": 35.2,
+      "tokens/train_per_sec_per_gpu": 35.26,
       "tokens/trainable": 67150
     },
     {
       "epoch": 0.578125,
-      "grad_norm": 0.5702983736991882,
+      "grad_norm": 0.2333666980266571,
       "learning_rate": 8.676844712937552e-05,
-      "loss": 0.00848381593823433,
+      "loss": 0.005151683464646339,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00852,
+      "ppl": 1.00516,
       "step": 148,
       "tokens/total": 4474464,
-      "tokens/train_per_sec_per_gpu": 35.02,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 67633
     },
     {
       "epoch": 0.58203125,
-      "grad_norm": 0.12609520554542542,
+      "grad_norm": 1.3599361181259155,
       "learning_rate": 8.656219011037509e-05,
-      "loss": 0.002346089808270335,
+      "loss": 0.010383867658674717,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00235,
+      "ppl": 1.01044,
       "step": 149,
       "tokens/total": 4504880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.82,
       "tokens/trainable": 68077
     },
     {
       "epoch": 0.5859375,
-      "grad_norm": 0.044832926243543625,
+      "grad_norm": 0.2275276631116867,
       "learning_rate": 8.63546196598125e-05,
-      "loss": 0.000866610265802592,
+      "loss": 0.0016249704640358686,
       "memory/device_reserved (GiB)": 35.79,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00087,
+      "ppl": 1.00163,
       "step": 150,
       "tokens/total": 4534976,
-      "tokens/train_per_sec_per_gpu": 33.21,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 68545
     },
     {
       "epoch": 0.58984375,
-      "grad_norm": 0.2726699113845825,
+      "grad_norm": 0.6304328441619873,
       "learning_rate": 8.614574441554145e-05,
-      "loss": 0.010688753798604012,
+      "loss": 0.02714025229215622,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.01075,
+      "ppl": 1.02751,
       "step": 151,
       "tokens/total": 4565088,
-      "tokens/train_per_sec_per_gpu": 30.17,
+      "tokens/train_per_sec_per_gpu": 30.26,
       "tokens/trainable": 69011
     },
     {
       "epoch": 0.59375,
-      "grad_norm": 0.6334059238433838,
+      "grad_norm": 0.3306228518486023,
       "learning_rate": 8.593557306971349e-05,
-      "loss": 0.007179228588938713,
+      "loss": 0.006482407916337252,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00721,
+      "ppl": 1.0065,
       "step": 152,
       "tokens/total": 4595600,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 69465
     },
     {
       "epoch": 0.59765625,
-      "grad_norm": 0.6207929849624634,
+      "grad_norm": 2.509195566177368,
       "learning_rate": 8.572411436841618e-05,
-      "loss": 0.005558713339269161,
+      "loss": 0.014876915141940117,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00557,
+      "ppl": 1.01499,
       "step": 153,
       "tokens/total": 4626032,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.89,
       "tokens/trainable": 69928
     },
     {
       "epoch": 0.6015625,
-      "grad_norm": 0.14352096617221832,
+      "grad_norm": 0.08186601847410202,
       "learning_rate": 8.551137711130922e-05,
-      "loss": 0.0009000393329188228,
+      "loss": 0.0014128303155303001,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.0009,
+      "ppl": 1.00141,
       "step": 154,
       "tokens/total": 4656352,
-      "tokens/train_per_sec_per_gpu": 36.64,
+      "tokens/train_per_sec_per_gpu": 36.67,
       "tokens/trainable": 70395
     },
     {
       "epoch": 0.60546875,
-      "grad_norm": 0.37536460161209106,
+      "grad_norm": 0.20300441980361938,
       "learning_rate": 8.529737015125824e-05,
-      "loss": 0.0032662933226674795,
+      "loss": 0.0016238696407526731,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00327,
+      "ppl": 1.00163,
       "step": 155,
       "tokens/total": 4684672,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.58,
       "tokens/trainable": 70829
     },
     {
       "epoch": 0.609375,
-      "grad_norm": 0.8182935118675232,
+      "grad_norm": 0.10475975275039673,
       "learning_rate": 8.508210239396639e-05,
-      "loss": 0.015120145864784718,
+      "loss": 0.0014675778802484274,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.01524,
+      "ppl": 1.00147,
       "step": 156,
       "tokens/total": 4715216,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.88,
       "tokens/trainable": 71287
     },
     {
       "epoch": 0.61328125,
-      "grad_norm": 0.05925029143691063,
+      "grad_norm": 0.2740918695926666,
       "learning_rate": 8.486558279760375e-05,
-      "loss": 0.00020593182125594467,
+      "loss": 0.004101475700736046,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00021,
+      "ppl": 1.00411,
       "step": 157,
       "tokens/total": 4745376,
       "tokens/train_per_sec_per_gpu": 34.95,
@@ -2209,405 +2209,405 @@
     },
     {
       "epoch": 0.6171875,
-      "grad_norm": 0.5880815982818604,
+      "grad_norm": 0.30794957280158997,
       "learning_rate": 8.464782037243449e-05,
-      "loss": 0.011222876608371735,
+      "loss": 0.0044078221544623375,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.01129,
+      "ppl": 1.00442,
       "step": 158,
       "tokens/total": 4775776,
-      "tokens/train_per_sec_per_gpu": 32.03,
+      "tokens/train_per_sec_per_gpu": 32.18,
       "tokens/trainable": 72185
     },
     {
       "epoch": 0.62109375,
-      "grad_norm": 0.1880187839269638,
+      "grad_norm": 0.3344379961490631,
       "learning_rate": 8.442882418044202e-05,
-      "loss": 0.0014226734638214111,
+      "loss": 0.0035248196218162775,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00142,
+      "ppl": 1.00353,
       "step": 159,
       "tokens/total": 4805856,
-      "tokens/train_per_sec_per_gpu": 30.51,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 72604
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.8338136076927185,
+      "grad_norm": 0.6211779117584229,
       "learning_rate": 8.420860333495179e-05,
-      "loss": 0.018916597589850426,
+      "loss": 0.015572377480566502,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0191,
+      "ppl": 1.01569,
       "step": 160,
       "tokens/total": 4836304,
-      "tokens/train_per_sec_per_gpu": 31.41,
+      "tokens/train_per_sec_per_gpu": 31.56,
       "tokens/trainable": 73061
     },
     {
       "epoch": 0.62890625,
-      "grad_norm": 0.11151792109012604,
+      "grad_norm": 0.5694647431373596,
       "learning_rate": 8.398716700025208e-05,
-      "loss": 0.001152656739577651,
+      "loss": 0.005395190790295601,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.00115,
+      "ppl": 1.00541,
       "step": 161,
       "tokens/total": 4866544,
-      "tokens/train_per_sec_per_gpu": 29.97,
+      "tokens/train_per_sec_per_gpu": 29.98,
       "tokens/trainable": 73474
     },
     {
       "epoch": 0.6328125,
-      "grad_norm": 0.319916695356369,
+      "grad_norm": 0.3221203088760376,
       "learning_rate": 8.376452439121266e-05,
-      "loss": 0.0030798939988017082,
+      "loss": 0.004907120484858751,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00308,
+      "ppl": 1.00492,
       "step": 162,
       "tokens/total": 4897120,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.65,
       "tokens/trainable": 73923
     },
     {
       "epoch": 0.63671875,
-      "grad_norm": 0.23228555917739868,
+      "grad_norm": 0.2562556862831116,
       "learning_rate": 8.354068477290124e-05,
-      "loss": 0.0029990714974701405,
+      "loss": 0.011767375282943249,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.003,
+      "ppl": 1.01184,
       "step": 163,
       "tokens/total": 4927440,
-      "tokens/train_per_sec_per_gpu": 35.29,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 74379
     },
     {
       "epoch": 0.640625,
-      "grad_norm": 0.16945457458496094,
+      "grad_norm": 0.37376344203948975,
       "learning_rate": 8.331565746019807e-05,
-      "loss": 0.0036612222902476788,
+      "loss": 0.014833642169833183,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00367,
+      "ppl": 1.01494,
       "step": 164,
       "tokens/total": 4958032,
-      "tokens/train_per_sec_per_gpu": 33.33,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 74830
     },
     {
       "epoch": 0.64453125,
-      "grad_norm": 0.08089865744113922,
+      "grad_norm": 0.07679534703493118,
       "learning_rate": 8.308945181740812e-05,
-      "loss": 0.0011908983578905463,
+      "loss": 0.0006800079718232155,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00119,
+      "ppl": 1.00068,
       "step": 165,
       "tokens/total": 4988368,
-      "tokens/train_per_sec_per_gpu": 31.57,
+      "tokens/train_per_sec_per_gpu": 31.73,
       "tokens/trainable": 75262
     },
     {
       "epoch": 0.6484375,
-      "grad_norm": 0.06332889199256897,
+      "grad_norm": 0.15940658748149872,
       "learning_rate": 8.286207725787153e-05,
-      "loss": 0.0010977290803566575,
+      "loss": 0.0005167147610336542,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
-      "ppl": 1.0011,
+      "ppl": 1.00052,
       "step": 166,
       "tokens/total": 5018672,
-      "tokens/train_per_sec_per_gpu": 31.75,
+      "tokens/train_per_sec_per_gpu": 31.84,
       "tokens/trainable": 75703
     },
     {
       "epoch": 0.65234375,
-      "grad_norm": 0.058639127761125565,
+      "grad_norm": 0.18865765631198883,
       "learning_rate": 8.263354324357182e-05,
-      "loss": 0.000854893063660711,
+      "loss": 0.003161062952131033,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00086,
+      "ppl": 1.00317,
       "step": 167,
       "tokens/total": 5049344,
-      "tokens/train_per_sec_per_gpu": 31.26,
+      "tokens/train_per_sec_per_gpu": 31.3,
       "tokens/trainable": 76164
     },
     {
       "epoch": 0.65625,
-      "grad_norm": 0.19198745489120483,
+      "grad_norm": 0.3047192096710205,
       "learning_rate": 8.240385928474219e-05,
-      "loss": 0.0024191238917410374,
+      "loss": 0.010941199027001858,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00242,
+      "ppl": 1.011,
       "step": 168,
       "tokens/total": 5079856,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 76623
     },
     {
       "epoch": 0.66015625,
-      "grad_norm": 0.033948056399822235,
+      "grad_norm": 0.39291632175445557,
       "learning_rate": 8.217303493946967e-05,
-      "loss": 0.000675962888635695,
+      "loss": 0.009001722559332848,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00068,
+      "ppl": 1.00904,
       "step": 169,
       "tokens/total": 5110192,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 77047
     },
     {
       "epoch": 0.6640625,
-      "grad_norm": 0.10515601187944412,
+      "grad_norm": 0.09876111894845963,
       "learning_rate": 8.194107981329746e-05,
-      "loss": 0.0006985474610701203,
+      "loss": 0.0014627808704972267,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0007,
+      "ppl": 1.00146,
       "step": 170,
       "tokens/total": 5140592,
-      "tokens/train_per_sec_per_gpu": 33.98,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 77496
     },
     {
       "epoch": 0.66796875,
-      "grad_norm": 0.01597985066473484,
+      "grad_norm": 0.02293679490685463,
       "learning_rate": 8.170800355882518e-05,
-      "loss": 0.0002490747720003128,
+      "loss": 0.00041422739741392434,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 171,
       "tokens/total": 5170848,
-      "tokens/train_per_sec_per_gpu": 33.13,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 77956
     },
     {
       "epoch": 0.671875,
-      "grad_norm": 0.28580594062805176,
+      "grad_norm": 0.2762586176395416,
       "learning_rate": 8.147381587530713e-05,
-      "loss": 0.002719157375395298,
+      "loss": 0.006397426128387451,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00272,
+      "ppl": 1.00642,
       "step": 172,
       "tokens/total": 5201280,
-      "tokens/train_per_sec_per_gpu": 29.75,
+      "tokens/train_per_sec_per_gpu": 29.78,
       "tokens/trainable": 78410
     },
     {
       "epoch": 0.67578125,
-      "grad_norm": 0.027172744274139404,
+      "grad_norm": 0.07085710763931274,
       "learning_rate": 8.123852650824877e-05,
-      "loss": 0.0005204043700359762,
+      "loss": 0.0017267607618123293,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00052,
+      "ppl": 1.00173,
       "step": 173,
       "tokens/total": 5231712,
-      "tokens/train_per_sec_per_gpu": 33.04,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 78849
     },
     {
       "epoch": 0.6796875,
-      "grad_norm": 2.8796133995056152,
+      "grad_norm": 0.4523530602455139,
       "learning_rate": 8.100214524900103e-05,
-      "loss": 0.015293785370886326,
+      "loss": 0.011936807073652744,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.62,
       "memory/max_allocated (GiB)": 33.62,
-      "ppl": 1.01541,
+      "ppl": 1.01201,
       "step": 174,
       "tokens/total": 5261440,
-      "tokens/train_per_sec_per_gpu": 29.53,
+      "tokens/train_per_sec_per_gpu": 30.07,
       "tokens/trainable": 79266
     },
     {
       "epoch": 0.68359375,
-      "grad_norm": 0.1005098894238472,
+      "grad_norm": 0.1293293535709381,
       "learning_rate": 8.076468193435301e-05,
-      "loss": 0.0010334225371479988,
+      "loss": 0.0021572443656623363,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00103,
+      "ppl": 1.00216,
       "step": 175,
       "tokens/total": 5291728,
-      "tokens/train_per_sec_per_gpu": 32.42,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 79736
     },
     {
       "epoch": 0.6875,
-      "grad_norm": 0.6506177186965942,
+      "grad_norm": 0.13209661841392517,
       "learning_rate": 8.052614644612253e-05,
-      "loss": 0.0038155291695147753,
+      "loss": 0.0017078389646485448,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00382,
+      "ppl": 1.00171,
       "step": 176,
       "tokens/total": 5322240,
-      "tokens/train_per_sec_per_gpu": 36.12,
+      "tokens/train_per_sec_per_gpu": 36.17,
       "tokens/trainable": 80214
     },
     {
       "epoch": 0.69140625,
-      "grad_norm": 0.19899244606494904,
+      "grad_norm": 0.09734465181827545,
       "learning_rate": 8.028654871074489e-05,
-      "loss": 0.0020266459323465824,
+      "loss": 0.0020819404162466526,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00203,
+      "ppl": 1.00208,
       "step": 177,
       "tokens/total": 5352576,
-      "tokens/train_per_sec_per_gpu": 32.51,
+      "tokens/train_per_sec_per_gpu": 32.56,
       "tokens/trainable": 80661
     },
     {
       "epoch": 0.6953125,
-      "grad_norm": 0.009929690510034561,
+      "grad_norm": 0.8079077005386353,
       "learning_rate": 8.004589869885986e-05,
-      "loss": 0.00014211784582585096,
+      "loss": 0.006122267805039883,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00014,
+      "ppl": 1.00614,
       "step": 178,
       "tokens/total": 5382784,
-      "tokens/train_per_sec_per_gpu": 34.81,
+      "tokens/train_per_sec_per_gpu": 34.82,
       "tokens/trainable": 81129
     },
     {
       "epoch": 0.69921875,
-      "grad_norm": 0.5209382176399231,
+      "grad_norm": 0.806120753288269,
       "learning_rate": 7.980420642489674e-05,
-      "loss": 0.023224791511893272,
+      "loss": 0.003698066109791398,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0235,
+      "ppl": 1.0037,
       "step": 179,
       "tokens/total": 5412784,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 81576
     },
     {
       "epoch": 0.703125,
-      "grad_norm": 0.02280755713582039,
+      "grad_norm": 0.0535414032638073,
       "learning_rate": 7.95614819466576e-05,
-      "loss": 0.0002463866549078375,
+      "loss": 0.0007224087603390217,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00025,
+      "ppl": 1.00072,
       "step": 180,
       "tokens/total": 5442960,
-      "tokens/train_per_sec_per_gpu": 34.49,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 82049
     },
     {
       "epoch": 0.70703125,
-      "grad_norm": 0.3769514262676239,
+      "grad_norm": 0.8787689208984375,
       "learning_rate": 7.931773536489872e-05,
-      "loss": 0.01167471706867218,
+      "loss": 0.014041105285286903,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.01174,
+      "ppl": 1.01414,
       "step": 181,
       "tokens/total": 5473216,
-      "tokens/train_per_sec_per_gpu": 38.75,
+      "tokens/train_per_sec_per_gpu": 38.79,
       "tokens/trainable": 82566
     },
     {
       "epoch": 0.7109375,
-      "grad_norm": 0.060761548578739166,
+      "grad_norm": 0.23288801312446594,
       "learning_rate": 7.907297682291035e-05,
-      "loss": 0.000216186250327155,
+      "loss": 0.0018348076846450567,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00022,
+      "ppl": 1.00184,
       "step": 182,
       "tokens/total": 5503568,
-      "tokens/train_per_sec_per_gpu": 33.83,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 83054
     },
     {
       "epoch": 0.71484375,
-      "grad_norm": 0.03373813256621361,
+      "grad_norm": 0.08039269596338272,
       "learning_rate": 7.882721650609442e-05,
-      "loss": 0.00041355923167429864,
+      "loss": 0.0011233665281906724,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00041,
+      "ppl": 1.00112,
       "step": 183,
       "tokens/total": 5533760,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 83501
     },
     {
       "epoch": 0.71875,
-      "grad_norm": 0.036661963909864426,
+      "grad_norm": 0.028009524568915367,
       "learning_rate": 7.85804646415409e-05,
-      "loss": 0.00041001036879606545,
+      "loss": 0.00045670082909055054,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00041,
+      "ppl": 1.00046,
       "step": 184,
       "tokens/total": 5564240,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 84007
     },
     {
       "epoch": 0.72265625,
-      "grad_norm": 0.03234419599175453,
+      "grad_norm": 0.04383641108870506,
       "learning_rate": 7.833273149760207e-05,
-      "loss": 0.0003714286140166223,
+      "loss": 0.00067467603366822,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00037,
+      "ppl": 1.00067,
       "step": 185,
       "tokens/total": 5594448,
-      "tokens/train_per_sec_per_gpu": 35.39,
+      "tokens/train_per_sec_per_gpu": 35.54,
       "tokens/trainable": 84491
     },
     {
       "epoch": 0.7265625,
-      "grad_norm": 0.03076508454978466,
+      "grad_norm": 0.1170438677072525,
       "learning_rate": 7.808402738346527e-05,
-      "loss": 0.0003759118844754994,
+      "loss": 0.0014282094780355692,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00038,
+      "ppl": 1.00143,
       "step": 186,
       "tokens/total": 5624816,
       "tokens/train_per_sec_per_gpu": 33.33,
@@ -2615,181 +2615,181 @@
     },
     {
       "epoch": 0.73046875,
-      "grad_norm": 0.6632063388824463,
+      "grad_norm": 0.05614401772618294,
       "learning_rate": 7.783436264872382e-05,
-      "loss": 0.002438098657876253,
+      "loss": 0.0002470523177180439,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00244,
+      "ppl": 1.00025,
       "step": 187,
       "tokens/total": 5655104,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.92,
       "tokens/trainable": 85400
     },
     {
       "epoch": 0.734375,
-      "grad_norm": 0.2651624083518982,
+      "grad_norm": 0.30610957741737366,
       "learning_rate": 7.758374768294647e-05,
-      "loss": 0.002341378014534712,
+      "loss": 0.004403322469443083,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00234,
+      "ppl": 1.00441,
       "step": 188,
       "tokens/total": 5685360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 85889
     },
     {
       "epoch": 0.73828125,
-      "grad_norm": 0.305477112531662,
+      "grad_norm": 0.433325856924057,
       "learning_rate": 7.733219291524489e-05,
-      "loss": 0.0016374983824789524,
+      "loss": 0.005490327253937721,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00164,
+      "ppl": 1.00551,
       "step": 189,
       "tokens/total": 5715648,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 86396
     },
     {
       "epoch": 0.7421875,
-      "grad_norm": 0.2344018816947937,
+      "grad_norm": 0.4976643919944763,
       "learning_rate": 7.707970881383977e-05,
-      "loss": 0.002069709589704871,
+      "loss": 0.006626967806369066,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00207,
+      "ppl": 1.00665,
       "step": 190,
       "tokens/total": 5746240,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.85,
       "tokens/trainable": 86867
     },
     {
       "epoch": 0.74609375,
-      "grad_norm": 0.09677010774612427,
+      "grad_norm": 1.0000759363174438,
       "learning_rate": 7.682630588562518e-05,
-      "loss": 0.0006176180904731154,
+      "loss": 0.009028773754835129,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00062,
+      "ppl": 1.00907,
       "step": 191,
       "tokens/total": 5776816,
-      "tokens/train_per_sec_per_gpu": 36.03,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 87314
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.05828038975596428,
+      "grad_norm": 0.006456763483583927,
       "learning_rate": 7.657199467573129e-05,
-      "loss": 0.0004249829798936844,
+      "loss": 5.810559741803445e-05,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00043,
+      "ppl": 1.00006,
       "step": 192,
       "tokens/total": 5807248,
-      "tokens/train_per_sec_per_gpu": 30.29,
+      "tokens/train_per_sec_per_gpu": 30.38,
       "tokens/trainable": 87736
     },
     {
       "epoch": 0.75390625,
-      "grad_norm": 0.09240972250699997,
+      "grad_norm": 0.6312516331672668,
       "learning_rate": 7.631678576708561e-05,
-      "loss": 0.0003209684509783983,
+      "loss": 0.0015132949920371175,
       "memory/device_reserved (GiB)": 36.02,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00032,
+      "ppl": 1.00151,
       "step": 193,
       "tokens/total": 5837680,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 88219
     },
     {
       "epoch": 0.7578125,
-      "grad_norm": 0.00590590201318264,
+      "grad_norm": 0.003887161612510681,
       "learning_rate": 7.606068977997255e-05,
-      "loss": 7.823687337804586e-05,
+      "loss": 3.5049586585955694e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00004,
       "step": 194,
       "tokens/total": 5868000,
-      "tokens/train_per_sec_per_gpu": 36.11,
+      "tokens/train_per_sec_per_gpu": 36.15,
       "tokens/trainable": 88699
     },
     {
       "epoch": 0.76171875,
-      "grad_norm": 0.45261114835739136,
+      "grad_norm": 0.05365428328514099,
       "learning_rate": 7.580371737159148e-05,
-      "loss": 0.0019616044592112303,
+      "loss": 0.00017142925935331732,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00196,
+      "ppl": 1.00017,
       "step": 195,
       "tokens/total": 5896224,
-      "tokens/train_per_sec_per_gpu": 30.36,
+      "tokens/train_per_sec_per_gpu": 30.39,
       "tokens/trainable": 89132
     },
     {
       "epoch": 0.765625,
-      "grad_norm": 0.03762947395443916,
+      "grad_norm": 1.5210427045822144,
       "learning_rate": 7.554587923561324e-05,
-      "loss": 0.0004910373827442527,
+      "loss": 0.007072226610034704,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00049,
+      "ppl": 1.0071,
       "step": 196,
       "tokens/total": 5926624,
-      "tokens/train_per_sec_per_gpu": 37.21,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 89594
     },
     {
       "epoch": 0.76953125,
-      "grad_norm": 0.37914252281188965,
+      "grad_norm": 0.01594001241028309,
       "learning_rate": 7.528718610173511e-05,
-      "loss": 0.005961633287370205,
+      "loss": 0.00012487987987697124,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.34,
       "memory/max_allocated (GiB)": 33.34,
-      "ppl": 1.00598,
+      "ppl": 1.00012,
       "step": 197,
       "tokens/total": 5954944,
-      "tokens/train_per_sec_per_gpu": 37.98,
+      "tokens/train_per_sec_per_gpu": 38.08,
       "tokens/trainable": 90042
     },
     {
       "epoch": 0.7734375,
-      "grad_norm": 0.061283551156520844,
+      "grad_norm": 0.48730939626693726,
       "learning_rate": 7.502764873523431e-05,
-      "loss": 0.0004046210669912398,
+      "loss": 0.004653760232031345,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0004,
+      "ppl": 1.00466,
       "step": 198,
       "tokens/total": 5985200,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 90503
     },
     {
       "epoch": 0.77734375,
-      "grad_norm": 0.012014869600534439,
+      "grad_norm": 0.06984082609415054,
       "learning_rate": 7.476727793652011e-05,
-      "loss": 0.00018992810510098934,
+      "loss": 0.0007038296316750348,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00019,
+      "ppl": 1.0007,
       "step": 199,
       "tokens/total": 6015952,
       "tokens/train_per_sec_per_gpu": 36.46,
@@ -2797,405 +2797,405 @@
     },
     {
       "epoch": 0.78125,
-      "grad_norm": 0.039138417690992355,
+      "grad_norm": 0.09596561640501022,
       "learning_rate": 7.450608454068415e-05,
-      "loss": 0.0003556829469744116,
+      "loss": 0.0010627463925629854,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00036,
+      "ppl": 1.00106,
       "step": 200,
       "tokens/total": 6046208,
-      "tokens/train_per_sec_per_gpu": 39.22,
+      "tokens/train_per_sec_per_gpu": 39.3,
       "tokens/trainable": 91466
     },
     {
       "epoch": 0.78515625,
-      "grad_norm": 0.0454350970685482,
+      "grad_norm": 0.03423153981566429,
       "learning_rate": 7.424407941704987e-05,
-      "loss": 0.00024555803975090384,
+      "loss": 0.00040728453313931823,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00025,
+      "ppl": 1.00041,
       "step": 201,
       "tokens/total": 6076208,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 91912
     },
     {
       "epoch": 0.7890625,
-      "grad_norm": 0.6254648566246033,
+      "grad_norm": 0.2397124022245407,
       "learning_rate": 7.398127346871986e-05,
-      "loss": 0.009876105934381485,
+      "loss": 0.0034314473159611225,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00993,
+      "ppl": 1.00344,
       "step": 202,
       "tokens/total": 6106560,
-      "tokens/train_per_sec_per_gpu": 37.79,
+      "tokens/train_per_sec_per_gpu": 37.86,
       "tokens/trainable": 92385
     },
     {
       "epoch": 0.79296875,
-      "grad_norm": 0.7776591181755066,
+      "grad_norm": 0.8799890875816345,
       "learning_rate": 7.371767763212238e-05,
-      "loss": 0.011619904078543186,
+      "loss": 0.021598415449261665,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.01169,
+      "ppl": 1.02183,
       "step": 203,
       "tokens/total": 6136640,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.55,
       "tokens/trainable": 92842
     },
     {
       "epoch": 0.796875,
-      "grad_norm": 0.07254572212696075,
+      "grad_norm": 0.12073227018117905,
       "learning_rate": 7.345330287655617e-05,
-      "loss": 0.0005007775034755468,
+      "loss": 0.0006239307695068419,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.0005,
+      "ppl": 1.00062,
       "step": 204,
       "tokens/total": 6167184,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.97,
       "tokens/trainable": 93323
     },
     {
       "epoch": 0.80078125,
-      "grad_norm": 0.6082178950309753,
+      "grad_norm": 0.3556082844734192,
       "learning_rate": 7.31881602037339e-05,
-      "loss": 0.007061033509671688,
+      "loss": 0.0026657599955797195,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00709,
+      "ppl": 1.00267,
       "step": 205,
       "tokens/total": 6197376,
-      "tokens/train_per_sec_per_gpu": 34.75,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 93756
     },
     {
       "epoch": 0.8046875,
-      "grad_norm": 0.4930607080459595,
+      "grad_norm": 0.24451178312301636,
       "learning_rate": 7.29222606473245e-05,
-      "loss": 0.006006345152854919,
+      "loss": 0.0015720583032816648,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00602,
+      "ppl": 1.00157,
       "step": 206,
       "tokens/total": 6227600,
-      "tokens/train_per_sec_per_gpu": 37.47,
+      "tokens/train_per_sec_per_gpu": 37.54,
       "tokens/trainable": 94235
     },
     {
       "epoch": 0.80859375,
-      "grad_norm": 0.06813386082649231,
+      "grad_norm": 0.23219481110572815,
       "learning_rate": 7.265561527249383e-05,
-      "loss": 0.0006858796114102006,
+      "loss": 0.0031918964814394712,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00069,
+      "ppl": 1.0032,
       "step": 207,
       "tokens/total": 6257760,
-      "tokens/train_per_sec_per_gpu": 36.48,
+      "tokens/train_per_sec_per_gpu": 36.52,
       "tokens/trainable": 94733
     },
     {
       "epoch": 0.8125,
-      "grad_norm": 0.06703829765319824,
+      "grad_norm": 0.13691213726997375,
       "learning_rate": 7.238823517544436e-05,
-      "loss": 0.001039305585436523,
+      "loss": 0.0018865278689190745,
       "memory/device_reserved (GiB)": 35.93,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00104,
+      "ppl": 1.00189,
       "step": 208,
       "tokens/total": 6288048,
-      "tokens/train_per_sec_per_gpu": 33.68,
+      "tokens/train_per_sec_per_gpu": 33.74,
       "tokens/trainable": 95218
     },
     {
       "epoch": 0.81640625,
-      "grad_norm": 0.2953227460384369,
+      "grad_norm": 0.2183372676372528,
       "learning_rate": 7.212013148295333e-05,
-      "loss": 0.0037906011566519737,
+      "loss": 0.0018021916039288044,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.0038,
+      "ppl": 1.0018,
       "step": 209,
       "tokens/total": 6318688,
-      "tokens/train_per_sec_per_gpu": 29.17,
+      "tokens/train_per_sec_per_gpu": 29.23,
       "tokens/trainable": 95619
     },
     {
       "epoch": 0.8203125,
-      "grad_norm": 0.32579171657562256,
+      "grad_norm": 0.8775327801704407,
       "learning_rate": 7.185131535190975e-05,
-      "loss": 0.002393842674791813,
+      "loss": 0.01249875407665968,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.0024,
+      "ppl": 1.01258,
       "step": 210,
       "tokens/total": 6348944,
-      "tokens/train_per_sec_per_gpu": 35.32,
+      "tokens/train_per_sec_per_gpu": 35.39,
       "tokens/trainable": 96076
     },
     {
       "epoch": 0.82421875,
-      "grad_norm": 0.4734557271003723,
+      "grad_norm": 0.4041156768798828,
       "learning_rate": 7.158179796885005e-05,
-      "loss": 0.006001632194966078,
+      "loss": 0.0069442871026694775,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00602,
+      "ppl": 1.00697,
       "step": 211,
       "tokens/total": 6378960,
-      "tokens/train_per_sec_per_gpu": 35.28,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 96505
     },
     {
       "epoch": 0.828125,
-      "grad_norm": 0.15732432901859283,
+      "grad_norm": 0.5414943099021912,
       "learning_rate": 7.131159054949273e-05,
-      "loss": 0.003803959349170327,
+      "loss": 0.0017384262755513191,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00381,
+      "ppl": 1.00174,
       "step": 212,
       "tokens/total": 6409472,
-      "tokens/train_per_sec_per_gpu": 32.57,
+      "tokens/train_per_sec_per_gpu": 32.68,
       "tokens/trainable": 96930
     },
     {
       "epoch": 0.83203125,
-      "grad_norm": 0.18325994908809662,
+      "grad_norm": 0.010344896465539932,
       "learning_rate": 7.104070433827139e-05,
-      "loss": 0.005038415547460318,
+      "loss": 0.0002013841731240973,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00505,
+      "ppl": 1.0002,
       "step": 213,
       "tokens/total": 6439904,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.72,
       "tokens/trainable": 97361
     },
     {
       "epoch": 0.8359375,
-      "grad_norm": 0.17816093564033508,
+      "grad_norm": 0.5833696722984314,
       "learning_rate": 7.076915060786705e-05,
-      "loss": 0.0020862463861703873,
+      "loss": 0.009723629802465439,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00209,
+      "ppl": 1.00977,
       "step": 214,
       "tokens/total": 6470176,
-      "tokens/train_per_sec_per_gpu": 36.26,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 97858
     },
     {
       "epoch": 0.83984375,
-      "grad_norm": 0.2746942937374115,
+      "grad_norm": 0.31467050313949585,
       "learning_rate": 7.049694065873882e-05,
-      "loss": 0.005029214546084404,
+      "loss": 0.004062815103679895,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00504,
+      "ppl": 1.00407,
       "step": 215,
       "tokens/total": 6500336,
-      "tokens/train_per_sec_per_gpu": 37.41,
+      "tokens/train_per_sec_per_gpu": 37.47,
       "tokens/trainable": 98308
     },
     {
       "epoch": 0.84375,
-      "grad_norm": 0.4508717358112335,
+      "grad_norm": 0.013619605451822281,
       "learning_rate": 7.022408581865382e-05,
-      "loss": 0.0054893530905246735,
+      "loss": 0.0002783117815852165,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0055,
+      "ppl": 1.00028,
       "step": 216,
       "tokens/total": 6530688,
-      "tokens/train_per_sec_per_gpu": 36.34,
+      "tokens/train_per_sec_per_gpu": 36.41,
       "tokens/trainable": 98781
     },
     {
       "epoch": 0.84765625,
-      "grad_norm": 0.23194828629493713,
+      "grad_norm": 0.02845357358455658,
       "learning_rate": 6.99505974422157e-05,
-      "loss": 0.0034187212586402893,
+      "loss": 0.00029517346411012113,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00342,
+      "ppl": 1.0003,
       "step": 217,
       "tokens/total": 6561040,
-      "tokens/train_per_sec_per_gpu": 36.67,
+      "tokens/train_per_sec_per_gpu": 36.76,
       "tokens/trainable": 99266
     },
     {
       "epoch": 0.8515625,
-      "grad_norm": 0.06469441950321198,
+      "grad_norm": 0.17569833993911743,
       "learning_rate": 6.967648691039213e-05,
-      "loss": 0.0009051199886016548,
+      "loss": 0.0029272970277816057,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00091,
+      "ppl": 1.00293,
       "step": 218,
       "tokens/total": 6589392,
-      "tokens/train_per_sec_per_gpu": 39.6,
+      "tokens/train_per_sec_per_gpu": 39.65,
       "tokens/trainable": 99731
     },
     {
       "epoch": 0.85546875,
-      "grad_norm": 0.06683141738176346,
+      "grad_norm": 0.25007033348083496,
       "learning_rate": 6.940176563004123e-05,
-      "loss": 0.0011187575291842222,
+      "loss": 0.005346981342881918,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00112,
+      "ppl": 1.00536,
       "step": 219,
       "tokens/total": 6619824,
-      "tokens/train_per_sec_per_gpu": 31.66,
+      "tokens/train_per_sec_per_gpu": 31.71,
       "tokens/trainable": 100164
     },
     {
       "epoch": 0.859375,
-      "grad_norm": 0.1525852382183075,
+      "grad_norm": 0.36598774790763855,
       "learning_rate": 6.912644503343682e-05,
-      "loss": 0.0017743443604558706,
+      "loss": 0.00675960024818778,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00178,
+      "ppl": 1.00678,
       "step": 220,
       "tokens/total": 6650224,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.18,
       "tokens/trainable": 100620
     },
     {
       "epoch": 0.86328125,
-      "grad_norm": 0.8795211911201477,
+      "grad_norm": 0.2870761454105377,
       "learning_rate": 6.885053657779273e-05,
-      "loss": 0.006786257494240999,
+      "loss": 0.004020639695227146,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00681,
+      "ppl": 1.00403,
       "step": 221,
       "tokens/total": 6680384,
-      "tokens/train_per_sec_per_gpu": 32.9,
+      "tokens/train_per_sec_per_gpu": 32.94,
       "tokens/trainable": 101080
     },
     {
       "epoch": 0.8671875,
-      "grad_norm": 0.19531063735485077,
+      "grad_norm": 0.05318336561322212,
       "learning_rate": 6.857405174478604e-05,
-      "loss": 0.0014687087386846542,
+      "loss": 0.0006697649951092899,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00147,
+      "ppl": 1.00067,
       "step": 222,
       "tokens/total": 6710896,
-      "tokens/train_per_sec_per_gpu": 37.76,
+      "tokens/train_per_sec_per_gpu": 37.8,
       "tokens/trainable": 101546
     },
     {
       "epoch": 0.87109375,
-      "grad_norm": 0.011134379543364048,
+      "grad_norm": 0.016319693997502327,
       "learning_rate": 6.82970020400792e-05,
-      "loss": 0.00012406683526933193,
+      "loss": 0.00020417713676579297,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00012,
+      "ppl": 1.0002,
       "step": 223,
       "tokens/total": 6741280,
-      "tokens/train_per_sec_per_gpu": 31.51,
+      "tokens/train_per_sec_per_gpu": 31.52,
       "tokens/trainable": 101987
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.41609230637550354,
+      "grad_norm": 0.09935661405324936,
       "learning_rate": 6.801939899284132e-05,
-      "loss": 0.00440242001786828,
+      "loss": 0.0018507638014853,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00441,
+      "ppl": 1.00185,
       "step": 224,
       "tokens/total": 6771488,
-      "tokens/train_per_sec_per_gpu": 35.45,
+      "tokens/train_per_sec_per_gpu": 35.37,
       "tokens/trainable": 102435
     },
     {
       "epoch": 0.87890625,
-      "grad_norm": 0.23292526602745056,
+      "grad_norm": 0.045450225472450256,
       "learning_rate": 6.774125415526827e-05,
-      "loss": 0.000571364420466125,
+      "loss": 0.0007717214175499976,
       "memory/device_reserved (GiB)": 35.94,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00057,
+      "ppl": 1.00077,
       "step": 225,
       "tokens/total": 6801744,
-      "tokens/train_per_sec_per_gpu": 28.91,
+      "tokens/train_per_sec_per_gpu": 28.86,
       "tokens/trainable": 102843
     },
     {
       "epoch": 0.8828125,
-      "grad_norm": 0.016833841800689697,
+      "grad_norm": 0.1693895012140274,
       "learning_rate": 6.746257910210214e-05,
-      "loss": 0.00012277837959118187,
+      "loss": 0.0015482499729841948,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00155,
       "step": 226,
       "tokens/total": 6832432,
-      "tokens/train_per_sec_per_gpu": 37.37,
+      "tokens/train_per_sec_per_gpu": 37.39,
       "tokens/trainable": 103342
     },
     {
       "epoch": 0.88671875,
-      "grad_norm": 0.35366129875183105,
+      "grad_norm": 0.011035765521228313,
       "learning_rate": 6.718338543014937e-05,
-      "loss": 0.004477445501834154,
+      "loss": 0.0002161034499295056,
       "memory/device_reserved (GiB)": 35.71,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00449,
+      "ppl": 1.00022,
       "step": 227,
       "tokens/total": 6862448,
-      "tokens/train_per_sec_per_gpu": 40.68,
+      "tokens/train_per_sec_per_gpu": 40.81,
       "tokens/trainable": 103844
     },
     {
       "epoch": 0.890625,
-      "grad_norm": 0.4743596017360687,
+      "grad_norm": 0.339762806892395,
       "learning_rate": 6.69036847577983e-05,
-      "loss": 0.002162824384868145,
+      "loss": 0.00712943309918046,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00217,
+      "ppl": 1.00715,
       "step": 228,
       "tokens/total": 6892832,
       "tokens/train_per_sec_per_gpu": 36.23,
@@ -3203,531 +3203,531 @@
     },
     {
       "epoch": 0.89453125,
-      "grad_norm": 0.4887899160385132,
+      "grad_norm": 0.022490613162517548,
       "learning_rate": 6.662348872453553e-05,
-      "loss": 0.0048948717303574085,
+      "loss": 0.000386083556804806,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.69,
       "memory/max_allocated (GiB)": 33.69,
-      "ppl": 1.00491,
+      "ppl": 1.00039,
       "step": 229,
       "tokens/total": 6922832,
-      "tokens/train_per_sec_per_gpu": 33.43,
+      "tokens/train_per_sec_per_gpu": 33.49,
       "tokens/trainable": 104772
     },
     {
       "epoch": 0.8984375,
-      "grad_norm": 0.017348933964967728,
+      "grad_norm": 0.030813584104180336,
       "learning_rate": 6.63428089904618e-05,
-      "loss": 7.635008660145104e-05,
+      "loss": 0.00022011388500686735,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00008,
+      "ppl": 1.00022,
       "step": 230,
       "tokens/total": 6953104,
-      "tokens/train_per_sec_per_gpu": 36.83,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 105240
     },
     {
       "epoch": 0.90234375,
-      "grad_norm": 0.3773981034755707,
+      "grad_norm": 0.13754640519618988,
       "learning_rate": 6.60616572358065e-05,
-      "loss": 0.003427914110943675,
+      "loss": 0.0012558111920952797,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00343,
+      "ppl": 1.00126,
       "step": 231,
       "tokens/total": 6983440,
-      "tokens/train_per_sec_per_gpu": 33.75,
+      "tokens/train_per_sec_per_gpu": 33.76,
       "tokens/trainable": 105703
     },
     {
       "epoch": 0.90625,
-      "grad_norm": 0.0006717262440361083,
+      "grad_norm": 0.003156182123348117,
       "learning_rate": 6.578004516044172e-05,
-      "loss": 1.8351973267272115e-05,
+      "loss": 6.873848178656772e-05,
       "memory/device_reserved (GiB)": 35.77,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 232,
       "tokens/total": 7013840,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.02,
       "tokens/trainable": 106179
     },
     {
       "epoch": 0.91015625,
-      "grad_norm": 0.0018952380632981658,
+      "grad_norm": 0.021215464919805527,
       "learning_rate": 6.549798448339548e-05,
-      "loss": 2.8159589419374242e-05,
+      "loss": 0.00016204064013436437,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00016,
       "step": 233,
       "tokens/total": 7044240,
-      "tokens/train_per_sec_per_gpu": 37.96,
+      "tokens/train_per_sec_per_gpu": 38.0,
       "tokens/trainable": 106683
     },
     {
       "epoch": 0.9140625,
-      "grad_norm": 0.04159875586628914,
+      "grad_norm": 0.03298190236091614,
       "learning_rate": 6.521548694236384e-05,
-      "loss": 0.0001036152389133349,
+      "loss": 0.0003434696700423956,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.0001,
+      "ppl": 1.00034,
       "step": 234,
       "tokens/total": 7074224,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 107121
     },
     {
       "epoch": 0.91796875,
-      "grad_norm": 0.005844940431416035,
+      "grad_norm": 0.08474111557006836,
       "learning_rate": 6.493256429322259e-05,
-      "loss": 5.149145727045834e-05,
+      "loss": 0.0006107806693762541,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00061,
       "step": 235,
       "tokens/total": 7104672,
-      "tokens/train_per_sec_per_gpu": 32.68,
+      "tokens/train_per_sec_per_gpu": 32.7,
       "tokens/trainable": 107561
     },
     {
       "epoch": 0.921875,
-      "grad_norm": 0.0022666200529783964,
+      "grad_norm": 0.012000355869531631,
       "learning_rate": 6.464922830953799e-05,
-      "loss": 2.886037327698432e-05,
+      "loss": 0.00010977509373333305,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00011,
       "step": 236,
       "tokens/total": 7134784,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 108018
     },
     {
       "epoch": 0.92578125,
-      "grad_norm": 0.27136701345443726,
+      "grad_norm": 0.011812024749815464,
       "learning_rate": 6.436549078207688e-05,
-      "loss": 0.0009024653700180352,
+      "loss": 6.36270415270701e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.0009,
+      "ppl": 1.00006,
       "step": 237,
       "tokens/total": 7164960,
-      "tokens/train_per_sec_per_gpu": 33.28,
+      "tokens/train_per_sec_per_gpu": 33.32,
       "tokens/trainable": 108446
     },
     {
       "epoch": 0.9296875,
-      "grad_norm": 0.13347801566123962,
+      "grad_norm": 0.055643483996391296,
       "learning_rate": 6.408136351831592e-05,
-      "loss": 0.00046705722343176603,
+      "loss": 0.0002590732474345714,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00047,
+      "ppl": 1.00026,
       "step": 238,
       "tokens/total": 7195344,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.15,
       "tokens/trainable": 108866
     },
     {
       "epoch": 0.93359375,
-      "grad_norm": 0.10074340552091599,
+      "grad_norm": 0.7225600481033325,
       "learning_rate": 6.379685834195036e-05,
-      "loss": 0.0006794088403694332,
+      "loss": 0.004054174292832613,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00068,
+      "ppl": 1.00406,
       "step": 239,
       "tokens/total": 7225680,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 109316
     },
     {
       "epoch": 0.9375,
-      "grad_norm": 0.12028707563877106,
+      "grad_norm": 0.06819978356361389,
       "learning_rate": 6.351198709240186e-05,
-      "loss": 0.000513301754835993,
+      "loss": 0.00046806156751699746,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00051,
+      "ppl": 1.00047,
       "step": 240,
       "tokens/total": 7256112,
-      "tokens/train_per_sec_per_gpu": 33.26,
+      "tokens/train_per_sec_per_gpu": 33.27,
       "tokens/trainable": 109787
     },
     {
       "epoch": 0.94140625,
-      "grad_norm": 0.0062753986567258835,
+      "grad_norm": 0.17400965094566345,
       "learning_rate": 6.32267616243259e-05,
-      "loss": 3.0990675440989435e-05,
+      "loss": 0.0013036895543336868,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00003,
+      "ppl": 1.0013,
       "step": 241,
       "tokens/total": 7286384,
-      "tokens/train_per_sec_per_gpu": 38.3,
+      "tokens/train_per_sec_per_gpu": 38.35,
       "tokens/trainable": 110246
     },
     {
       "epoch": 0.9453125,
-      "grad_norm": 0.05608534440398216,
+      "grad_norm": 0.0573316365480423,
       "learning_rate": 6.294119380711849e-05,
-      "loss": 0.00039500967250205576,
+      "loss": 0.00048713054275140166,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0004,
+      "ppl": 1.00049,
       "step": 242,
       "tokens/total": 7316688,
-      "tokens/train_per_sec_per_gpu": 29.14,
+      "tokens/train_per_sec_per_gpu": 29.22,
       "tokens/trainable": 110667
     },
     {
       "epoch": 0.94921875,
-      "grad_norm": 0.05409131199121475,
+      "grad_norm": 0.8039842247962952,
       "learning_rate": 6.265529552442209e-05,
-      "loss": 0.0002061743725789711,
+      "loss": 0.013966547325253487,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00021,
+      "ppl": 1.01406,
       "step": 243,
       "tokens/total": 7346992,
-      "tokens/train_per_sec_per_gpu": 38.09,
+      "tokens/train_per_sec_per_gpu": 38.13,
       "tokens/trainable": 111154
     },
     {
       "epoch": 0.953125,
-      "grad_norm": 0.0556725338101387,
+      "grad_norm": 0.03269768878817558,
       "learning_rate": 6.236907867363127e-05,
-      "loss": 0.0005747157847508788,
+      "loss": 0.00011441730748629197,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00057,
+      "ppl": 1.00011,
       "step": 244,
       "tokens/total": 7377632,
-      "tokens/train_per_sec_per_gpu": 32.97,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 111611
     },
     {
       "epoch": 0.95703125,
-      "grad_norm": 0.16125747561454773,
+      "grad_norm": 0.006486211437731981,
       "learning_rate": 6.208255516539749e-05,
-      "loss": 0.0008788461564108729,
+      "loss": 4.603556953952648e-05,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00088,
+      "ppl": 1.00005,
       "step": 245,
       "tokens/total": 7407872,
-      "tokens/train_per_sec_per_gpu": 34.13,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 112102
     },
     {
       "epoch": 0.9609375,
-      "grad_norm": 0.00565892830491066,
+      "grad_norm": 0.04383692517876625,
       "learning_rate": 6.179573692313344e-05,
-      "loss": 4.069084388902411e-05,
+      "loss": 0.0002824773546308279,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00004,
+      "ppl": 1.00028,
       "step": 246,
       "tokens/total": 7438448,
-      "tokens/train_per_sec_per_gpu": 33.16,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 112569
     },
     {
       "epoch": 0.96484375,
-      "grad_norm": 0.1774720549583435,
+      "grad_norm": 0.16995811462402344,
       "learning_rate": 6.150863588251694e-05,
-      "loss": 0.0005083430442027748,
+      "loss": 0.0024534151889383793,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00051,
+      "ppl": 1.00246,
       "step": 247,
       "tokens/total": 7466768,
-      "tokens/train_per_sec_per_gpu": 37.68,
+      "tokens/train_per_sec_per_gpu": 37.71,
       "tokens/trainable": 113023
     },
     {
       "epoch": 0.96875,
-      "grad_norm": 0.0028223118279129267,
+      "grad_norm": 0.016517437994480133,
       "learning_rate": 6.122126399099419e-05,
-      "loss": 3.470741648925468e-05,
+      "loss": 0.00018760509556159377,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00003,
+      "ppl": 1.00019,
       "step": 248,
       "tokens/total": 7497360,
-      "tokens/train_per_sec_per_gpu": 35.17,
+      "tokens/train_per_sec_per_gpu": 35.12,
       "tokens/trainable": 113464
     },
     {
       "epoch": 0.97265625,
-      "grad_norm": 0.0367576889693737,
+      "grad_norm": 0.27915316820144653,
       "learning_rate": 6.0933633207282615e-05,
-      "loss": 8.433192851953208e-05,
+      "loss": 0.0029680300503969193,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00008,
+      "ppl": 1.00297,
       "step": 249,
       "tokens/total": 7527504,
-      "tokens/train_per_sec_per_gpu": 32.65,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 113908
     },
     {
       "epoch": 0.9765625,
-      "grad_norm": 0.11235758662223816,
+      "grad_norm": 0.45557740330696106,
       "learning_rate": 6.064575550087316e-05,
-      "loss": 0.0006511391839012504,
+      "loss": 0.005397973116487265,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00065,
+      "ppl": 1.00541,
       "step": 250,
       "tokens/total": 7555808,
-      "tokens/train_per_sec_per_gpu": 35.27,
+      "tokens/train_per_sec_per_gpu": 35.3,
       "tokens/trainable": 114348
     },
     {
       "epoch": 0.98046875,
-      "grad_norm": 0.049232710152864456,
+      "grad_norm": 0.03177047520875931,
       "learning_rate": 6.0357642851532245e-05,
-      "loss": 0.00024149491218850017,
+      "loss": 0.00033563212491571903,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00024,
+      "ppl": 1.00034,
       "step": 251,
       "tokens/total": 7586320,
-      "tokens/train_per_sec_per_gpu": 35.15,
+      "tokens/train_per_sec_per_gpu": 35.23,
       "tokens/trainable": 114830
     },
     {
       "epoch": 0.984375,
-      "grad_norm": 0.4833732545375824,
+      "grad_norm": 0.2969627380371094,
       "learning_rate": 6.0069307248803294e-05,
-      "loss": 0.005053690634667873,
+      "loss": 0.0029213305097073317,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00507,
+      "ppl": 1.00293,
       "step": 252,
       "tokens/total": 7616576,
-      "tokens/train_per_sec_per_gpu": 35.1,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 115278
     },
     {
       "epoch": 0.98828125,
-      "grad_norm": 0.013355270028114319,
+      "grad_norm": 0.09316814690828323,
       "learning_rate": 5.9780760691507635e-05,
-      "loss": 8.867261931300163e-05,
+      "loss": 0.00040648062713444233,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00009,
+      "ppl": 1.00041,
       "step": 253,
       "tokens/total": 7647152,
-      "tokens/train_per_sec_per_gpu": 32.21,
+      "tokens/train_per_sec_per_gpu": 32.29,
       "tokens/trainable": 115699
     },
     {
       "epoch": 0.9921875,
-      "grad_norm": 0.09679444134235382,
+      "grad_norm": 0.2160760760307312,
       "learning_rate": 5.9492015187245334e-05,
-      "loss": 0.0006353833014145494,
+      "loss": 0.0019942133221775293,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00064,
+      "ppl": 1.002,
       "step": 254,
       "tokens/total": 7677584,
-      "tokens/train_per_sec_per_gpu": 30.73,
+      "tokens/train_per_sec_per_gpu": 30.81,
       "tokens/trainable": 116133
     },
     {
       "epoch": 0.99609375,
-      "grad_norm": 0.5139889717102051,
+      "grad_norm": 0.08144493401050568,
       "learning_rate": 5.920308275189541e-05,
-      "loss": 0.001129151787608862,
+      "loss": 0.0008475447539240122,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00113,
+      "ppl": 1.00085,
       "step": 255,
       "tokens/total": 7708016,
-      "tokens/train_per_sec_per_gpu": 36.73,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 116591
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.00027604683418758214,
+      "grad_norm": 0.06008912995457649,
       "learning_rate": 5.8913975409115874e-05,
-      "loss": 7.1114727688836865e-06,
+      "loss": 0.0007099541835486889,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00071,
       "step": 256,
       "tokens/total": 7738608,
-      "tokens/train_per_sec_per_gpu": 32.04,
+      "tokens/train_per_sec_per_gpu": 32.11,
       "tokens/trainable": 117039
     },
     {
       "epoch": 1.00390625,
-      "grad_norm": 0.005449575372040272,
+      "grad_norm": 0.01939917728304863,
       "learning_rate": 5.8624705189843395e-05,
-      "loss": 2.958515688078478e-05,
+      "loss": 0.00010193810157943517,
       "memory/device_reserved (GiB)": 36.14,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.0001,
       "step": 257,
       "tokens/total": 7768944,
-      "tokens/train_per_sec_per_gpu": 32.24,
+      "tokens/train_per_sec_per_gpu": 32.49,
       "tokens/trainable": 117495
     },
     {
       "epoch": 1.0078125,
-      "grad_norm": 0.003980543464422226,
+      "grad_norm": 0.020724618807435036,
       "learning_rate": 5.833528413179249e-05,
-      "loss": 2.851396129699424e-05,
+      "loss": 0.00015316747885663062,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00003,
+      "ppl": 1.00015,
       "step": 258,
       "tokens/total": 7799232,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.54,
       "tokens/trainable": 117925
     },
     {
       "epoch": 1.01171875,
-      "grad_norm": 0.001549297128804028,
+      "grad_norm": 0.002170708030462265,
       "learning_rate": 5.80457242789548e-05,
-      "loss": 1.1607408850977663e-05,
+      "loss": 4.118717333767563e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 259,
       "tokens/total": 7829568,
-      "tokens/train_per_sec_per_gpu": 31.58,
+      "tokens/train_per_sec_per_gpu": 31.66,
       "tokens/trainable": 118373
     },
     {
       "epoch": 1.015625,
-      "grad_norm": 0.00039364350959658623,
+      "grad_norm": 0.0011113203363493085,
       "learning_rate": 5.77560376810977e-05,
-      "loss": 9.399981536262203e-06,
+      "loss": 2.2629777959082276e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 260,
       "tokens/total": 7860000,
-      "tokens/train_per_sec_per_gpu": 31.46,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 118816
     },
     {
       "epoch": 1.01953125,
-      "grad_norm": 0.024472124874591827,
+      "grad_norm": 0.01329761277884245,
       "learning_rate": 5.7466236393263005e-05,
-      "loss": 0.00013395686983130872,
+      "loss": 7.257944525917992e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 261,
       "tokens/total": 7890464,
-      "tokens/train_per_sec_per_gpu": 33.8,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 119263
     },
     {
       "epoch": 1.0234375,
-      "grad_norm": 0.3619483709335327,
+      "grad_norm": 0.060381509363651276,
       "learning_rate": 5.717633247526522e-05,
-      "loss": 0.0023931623436510563,
+      "loss": 5.40789587830659e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0024,
+      "ppl": 1.00005,
       "step": 262,
       "tokens/total": 7920944,
-      "tokens/train_per_sec_per_gpu": 34.66,
+      "tokens/train_per_sec_per_gpu": 34.79,
       "tokens/trainable": 119772
     },
     {
       "epoch": 1.02734375,
-      "grad_norm": 0.17659051716327667,
+      "grad_norm": 0.003907898440957069,
       "learning_rate": 5.688633799118971e-05,
-      "loss": 0.0005250183749012649,
+      "loss": 5.063481512479484e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 263,
       "tokens/total": 7951296,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.87,
       "tokens/trainable": 120269
     },
     {
       "epoch": 1.03125,
-      "grad_norm": 0.000787933764513582,
+      "grad_norm": 0.008468744345009327,
       "learning_rate": 5.659626500889066e-05,
-      "loss": 1.1148193152621388e-05,
+      "loss": 3.1418872822541744e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 264,
       "tokens/total": 7981856,
-      "tokens/train_per_sec_per_gpu": 36.96,
+      "tokens/train_per_sec_per_gpu": 37.05,
       "tokens/trainable": 120746
     },
     {
       "epoch": 1.03515625,
-      "grad_norm": 0.00879898015409708,
+      "grad_norm": 0.276324599981308,
       "learning_rate": 5.6306125599488905e-05,
-      "loss": 4.70993691124022e-05,
+      "loss": 0.0011993877124041319,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00005,
+      "ppl": 1.0012,
       "step": 265,
       "tokens/total": 8012192,
-      "tokens/train_per_sec_per_gpu": 38.22,
+      "tokens/train_per_sec_per_gpu": 38.3,
       "tokens/trainable": 121204
     },
     {
       "epoch": 1.0390625,
-      "grad_norm": 0.0005215261480771005,
+      "grad_norm": 0.3999474346637726,
       "learning_rate": 5.601593183686955e-05,
-      "loss": 8.791243999439757e-06,
+      "loss": 0.01104999240487814,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00001,
+      "ppl": 1.01111,
       "step": 266,
       "tokens/total": 8042736,
       "tokens/train_per_sec_per_gpu": 31.23,
@@ -3735,1525 +3735,1525 @@
     },
     {
       "epoch": 1.04296875,
-      "grad_norm": 0.1482085883617401,
+      "grad_norm": 0.16733741760253906,
       "learning_rate": 5.572569579717961e-05,
-      "loss": 0.000805735879112035,
+      "loss": 0.0020670127123594284,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00081,
+      "ppl": 1.00207,
       "step": 267,
       "tokens/total": 8073024,
-      "tokens/train_per_sec_per_gpu": 40.04,
+      "tokens/train_per_sec_per_gpu": 40.11,
       "tokens/trainable": 122183
     },
     {
       "epoch": 1.046875,
-      "grad_norm": 0.08614944666624069,
+      "grad_norm": 0.03506140410900116,
       "learning_rate": 5.543542955832538e-05,
-      "loss": 0.00032304273918271065,
+      "loss": 0.0002631399256642908,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00032,
+      "ppl": 1.00026,
       "step": 268,
       "tokens/total": 8103280,
-      "tokens/train_per_sec_per_gpu": 35.38,
+      "tokens/train_per_sec_per_gpu": 35.41,
       "tokens/trainable": 122645
     },
     {
       "epoch": 1.05078125,
-      "grad_norm": 0.0005171536467969418,
+      "grad_norm": 0.006357306614518166,
       "learning_rate": 5.514514519946986e-05,
-      "loss": 1.1340615856170189e-05,
+      "loss": 7.129507866920903e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 269,
       "tokens/total": 8133440,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.09,
       "tokens/trainable": 123105
     },
     {
       "epoch": 1.0546875,
-      "grad_norm": 0.00104080094024539,
+      "grad_norm": 0.06588926911354065,
       "learning_rate": 5.485485480053015e-05,
-      "loss": 1.1951466149184853e-05,
+      "loss": 0.0005020307726226747,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00001,
+      "ppl": 1.0005,
       "step": 270,
       "tokens/total": 8163920,
-      "tokens/train_per_sec_per_gpu": 36.18,
+      "tokens/train_per_sec_per_gpu": 36.24,
       "tokens/trainable": 123595
     },
     {
       "epoch": 1.05859375,
-      "grad_norm": 0.005696228239685297,
+      "grad_norm": 0.056961171329021454,
       "learning_rate": 5.4564570441674645e-05,
-      "loss": 1.725989204715006e-05,
+      "loss": 0.0007095756009221077,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 271,
       "tokens/total": 8194032,
-      "tokens/train_per_sec_per_gpu": 37.59,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 124077
     },
     {
       "epoch": 1.0625,
-      "grad_norm": 0.0005264461506158113,
+      "grad_norm": 0.09018661081790924,
       "learning_rate": 5.42743042028204e-05,
-      "loss": 1.0610991921566892e-05,
+      "loss": 0.0002944314619526267,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00029,
       "step": 272,
       "tokens/total": 8224320,
-      "tokens/train_per_sec_per_gpu": 34.65,
+      "tokens/train_per_sec_per_gpu": 34.77,
       "tokens/trainable": 124547
     },
     {
       "epoch": 1.06640625,
-      "grad_norm": 0.08169055730104446,
+      "grad_norm": 0.13114672899246216,
       "learning_rate": 5.3984068163130464e-05,
-      "loss": 0.0003627589321695268,
+      "loss": 0.0013651850167661905,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00036,
+      "ppl": 1.00137,
       "step": 273,
       "tokens/total": 8254368,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 34.0,
       "tokens/trainable": 124977
     },
     {
       "epoch": 1.0703125,
-      "grad_norm": 0.0005202546599321067,
+      "grad_norm": 0.35149261355400085,
       "learning_rate": 5.369387440051111e-05,
-      "loss": 1.123354013543576e-05,
+      "loss": 0.003347770543769002,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00335,
       "step": 274,
       "tokens/total": 8284944,
-      "tokens/train_per_sec_per_gpu": 34.51,
+      "tokens/train_per_sec_per_gpu": 34.56,
       "tokens/trainable": 125438
     },
     {
       "epoch": 1.07421875,
-      "grad_norm": 0.0031104006338864565,
+      "grad_norm": 0.33942294120788574,
       "learning_rate": 5.340373499110935e-05,
-      "loss": 1.4764596926397644e-05,
+      "loss": 0.0012429209891706705,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00124,
       "step": 275,
       "tokens/total": 8315584,
-      "tokens/train_per_sec_per_gpu": 31.47,
+      "tokens/train_per_sec_per_gpu": 31.49,
       "tokens/trainable": 125889
     },
     {
       "epoch": 1.078125,
-      "grad_norm": 0.01905728504061699,
+      "grad_norm": 0.39785870909690857,
       "learning_rate": 5.3113662008810304e-05,
-      "loss": 9.163993672700599e-05,
+      "loss": 0.004628765396773815,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00464,
       "step": 276,
       "tokens/total": 8345952,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 126363
     },
     {
       "epoch": 1.08203125,
-      "grad_norm": 0.0020019100047647953,
+      "grad_norm": 0.2942987382411957,
       "learning_rate": 5.282366752473479e-05,
-      "loss": 1.796903961803764e-05,
+      "loss": 0.004942136351019144,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00495,
       "step": 277,
       "tokens/total": 8376416,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 126828
     },
     {
       "epoch": 1.0859375,
-      "grad_norm": 0.0011239969171583652,
+      "grad_norm": 0.04225993901491165,
       "learning_rate": 5.2533763606737005e-05,
-      "loss": 1.2794653230230324e-05,
+      "loss": 0.00015706325939390808,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 278,
       "tokens/total": 8406640,
-      "tokens/train_per_sec_per_gpu": 38.16,
+      "tokens/train_per_sec_per_gpu": 38.21,
       "tokens/trainable": 127331
     },
     {
       "epoch": 1.08984375,
-      "grad_norm": 0.07841381430625916,
+      "grad_norm": 0.008294228464365005,
       "learning_rate": 5.224396231890232e-05,
-      "loss": 0.0005632906104438007,
+      "loss": 9.9621967819985e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00056,
+      "ppl": 1.0001,
       "step": 279,
       "tokens/total": 8436784,
-      "tokens/train_per_sec_per_gpu": 34.34,
+      "tokens/train_per_sec_per_gpu": 34.39,
       "tokens/trainable": 127798
     },
     {
       "epoch": 1.09375,
-      "grad_norm": 0.0066568912006914616,
+      "grad_norm": 0.017988065257668495,
       "learning_rate": 5.195427572104522e-05,
-      "loss": 4.4691078073810786e-05,
+      "loss": 0.00021193400607444346,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00004,
+      "ppl": 1.00021,
       "step": 280,
       "tokens/total": 8467152,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.84,
       "tokens/trainable": 128247
     },
     {
       "epoch": 1.09765625,
-      "grad_norm": 0.0006084765191189945,
+      "grad_norm": 0.055866386741399765,
       "learning_rate": 5.166471586820751e-05,
-      "loss": 9.42349561228184e-06,
+      "loss": 0.0003000127908308059,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.0003,
       "step": 281,
       "tokens/total": 8497424,
-      "tokens/train_per_sec_per_gpu": 36.17,
+      "tokens/train_per_sec_per_gpu": 36.22,
       "tokens/trainable": 128729
     },
     {
       "epoch": 1.1015625,
-      "grad_norm": 0.4024369716644287,
+      "grad_norm": 0.03349864110350609,
       "learning_rate": 5.1375294810156615e-05,
-      "loss": 0.002225687028840184,
+      "loss": 0.0003475213306955993,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00223,
+      "ppl": 1.00035,
       "step": 282,
       "tokens/total": 8527616,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 129203
     },
     {
       "epoch": 1.10546875,
-      "grad_norm": 0.0008598315180279315,
+      "grad_norm": 0.0013503885129466653,
       "learning_rate": 5.1086024590884144e-05,
-      "loss": 9.10060407477431e-06,
+      "loss": 2.522556678741239e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 283,
       "tokens/total": 8557984,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 129653
     },
     {
       "epoch": 1.109375,
-      "grad_norm": 0.13851912319660187,
+      "grad_norm": 0.03435222804546356,
       "learning_rate": 5.079691724810461e-05,
-      "loss": 0.00045336864423006773,
+      "loss": 0.00024076660338323563,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00045,
+      "ppl": 1.00024,
       "step": 284,
       "tokens/total": 8588384,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 130100
     },
     {
       "epoch": 1.11328125,
-      "grad_norm": 0.08453445136547089,
+      "grad_norm": 0.06312273442745209,
       "learning_rate": 5.0507984812754684e-05,
-      "loss": 0.0002941025304608047,
+      "loss": 0.0001932874001795426,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00029,
+      "ppl": 1.00019,
       "step": 285,
       "tokens/total": 8618768,
-      "tokens/train_per_sec_per_gpu": 30.44,
+      "tokens/train_per_sec_per_gpu": 30.49,
       "tokens/trainable": 130543
     },
     {
       "epoch": 1.1171875,
-      "grad_norm": 0.018670005723834038,
+      "grad_norm": 0.01356665138155222,
       "learning_rate": 5.021923930849237e-05,
-      "loss": 0.00010185746214119717,
+      "loss": 0.00014943527639843524,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.0001,
+      "ppl": 1.00015,
       "step": 286,
       "tokens/total": 8648976,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 131000
     },
     {
       "epoch": 1.12109375,
-      "grad_norm": 0.0015903875464573503,
+      "grad_norm": 0.008714325726032257,
       "learning_rate": 4.99306927511967e-05,
-      "loss": 1.1978466318396386e-05,
+      "loss": 9.58888849709183e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00001,
+      "ppl": 1.0001,
       "step": 287,
       "tokens/total": 8679152,
-      "tokens/train_per_sec_per_gpu": 33.96,
+      "tokens/train_per_sec_per_gpu": 34.04,
       "tokens/trainable": 131449
     },
     {
       "epoch": 1.125,
-      "grad_norm": 0.0004093375173397362,
+      "grad_norm": 0.004380185157060623,
       "learning_rate": 4.964235714846775e-05,
-      "loss": 7.070232186379144e-06,
+      "loss": 3.911522071575746e-05,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 288,
       "tokens/total": 8709504,
-      "tokens/train_per_sec_per_gpu": 35.94,
+      "tokens/train_per_sec_per_gpu": 35.97,
       "tokens/trainable": 131923
     },
     {
       "epoch": 1.12890625,
-      "grad_norm": 0.12260667979717255,
+      "grad_norm": 0.3479631245136261,
       "learning_rate": 4.9354244499126866e-05,
-      "loss": 0.0006339149549603462,
+      "loss": 0.003404036397114396,
       "memory/device_reserved (GiB)": 35.91,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00063,
+      "ppl": 1.00341,
       "step": 289,
       "tokens/total": 8739680,
-      "tokens/train_per_sec_per_gpu": 33.11,
+      "tokens/train_per_sec_per_gpu": 33.15,
       "tokens/trainable": 132397
     },
     {
       "epoch": 1.1328125,
-      "grad_norm": 0.00021185794321354479,
+      "grad_norm": 0.0024364532437175512,
       "learning_rate": 4.90663667927174e-05,
-      "loss": 5.396935193857644e-06,
+      "loss": 2.929499896708876e-05,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 290,
       "tokens/total": 8770112,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 132846
     },
     {
       "epoch": 1.13671875,
-      "grad_norm": 0.2777019739151001,
+      "grad_norm": 0.5378495454788208,
       "learning_rate": 4.877873600900581e-05,
-      "loss": 0.0013501858338713646,
+      "loss": 0.0011512563796713948,
       "memory/device_reserved (GiB)": 35.57,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00135,
+      "ppl": 1.00115,
       "step": 291,
       "tokens/total": 8800368,
-      "tokens/train_per_sec_per_gpu": 35.65,
+      "tokens/train_per_sec_per_gpu": 35.74,
       "tokens/trainable": 133347
     },
     {
       "epoch": 1.140625,
-      "grad_norm": 0.0012349279131740332,
+      "grad_norm": 0.005875240080058575,
       "learning_rate": 4.849136411748306e-05,
-      "loss": 8.635176527604926e-06,
+      "loss": 4.1250437789130956e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 292,
       "tokens/total": 8830688,
-      "tokens/train_per_sec_per_gpu": 34.18,
+      "tokens/train_per_sec_per_gpu": 34.29,
       "tokens/trainable": 133779
     },
     {
       "epoch": 1.14453125,
-      "grad_norm": 0.00026181069551967084,
+      "grad_norm": 0.04661871865391731,
       "learning_rate": 4.8204263076866574e-05,
-      "loss": 4.307490144128678e-06,
+      "loss": 0.00011396713671274483,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0,
+      "ppl": 1.00011,
       "step": 293,
       "tokens/total": 8861008,
-      "tokens/train_per_sec_per_gpu": 36.23,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 134251
     },
     {
       "epoch": 1.1484375,
-      "grad_norm": 0.00046773714711889625,
+      "grad_norm": 0.0046598357148468494,
       "learning_rate": 4.791744483460251e-05,
-      "loss": 4.467175585887162e-06,
+      "loss": 3.974197170464322e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00004,
       "step": 294,
       "tokens/total": 8891760,
-      "tokens/train_per_sec_per_gpu": 33.41,
+      "tokens/train_per_sec_per_gpu": 33.55,
       "tokens/trainable": 134683
     },
     {
       "epoch": 1.15234375,
-      "grad_norm": 0.0002164130419259891,
+      "grad_norm": 0.0034782576840370893,
       "learning_rate": 4.7630921326368736e-05,
-      "loss": 4.914254532195628e-06,
+      "loss": 1.7245467461179942e-05,
       "memory/device_reserved (GiB)": 35.59,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00002,
       "step": 295,
       "tokens/total": 8922192,
-      "tokens/train_per_sec_per_gpu": 35.51,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 135165
     },
     {
       "epoch": 1.15625,
-      "grad_norm": 0.0002447084116283804,
+      "grad_norm": 0.007050055079162121,
       "learning_rate": 4.7344704475577916e-05,
-      "loss": 4.242253453412559e-06,
+      "loss": 6.337351078400388e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.0,
+      "ppl": 1.00006,
       "step": 296,
       "tokens/total": 8952496,
-      "tokens/train_per_sec_per_gpu": 33.18,
+      "tokens/train_per_sec_per_gpu": 33.24,
       "tokens/trainable": 135599
     },
     {
       "epoch": 1.16015625,
-      "grad_norm": 0.00028775070677511394,
+      "grad_norm": 0.0338653139770031,
       "learning_rate": 4.705880619288153e-05,
-      "loss": 4.674366209656e-06,
+      "loss": 0.00016246488667093217,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.0,
+      "ppl": 1.00016,
       "step": 297,
       "tokens/total": 8982768,
-      "tokens/train_per_sec_per_gpu": 32.56,
+      "tokens/train_per_sec_per_gpu": 32.67,
       "tokens/trainable": 136038
     },
     {
       "epoch": 1.1640625,
-      "grad_norm": 0.001555647817440331,
+      "grad_norm": 0.014626193791627884,
       "learning_rate": 4.677323837567412e-05,
-      "loss": 8.829203579807654e-06,
+      "loss": 0.00016142457025125623,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00016,
       "step": 298,
       "tokens/total": 9013248,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.9,
       "tokens/trainable": 136495
     },
     {
       "epoch": 1.16796875,
-      "grad_norm": 0.0002536054525990039,
+      "grad_norm": 0.0024293530732393265,
       "learning_rate": 4.6488012907598146e-05,
-      "loss": 5.5865721151349135e-06,
+      "loss": 2.855894126696512e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 299,
       "tokens/total": 9043632,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.34,
       "tokens/trainable": 136953
     },
     {
       "epoch": 1.171875,
-      "grad_norm": 0.00031900990870781243,
+      "grad_norm": 0.027738185599446297,
       "learning_rate": 4.620314165804964e-05,
-      "loss": 5.917333055549534e-06,
+      "loss": 0.00019799786969088018,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.0002,
       "step": 300,
       "tokens/total": 9073936,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 137412
     },
     {
       "epoch": 1.17578125,
-      "grad_norm": 0.0005231397226452827,
+      "grad_norm": 0.22624947130680084,
       "learning_rate": 4.591863648168407e-05,
-      "loss": 9.367842721985653e-06,
+      "loss": 0.0013185559073463082,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00001,
+      "ppl": 1.00132,
       "step": 301,
       "tokens/total": 9104416,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.68,
       "tokens/trainable": 137907
     },
     {
       "epoch": 1.1796875,
-      "grad_norm": 0.0010045799426734447,
+      "grad_norm": 0.034625228494405746,
       "learning_rate": 4.5634509217923135e-05,
-      "loss": 9.613295333110727e-06,
+      "loss": 0.00012454690295271575,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00012,
       "step": 302,
       "tokens/total": 9134688,
-      "tokens/train_per_sec_per_gpu": 34.2,
+      "tokens/train_per_sec_per_gpu": 34.33,
       "tokens/trainable": 138383
     },
     {
       "epoch": 1.18359375,
-      "grad_norm": 0.0054819462820887566,
+      "grad_norm": 0.005657304544001818,
       "learning_rate": 4.535077169046201e-05,
-      "loss": 2.4776203645160422e-05,
+      "loss": 5.558782140724361e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00006,
       "step": 303,
       "tokens/total": 9165088,
-      "tokens/train_per_sec_per_gpu": 36.35,
+      "tokens/train_per_sec_per_gpu": 36.4,
       "tokens/trainable": 138842
     },
     {
       "epoch": 1.1875,
-      "grad_norm": 0.0005682530463673174,
+      "grad_norm": 0.0075146509334445,
       "learning_rate": 4.506743570677743e-05,
-      "loss": 5.2244549806346186e-06,
+      "loss": 1.902109943330288e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 304,
       "tokens/total": 9195584,
-      "tokens/train_per_sec_per_gpu": 30.61,
+      "tokens/train_per_sec_per_gpu": 30.67,
       "tokens/trainable": 139269
     },
     {
       "epoch": 1.19140625,
-      "grad_norm": 0.003496676916256547,
+      "grad_norm": 0.0024459187407046556,
       "learning_rate": 4.478451305763618e-05,
-      "loss": 2.194346598116681e-05,
+      "loss": 2.344881431781687e-05,
       "memory/device_reserved (GiB)": 35.67,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00002,
       "step": 305,
       "tokens/total": 9225760,
-      "tokens/train_per_sec_per_gpu": 34.36,
+      "tokens/train_per_sec_per_gpu": 34.45,
       "tokens/trainable": 139736
     },
     {
       "epoch": 1.1953125,
-      "grad_norm": 0.000201961855054833,
+      "grad_norm": 0.18014587461948395,
       "learning_rate": 4.450201551660454e-05,
-      "loss": 4.178235940344166e-06,
+      "loss": 0.0005937310634180903,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.0,
+      "ppl": 1.00059,
       "step": 306,
       "tokens/total": 9256352,
-      "tokens/train_per_sec_per_gpu": 36.91,
+      "tokens/train_per_sec_per_gpu": 37.01,
       "tokens/trainable": 140226
     },
     {
       "epoch": 1.19921875,
-      "grad_norm": 0.0004581330285873264,
+      "grad_norm": 0.0010275261010974646,
       "learning_rate": 4.4219954839558276e-05,
-      "loss": 5.6062099247355945e-06,
+      "loss": 2.1100560843478888e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.19,
       "memory/max_allocated (GiB)": 33.19,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 307,
       "tokens/total": 9284256,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 140664
     },
     {
       "epoch": 1.203125,
-      "grad_norm": 0.0016614671330899,
+      "grad_norm": 0.0054944949224591255,
       "learning_rate": 4.393834276419352e-05,
-      "loss": 7.163904228946194e-06,
+      "loss": 5.6230866903206334e-05,
       "memory/device_reserved (GiB)": 35.9,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00001,
+      "ppl": 1.00006,
       "step": 308,
       "tokens/total": 9314768,
-      "tokens/train_per_sec_per_gpu": 29.85,
+      "tokens/train_per_sec_per_gpu": 29.84,
       "tokens/trainable": 141115
     },
     {
       "epoch": 1.20703125,
-      "grad_norm": 0.00019009379320777953,
+      "grad_norm": 0.0006827972829341888,
       "learning_rate": 4.36571910095382e-05,
-      "loss": 3.8271318771876395e-06,
+      "loss": 1.4492828995571472e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 309,
       "tokens/total": 9344976,
-      "tokens/train_per_sec_per_gpu": 28.88,
+      "tokens/train_per_sec_per_gpu": 28.87,
       "tokens/trainable": 141530
     },
     {
       "epoch": 1.2109375,
-      "grad_norm": 0.0038869199343025684,
+      "grad_norm": 0.7971475720405579,
       "learning_rate": 4.337651127546448e-05,
-      "loss": 2.814065919665154e-05,
+      "loss": 0.0021856159437447786,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00003,
+      "ppl": 1.00219,
       "step": 310,
       "tokens/total": 9375280,
-      "tokens/train_per_sec_per_gpu": 34.15,
+      "tokens/train_per_sec_per_gpu": 34.2,
       "tokens/trainable": 141964
     },
     {
       "epoch": 1.21484375,
-      "grad_norm": 0.024828940629959106,
+      "grad_norm": 0.00386450975202024,
       "learning_rate": 4.3096315242201736e-05,
-      "loss": 0.00011992595682386309,
+      "loss": 1.9838389562210068e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00012,
+      "ppl": 1.00002,
       "step": 311,
       "tokens/total": 9405536,
-      "tokens/train_per_sec_per_gpu": 33.95,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 142428
     },
     {
       "epoch": 1.21875,
-      "grad_norm": 0.0017960596596822143,
+      "grad_norm": 0.058126050978899,
       "learning_rate": 4.2816614569850635e-05,
-      "loss": 9.508907169220038e-06,
+      "loss": 0.0002090351772494614,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00001,
+      "ppl": 1.00021,
       "step": 312,
       "tokens/total": 9435696,
-      "tokens/train_per_sec_per_gpu": 38.07,
+      "tokens/train_per_sec_per_gpu": 38.16,
       "tokens/trainable": 142894
     },
     {
       "epoch": 1.22265625,
-      "grad_norm": 0.0007181956898421049,
+      "grad_norm": 0.019857943058013916,
       "learning_rate": 4.2537420897897864e-05,
-      "loss": 8.858227374730632e-06,
+      "loss": 0.00014800178178120404,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00001,
+      "ppl": 1.00015,
       "step": 313,
       "tokens/total": 9466080,
-      "tokens/train_per_sec_per_gpu": 36.3,
+      "tokens/train_per_sec_per_gpu": 36.37,
       "tokens/trainable": 143368
     },
     {
       "epoch": 1.2265625,
-      "grad_norm": 0.7545468211174011,
+      "grad_norm": 0.05821572616696358,
       "learning_rate": 4.225874584473174e-05,
-      "loss": 0.015099202282726765,
+      "loss": 0.00014745524094905704,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.01521,
+      "ppl": 1.00015,
       "step": 314,
       "tokens/total": 9496400,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 143824
     },
     {
       "epoch": 1.23046875,
-      "grad_norm": 0.2810213565826416,
+      "grad_norm": 0.00997456256300211,
       "learning_rate": 4.19806010071587e-05,
-      "loss": 0.0007513084565289319,
+      "loss": 0.00011511147022247314,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00075,
+      "ppl": 1.00012,
       "step": 315,
       "tokens/total": 9526880,
-      "tokens/train_per_sec_per_gpu": 34.45,
+      "tokens/train_per_sec_per_gpu": 34.58,
       "tokens/trainable": 144306
     },
     {
       "epoch": 1.234375,
-      "grad_norm": 0.7471011281013489,
+      "grad_norm": 0.009190126322209835,
       "learning_rate": 4.170299795992081e-05,
-      "loss": 0.0005616231937892735,
+      "loss": 6.155027949716896e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00056,
+      "ppl": 1.00006,
       "step": 316,
       "tokens/total": 9557280,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 144763
     },
     {
       "epoch": 1.23828125,
-      "grad_norm": 0.0004946920089423656,
+      "grad_norm": 0.0016322726150974631,
       "learning_rate": 4.142594825521398e-05,
-      "loss": 5.734345904784277e-06,
+      "loss": 2.0669946025009267e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 317,
       "tokens/total": 9587712,
-      "tokens/train_per_sec_per_gpu": 37.4,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 145255
     },
     {
       "epoch": 1.2421875,
-      "grad_norm": 0.0032017722260206938,
+      "grad_norm": 0.00745917996391654,
       "learning_rate": 4.114946342220728e-05,
-      "loss": 2.6251871531712823e-05,
+      "loss": 5.585006874753162e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00003,
+      "ppl": 1.00006,
       "step": 318,
       "tokens/total": 9618064,
-      "tokens/train_per_sec_per_gpu": 36.53,
+      "tokens/train_per_sec_per_gpu": 36.7,
       "tokens/trainable": 145714
     },
     {
       "epoch": 1.24609375,
-      "grad_norm": 1.1724350452423096,
+      "grad_norm": 0.07343094050884247,
       "learning_rate": 4.087355496656321e-05,
-      "loss": 0.026299288496375084,
+      "loss": 0.00043384716263972223,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.02665,
+      "ppl": 1.00043,
       "step": 319,
       "tokens/total": 9648336,
-      "tokens/train_per_sec_per_gpu": 35.96,
+      "tokens/train_per_sec_per_gpu": 36.04,
       "tokens/trainable": 146200
     },
     {
       "epoch": 1.25,
-      "grad_norm": 0.00045995338587090373,
+      "grad_norm": 0.007983052171766758,
       "learning_rate": 4.05982343699588e-05,
-      "loss": 8.491813787259161e-06,
+      "loss": 4.524823816609569e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00001,
+      "ppl": 1.00005,
       "step": 320,
       "tokens/total": 9678688,
-      "tokens/train_per_sec_per_gpu": 34.23,
+      "tokens/train_per_sec_per_gpu": 34.28,
       "tokens/trainable": 146668
     },
     {
       "epoch": 1.25390625,
-      "grad_norm": 0.0067353262566030025,
+      "grad_norm": 0.01599739119410515,
       "learning_rate": 4.0323513089607876e-05,
-      "loss": 6.0428461438277736e-05,
+      "loss": 6.388167093973607e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
       "ppl": 1.00006,
       "step": 321,
       "tokens/total": 9708848,
-      "tokens/train_per_sec_per_gpu": 35.24,
+      "tokens/train_per_sec_per_gpu": 35.35,
       "tokens/trainable": 147114
     },
     {
       "epoch": 1.2578125,
-      "grad_norm": 0.05487794429063797,
+      "grad_norm": 0.0042083412408828735,
       "learning_rate": 4.004940255778431e-05,
-      "loss": 0.0002498509711585939,
+      "loss": 2.1792850020574406e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00025,
+      "ppl": 1.00002,
       "step": 322,
       "tokens/total": 9739360,
-      "tokens/train_per_sec_per_gpu": 35.08,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 147564
     },
     {
       "epoch": 1.26171875,
-      "grad_norm": 0.0011818762868642807,
+      "grad_norm": 0.009429940953850746,
       "learning_rate": 3.977591418134619e-05,
-      "loss": 1.1004886800947133e-05,
+      "loss": 3.834946619463153e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 323,
       "tokens/total": 9769776,
-      "tokens/train_per_sec_per_gpu": 34.14,
+      "tokens/train_per_sec_per_gpu": 34.24,
       "tokens/trainable": 148013
     },
     {
       "epoch": 1.265625,
-      "grad_norm": 0.018019674345850945,
+      "grad_norm": 0.05088210478425026,
       "learning_rate": 3.95030593412612e-05,
-      "loss": 0.00021162032498978078,
+      "loss": 0.00018766832363326102,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00021,
+      "ppl": 1.00019,
       "step": 324,
       "tokens/total": 9800096,
-      "tokens/train_per_sec_per_gpu": 37.99,
+      "tokens/train_per_sec_per_gpu": 38.06,
       "tokens/trainable": 148470
     },
     {
       "epoch": 1.26953125,
-      "grad_norm": 1.5907806158065796,
+      "grad_norm": 0.9416317939758301,
       "learning_rate": 3.923084939213296e-05,
-      "loss": 0.016217660158872604,
+      "loss": 0.01083211787045002,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.01635,
+      "ppl": 1.01089,
       "step": 325,
       "tokens/total": 9830304,
-      "tokens/train_per_sec_per_gpu": 36.44,
+      "tokens/train_per_sec_per_gpu": 36.53,
       "tokens/trainable": 148926
     },
     {
       "epoch": 1.2734375,
-      "grad_norm": 0.004153774119913578,
+      "grad_norm": 0.00033186975633725524,
       "learning_rate": 3.895929566172861e-05,
-      "loss": 3.801286584348418e-05,
+      "loss": 8.612486453785095e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 326,
       "tokens/total": 9860608,
-      "tokens/train_per_sec_per_gpu": 37.72,
+      "tokens/train_per_sec_per_gpu": 37.82,
       "tokens/trainable": 149420
     },
     {
       "epoch": 1.27734375,
-      "grad_norm": 0.0029778245370835066,
+      "grad_norm": 0.0005477021913975477,
       "learning_rate": 3.868840945050728e-05,
-      "loss": 5.526735549210571e-05,
+      "loss": 1.5825649825274013e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00006,
+      "ppl": 1.00002,
       "step": 327,
       "tokens/total": 9890560,
-      "tokens/train_per_sec_per_gpu": 39.32,
+      "tokens/train_per_sec_per_gpu": 39.41,
       "tokens/trainable": 149873
     },
     {
       "epoch": 1.28125,
-      "grad_norm": 0.028163742274045944,
+      "grad_norm": 0.0008259841124527156,
       "learning_rate": 3.841820203114995e-05,
-      "loss": 0.00039056455716490746,
+      "loss": 1.642670395085588e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00039,
+      "ppl": 1.00002,
       "step": 328,
       "tokens/total": 9920880,
-      "tokens/train_per_sec_per_gpu": 33.78,
+      "tokens/train_per_sec_per_gpu": 33.86,
       "tokens/trainable": 150347
     },
     {
       "epoch": 1.28515625,
-      "grad_norm": 0.04141407459974289,
+      "grad_norm": 0.0011461537797003984,
       "learning_rate": 3.814868464809027e-05,
-      "loss": 0.0005860928213223815,
+      "loss": 1.7516158550279215e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00059,
+      "ppl": 1.00002,
       "step": 329,
       "tokens/total": 9951280,
-      "tokens/train_per_sec_per_gpu": 31.72,
+      "tokens/train_per_sec_per_gpu": 31.83,
       "tokens/trainable": 150822
     },
     {
       "epoch": 1.2890625,
-      "grad_norm": 0.004758972208946943,
+      "grad_norm": 0.0022519829217344522,
       "learning_rate": 3.787986851704667e-05,
-      "loss": 9.08115180209279e-05,
+      "loss": 2.000835411308799e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 330,
       "tokens/total": 9981600,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.06,
       "tokens/trainable": 151266
     },
     {
       "epoch": 1.29296875,
-      "grad_norm": 0.04043704643845558,
+      "grad_norm": 0.08718931674957275,
       "learning_rate": 3.7611764824555654e-05,
-      "loss": 0.0005757657927460968,
+      "loss": 0.0003773289208766073,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00058,
+      "ppl": 1.00038,
       "step": 331,
       "tokens/total": 10012016,
-      "tokens/train_per_sec_per_gpu": 37.11,
+      "tokens/train_per_sec_per_gpu": 37.15,
       "tokens/trainable": 151734
     },
     {
       "epoch": 1.296875,
-      "grad_norm": 0.064565509557724,
+      "grad_norm": 0.000883373199030757,
       "learning_rate": 3.734438472750619e-05,
-      "loss": 0.00043528492096811533,
+      "loss": 1.1746728887374047e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00044,
+      "ppl": 1.00001,
       "step": 332,
       "tokens/total": 10042448,
-      "tokens/train_per_sec_per_gpu": 33.99,
+      "tokens/train_per_sec_per_gpu": 34.01,
       "tokens/trainable": 152184
     },
     {
       "epoch": 1.30078125,
-      "grad_norm": 0.051213983446359634,
+      "grad_norm": 0.011426394805312157,
       "learning_rate": 3.707773935267552e-05,
-      "loss": 0.000528274686075747,
+      "loss": 5.219353988650255e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00053,
+      "ppl": 1.00005,
       "step": 333,
       "tokens/total": 10073024,
-      "tokens/train_per_sec_per_gpu": 31.79,
+      "tokens/train_per_sec_per_gpu": 31.85,
       "tokens/trainable": 152623
     },
     {
       "epoch": 1.3046875,
-      "grad_norm": 0.0067980666644871235,
+      "grad_norm": 0.5781233310699463,
       "learning_rate": 3.68118397962661e-05,
-      "loss": 0.00013989521539770067,
+      "loss": 0.0034865224733948708,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00014,
+      "ppl": 1.00349,
       "step": 334,
       "tokens/total": 10103504,
-      "tokens/train_per_sec_per_gpu": 33.34,
+      "tokens/train_per_sec_per_gpu": 33.44,
       "tokens/trainable": 153040
     },
     {
       "epoch": 1.30859375,
-      "grad_norm": 0.28768453001976013,
+      "grad_norm": 0.7213758230209351,
       "learning_rate": 3.654669712344384e-05,
-      "loss": 0.0035491236485540867,
+      "loss": 0.004715530201792717,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.31,
       "memory/max_allocated (GiB)": 33.31,
-      "ppl": 1.00356,
+      "ppl": 1.00473,
       "step": 335,
       "tokens/total": 10131632,
-      "tokens/train_per_sec_per_gpu": 31.7,
+      "tokens/train_per_sec_per_gpu": 31.78,
       "tokens/trainable": 153463
     },
     {
       "epoch": 1.3125,
-      "grad_norm": 0.04292941838502884,
+      "grad_norm": 0.03131686523556709,
       "learning_rate": 3.628232236787763e-05,
-      "loss": 0.0007545957923866808,
+      "loss": 0.0002346257824683562,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00075,
+      "ppl": 1.00023,
       "step": 336,
       "tokens/total": 10161824,
-      "tokens/train_per_sec_per_gpu": 33.22,
+      "tokens/train_per_sec_per_gpu": 33.31,
       "tokens/trainable": 153908
     },
     {
       "epoch": 1.31640625,
-      "grad_norm": 0.13187745213508606,
+      "grad_norm": 0.03332929685711861,
       "learning_rate": 3.6018726531280144e-05,
-      "loss": 0.0011231126263737679,
+      "loss": 0.00011981122952420264,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00112,
+      "ppl": 1.00012,
       "step": 337,
       "tokens/total": 10192448,
-      "tokens/train_per_sec_per_gpu": 33.63,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 154387
     },
     {
       "epoch": 1.3203125,
-      "grad_norm": 0.06429488956928253,
+      "grad_norm": 0.0045038131065666676,
       "learning_rate": 3.575592058295017e-05,
-      "loss": 0.0004924655659124255,
+      "loss": 3.5494955227477476e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00049,
+      "ppl": 1.00004,
       "step": 338,
       "tokens/total": 10222704,
-      "tokens/train_per_sec_per_gpu": 33.82,
+      "tokens/train_per_sec_per_gpu": 33.94,
       "tokens/trainable": 154836
     },
     {
       "epoch": 1.32421875,
-      "grad_norm": 0.006816778797656298,
+      "grad_norm": 0.0012434936361387372,
       "learning_rate": 3.549391545931585e-05,
-      "loss": 8.63251625560224e-05,
+      "loss": 1.7894679331220686e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 339,
       "tokens/total": 10253168,
-      "tokens/train_per_sec_per_gpu": 29.73,
+      "tokens/train_per_sec_per_gpu": 29.81,
       "tokens/trainable": 155261
     },
     {
       "epoch": 1.328125,
-      "grad_norm": 0.04606792330741882,
+      "grad_norm": 0.3250766098499298,
       "learning_rate": 3.5232722063479914e-05,
-      "loss": 0.0003515402786433697,
+      "loss": 0.00279021542519331,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00035,
+      "ppl": 1.00279,
       "step": 340,
       "tokens/total": 10283552,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.24,
       "tokens/trainable": 155674
     },
     {
       "epoch": 1.33203125,
-      "grad_norm": 0.24907881021499634,
+      "grad_norm": 0.036490436643362045,
       "learning_rate": 3.49723512647657e-05,
-      "loss": 0.000742567703127861,
+      "loss": 0.00013251493510324508,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00074,
+      "ppl": 1.00013,
       "step": 341,
       "tokens/total": 10313872,
-      "tokens/train_per_sec_per_gpu": 35.67,
+      "tokens/train_per_sec_per_gpu": 35.78,
       "tokens/trainable": 156164
     },
     {
       "epoch": 1.3359375,
-      "grad_norm": 0.011238012462854385,
+      "grad_norm": 0.2706057131290436,
       "learning_rate": 3.471281389826491e-05,
-      "loss": 0.0001773187832441181,
+      "loss": 0.0021429890766739845,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00018,
+      "ppl": 1.00215,
       "step": 342,
       "tokens/total": 10344256,
-      "tokens/train_per_sec_per_gpu": 33.47,
+      "tokens/train_per_sec_per_gpu": 33.65,
       "tokens/trainable": 156611
     },
     {
       "epoch": 1.33984375,
-      "grad_norm": 0.00949561782181263,
+      "grad_norm": 0.01915346086025238,
       "learning_rate": 3.4454120764386764e-05,
-      "loss": 0.00018969883967656642,
+      "loss": 0.00013777356070932,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00019,
+      "ppl": 1.00014,
       "step": 343,
       "tokens/total": 10374544,
-      "tokens/train_per_sec_per_gpu": 33.58,
+      "tokens/train_per_sec_per_gpu": 33.68,
       "tokens/trainable": 157073
     },
     {
       "epoch": 1.34375,
-      "grad_norm": 0.10763411223888397,
+      "grad_norm": 0.02540464885532856,
       "learning_rate": 3.4196282628408526e-05,
-      "loss": 0.0007583287660963833,
+      "loss": 5.135099490871653e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00076,
+      "ppl": 1.00005,
       "step": 344,
       "tokens/total": 10405040,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 157523
     },
     {
       "epoch": 1.34765625,
-      "grad_norm": 0.15938052535057068,
+      "grad_norm": 0.39257746934890747,
       "learning_rate": 3.3939310220027456e-05,
-      "loss": 0.0014999746344983578,
+      "loss": 0.009803109802305698,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0015,
+      "ppl": 1.00985,
       "step": 345,
       "tokens/total": 10435424,
-      "tokens/train_per_sec_per_gpu": 35.07,
+      "tokens/train_per_sec_per_gpu": 35.16,
       "tokens/trainable": 157964
     },
     {
       "epoch": 1.3515625,
-      "grad_norm": 0.00165572261903435,
+      "grad_norm": 0.0029070202726870775,
       "learning_rate": 3.3683214232914404e-05,
-      "loss": 3.833783557638526e-05,
+      "loss": 2.9972559786983766e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00004,
+      "ppl": 1.00003,
       "step": 346,
       "tokens/total": 10465760,
-      "tokens/train_per_sec_per_gpu": 30.35,
+      "tokens/train_per_sec_per_gpu": 30.44,
       "tokens/trainable": 158390
     },
     {
       "epoch": 1.35546875,
-      "grad_norm": 0.02406073547899723,
+      "grad_norm": 0.020276719704270363,
       "learning_rate": 3.342800532426873e-05,
-      "loss": 0.00039186165668070316,
+      "loss": 7.613154593855143e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00039,
+      "ppl": 1.00008,
       "step": 347,
       "tokens/total": 10495904,
-      "tokens/train_per_sec_per_gpu": 35.47,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 158848
     },
     {
       "epoch": 1.359375,
-      "grad_norm": 0.045792482793331146,
+      "grad_norm": 0.14866818487644196,
       "learning_rate": 3.317369411437484e-05,
-      "loss": 0.0006161610363051295,
+      "loss": 0.0011244683992117643,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 34.0,
       "memory/max_allocated (GiB)": 34.0,
-      "ppl": 1.00062,
+      "ppl": 1.00113,
       "step": 348,
       "tokens/total": 10526624,
-      "tokens/train_per_sec_per_gpu": 28.63,
+      "tokens/train_per_sec_per_gpu": 28.71,
       "tokens/trainable": 159270
     },
     {
       "epoch": 1.36328125,
-      "grad_norm": 0.18648307025432587,
+      "grad_norm": 0.015379665419459343,
       "learning_rate": 3.292029118616024e-05,
-      "loss": 0.0012868957128375769,
+      "loss": 7.920589268906042e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00129,
+      "ppl": 1.00008,
       "step": 349,
       "tokens/total": 10556752,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.51,
       "tokens/trainable": 159720
     },
     {
       "epoch": 1.3671875,
-      "grad_norm": 0.015933886170387268,
+      "grad_norm": 0.0023695260751992464,
       "learning_rate": 3.266780708475511e-05,
-      "loss": 0.0002091687492793426,
+      "loss": 1.4566459867637604e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.99,
       "memory/max_allocated (GiB)": 33.99,
-      "ppl": 1.00021,
+      "ppl": 1.00001,
       "step": 350,
       "tokens/total": 10587264,
-      "tokens/train_per_sec_per_gpu": 35.09,
+      "tokens/train_per_sec_per_gpu": 35.14,
       "tokens/trainable": 160187
     },
     {
       "epoch": 1.37109375,
-      "grad_norm": 0.008724886924028397,
+      "grad_norm": 0.00954374298453331,
       "learning_rate": 3.241625231705354e-05,
-      "loss": 0.00013179841334931552,
+      "loss": 6.530048267450184e-05,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00013,
+      "ppl": 1.00007,
       "step": 351,
       "tokens/total": 10617504,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.63,
       "tokens/trainable": 160645
     },
     {
       "epoch": 1.375,
-      "grad_norm": 0.06377559900283813,
+      "grad_norm": 0.6504904627799988,
       "learning_rate": 3.216563735127618e-05,
-      "loss": 0.0008062056731432676,
+      "loss": 0.01059151440858841,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00081,
+      "ppl": 1.01065,
       "step": 352,
       "tokens/total": 10647760,
-      "tokens/train_per_sec_per_gpu": 34.42,
+      "tokens/train_per_sec_per_gpu": 34.49,
       "tokens/trainable": 161102
     },
     {
       "epoch": 1.37890625,
-      "grad_norm": 0.0037202269304543734,
+      "grad_norm": 0.024732626974582672,
       "learning_rate": 3.191597261653475e-05,
-      "loss": 7.238816760946065e-05,
+      "loss": 0.0001429672265658155,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00007,
+      "ppl": 1.00014,
       "step": 353,
       "tokens/total": 10678080,
-      "tokens/train_per_sec_per_gpu": 28.86,
+      "tokens/train_per_sec_per_gpu": 28.93,
       "tokens/trainable": 161555
     },
     {
       "epoch": 1.3828125,
-      "grad_norm": 0.0021510994993150234,
+      "grad_norm": 0.037309419363737106,
       "learning_rate": 3.166726850239794e-05,
-      "loss": 3.6805788113269955e-05,
+      "loss": 7.214388460852206e-05,
       "memory/device_reserved (GiB)": 35.41,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00004,
+      "ppl": 1.00007,
       "step": 354,
       "tokens/total": 10708544,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.75,
       "tokens/trainable": 162020
     },
     {
       "epoch": 1.38671875,
-      "grad_norm": 0.002642499515786767,
+      "grad_norm": 0.00885045062750578,
       "learning_rate": 3.141953535845912e-05,
-      "loss": 4.371708200778812e-05,
+      "loss": 0.00011830432777060196,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00012,
       "step": 355,
       "tokens/total": 10739024,
-      "tokens/train_per_sec_per_gpu": 32.05,
+      "tokens/train_per_sec_per_gpu": 32.1,
       "tokens/trainable": 162448
     },
     {
       "epoch": 1.390625,
-      "grad_norm": 0.0038646068423986435,
+      "grad_norm": 0.06791900098323822,
       "learning_rate": 3.11727834939056e-05,
-      "loss": 7.11614266037941e-05,
+      "loss": 0.0003609730047173798,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.61,
       "memory/max_allocated (GiB)": 33.61,
-      "ppl": 1.00007,
+      "ppl": 1.00036,
       "step": 356,
       "tokens/total": 10769024,
-      "tokens/train_per_sec_per_gpu": 28.2,
+      "tokens/train_per_sec_per_gpu": 28.26,
       "tokens/trainable": 162863
     },
     {
       "epoch": 1.39453125,
-      "grad_norm": 0.017895536497235298,
+      "grad_norm": 0.0019508687546476722,
       "learning_rate": 3.092702317708967e-05,
-      "loss": 5.4065520089352503e-05,
+      "loss": 3.1739040423417464e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 357,
       "tokens/total": 10799328,
-      "tokens/train_per_sec_per_gpu": 34.94,
+      "tokens/train_per_sec_per_gpu": 34.89,
       "tokens/trainable": 163318
     },
     {
       "epoch": 1.3984375,
-      "grad_norm": 0.8479411602020264,
+      "grad_norm": 0.7139555811882019,
       "learning_rate": 3.0682264635101276e-05,
-      "loss": 0.01497839204967022,
+      "loss": 0.010268578305840492,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.01509,
+      "ppl": 1.01032,
       "step": 358,
       "tokens/total": 10829488,
-      "tokens/train_per_sec_per_gpu": 32.89,
+      "tokens/train_per_sec_per_gpu": 33.08,
       "tokens/trainable": 163767
     },
     {
       "epoch": 1.40234375,
-      "grad_norm": 0.0015704578254371881,
+      "grad_norm": 0.006852464284747839,
       "learning_rate": 3.0438518053342407e-05,
-      "loss": 3.0390210667974316e-05,
+      "loss": 4.873241778113879e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00005,
       "step": 359,
       "tokens/total": 10859936,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 164208
     },
     {
       "epoch": 1.40625,
-      "grad_norm": 0.014669318683445454,
+      "grad_norm": 0.0018247900297865272,
       "learning_rate": 3.0195793575103266e-05,
-      "loss": 0.00010030368866864592,
+      "loss": 3.88835760531947e-05,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.0001,
+      "ppl": 1.00004,
       "step": 360,
       "tokens/total": 10890400,
-      "tokens/train_per_sec_per_gpu": 34.08,
+      "tokens/train_per_sec_per_gpu": 34.18,
       "tokens/trainable": 164655
     },
     {
       "epoch": 1.41015625,
-      "grad_norm": 0.3479421138763428,
+      "grad_norm": 0.0436873696744442,
       "learning_rate": 2.9954101301140146e-05,
-      "loss": 0.0026096655055880547,
+      "loss": 0.0003687943681143224,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00261,
+      "ppl": 1.00037,
       "step": 361,
       "tokens/total": 10920624,
-      "tokens/train_per_sec_per_gpu": 34.1,
+      "tokens/train_per_sec_per_gpu": 34.15,
       "tokens/trainable": 165100
     },
     {
       "epoch": 1.4140625,
-      "grad_norm": 0.004394443240016699,
+      "grad_norm": 0.26735150814056396,
       "learning_rate": 2.9713451289255123e-05,
-      "loss": 6.116658914834261e-05,
+      "loss": 0.00039605735219083726,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00006,
+      "ppl": 1.0004,
       "step": 362,
       "tokens/total": 10951136,
-      "tokens/train_per_sec_per_gpu": 34.84,
+      "tokens/train_per_sec_per_gpu": 34.91,
       "tokens/trainable": 165587
     },
     {
       "epoch": 1.41796875,
-      "grad_norm": 0.016197621822357178,
+      "grad_norm": 0.0964483991265297,
       "learning_rate": 2.9473853553877484e-05,
-      "loss": 0.0001775357231963426,
+      "loss": 0.0007330900407396257,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00018,
+      "ppl": 1.00073,
       "step": 363,
       "tokens/total": 10981344,
-      "tokens/train_per_sec_per_gpu": 34.09,
+      "tokens/train_per_sec_per_gpu": 34.14,
       "tokens/trainable": 166034
     },
     {
       "epoch": 1.421875,
-      "grad_norm": 0.09478334337472916,
+      "grad_norm": 0.05367618799209595,
       "learning_rate": 2.9235318065647e-05,
-      "loss": 0.0005033796769566834,
+      "loss": 0.0005384897813200951,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.0005,
+      "ppl": 1.00054,
       "step": 364,
       "tokens/total": 11011552,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.1,
       "tokens/trainable": 166485
     },
     {
       "epoch": 1.42578125,
-      "grad_norm": 0.014899010770022869,
+      "grad_norm": 0.10067912936210632,
       "learning_rate": 2.8997854750998964e-05,
-      "loss": 0.0001904651871882379,
+      "loss": 0.0006868956843391061,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00019,
+      "ppl": 1.00069,
       "step": 365,
       "tokens/total": 11041872,
-      "tokens/train_per_sec_per_gpu": 36.32,
+      "tokens/train_per_sec_per_gpu": 36.36,
       "tokens/trainable": 166973
     },
     {
       "epoch": 1.4296875,
-      "grad_norm": 0.005555527750402689,
+      "grad_norm": 0.02697882056236267,
       "learning_rate": 2.8761473491751258e-05,
-      "loss": 0.00010058133921120316,
+      "loss": 0.0002464794088155031,
       "memory/device_reserved (GiB)": 35.84,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.0001,
+      "ppl": 1.00025,
       "step": 366,
       "tokens/total": 11072192,
-      "tokens/train_per_sec_per_gpu": 29.1,
+      "tokens/train_per_sec_per_gpu": 29.12,
       "tokens/trainable": 167414
     },
     {
       "epoch": 1.43359375,
-      "grad_norm": 0.20706292986869812,
+      "grad_norm": 0.0026214183308184147,
       "learning_rate": 2.8526184124692883e-05,
-      "loss": 0.0005888720043003559,
+      "loss": 5.2629769925260916e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00059,
+      "ppl": 1.00005,
       "step": 367,
       "tokens/total": 11102736,
-      "tokens/train_per_sec_per_gpu": 30.11,
+      "tokens/train_per_sec_per_gpu": 30.23,
       "tokens/trainable": 167851
     },
     {
       "epoch": 1.4375,
-      "grad_norm": 0.0007191727054305375,
+      "grad_norm": 0.0018140808679163456,
       "learning_rate": 2.829199644117484e-05,
-      "loss": 1.2304372830840293e-05,
+      "loss": 2.9512597393477336e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00001,
+      "ppl": 1.00003,
       "step": 368,
       "tokens/total": 11133056,
-      "tokens/train_per_sec_per_gpu": 35.48,
+      "tokens/train_per_sec_per_gpu": 35.56,
       "tokens/trainable": 168294
     },
     {
       "epoch": 1.44140625,
-      "grad_norm": 0.030878128483891487,
+      "grad_norm": 0.012045321986079216,
       "learning_rate": 2.8058920186702553e-05,
-      "loss": 0.0002043281274382025,
+      "loss": 7.606908184243366e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.0002,
+      "ppl": 1.00008,
       "step": 369,
       "tokens/total": 11163568,
-      "tokens/train_per_sec_per_gpu": 32.08,
+      "tokens/train_per_sec_per_gpu": 32.23,
       "tokens/trainable": 168769
     },
     {
       "epoch": 1.4453125,
-      "grad_norm": 0.008817057125270367,
+      "grad_norm": 0.013453424908220768,
       "learning_rate": 2.782696506053033e-05,
-      "loss": 0.00011464582348708063,
+      "loss": 0.00015325279673561454,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00011,
+      "ppl": 1.00015,
       "step": 370,
       "tokens/total": 11193936,
-      "tokens/train_per_sec_per_gpu": 36.59,
+      "tokens/train_per_sec_per_gpu": 36.83,
       "tokens/trainable": 169270
     },
     {
       "epoch": 1.44921875,
-      "grad_norm": 0.023856336250901222,
+      "grad_norm": 0.003458227962255478,
       "learning_rate": 2.7596140715257824e-05,
-      "loss": 0.00013431125262286514,
+      "loss": 5.63332432648167e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00013,
+      "ppl": 1.00006,
       "step": 371,
       "tokens/total": 11224480,
-      "tokens/train_per_sec_per_gpu": 35.06,
+      "tokens/train_per_sec_per_gpu": 35.08,
       "tokens/trainable": 169711
     },
     {
       "epoch": 1.453125,
-      "grad_norm": 0.010502993129193783,
+      "grad_norm": 0.0063622924499213696,
       "learning_rate": 2.7366456756428184e-05,
-      "loss": 0.00015104333579074591,
+      "loss": 9.534892160445452e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00015,
+      "ppl": 1.0001,
       "step": 372,
       "tokens/total": 11254912,
-      "tokens/train_per_sec_per_gpu": 32.81,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 170185
     },
     {
       "epoch": 1.45703125,
-      "grad_norm": 0.026970423758029938,
+      "grad_norm": 0.04558980092406273,
       "learning_rate": 2.7137922742128486e-05,
-      "loss": 0.00026081278338097036,
+      "loss": 0.000613630109000951,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00026,
+      "ppl": 1.00061,
       "step": 373,
       "tokens/total": 11285104,
-      "tokens/train_per_sec_per_gpu": 33.08,
+      "tokens/train_per_sec_per_gpu": 33.12,
       "tokens/trainable": 170650
     },
     {
       "epoch": 1.4609375,
-      "grad_norm": 0.020325109362602234,
+      "grad_norm": 0.014873786829411983,
       "learning_rate": 2.691054818259188e-05,
-      "loss": 0.00014532633940689266,
+      "loss": 0.0002038514503510669,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00015,
+      "ppl": 1.0002,
       "step": 374,
       "tokens/total": 11315600,
-      "tokens/train_per_sec_per_gpu": 31.98,
+      "tokens/train_per_sec_per_gpu": 31.99,
       "tokens/trainable": 171126
     },
     {
       "epoch": 1.46484375,
-      "grad_norm": 0.033260464668273926,
+      "grad_norm": 0.058739252388477325,
       "learning_rate": 2.6684342539801933e-05,
-      "loss": 0.00045571548980660737,
+      "loss": 0.0010346119524911046,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00046,
+      "ppl": 1.00104,
       "step": 375,
       "tokens/total": 11345776,
       "tokens/train_per_sec_per_gpu": 35.6,
@@ -5261,713 +5261,713 @@
     },
     {
       "epoch": 1.46875,
-      "grad_norm": 0.017290709540247917,
+      "grad_norm": 0.24105991423130035,
       "learning_rate": 2.645931522709877e-05,
-      "loss": 0.00014910403115209192,
+      "loss": 0.0035716122947633266,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00015,
+      "ppl": 1.00358,
       "step": 376,
       "tokens/total": 11376208,
-      "tokens/train_per_sec_per_gpu": 28.09,
+      "tokens/train_per_sec_per_gpu": 28.14,
       "tokens/trainable": 172017
     },
     {
       "epoch": 1.47265625,
-      "grad_norm": 0.04457363486289978,
+      "grad_norm": 0.08793950080871582,
       "learning_rate": 2.6235475608787365e-05,
-      "loss": 6.834026135038584e-05,
+      "loss": 0.0007833336712792516,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00007,
+      "ppl": 1.00078,
       "step": 377,
       "tokens/total": 11406512,
-      "tokens/train_per_sec_per_gpu": 35.99,
+      "tokens/train_per_sec_per_gpu": 36.07,
       "tokens/trainable": 172493
     },
     {
       "epoch": 1.4765625,
-      "grad_norm": 0.002268735785037279,
+      "grad_norm": 0.009154544211924076,
       "learning_rate": 2.6012832999747916e-05,
-      "loss": 3.245133120799437e-05,
+      "loss": 0.00011752049613278359,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00003,
+      "ppl": 1.00012,
       "step": 378,
       "tokens/total": 11436544,
-      "tokens/train_per_sec_per_gpu": 34.27,
+      "tokens/train_per_sec_per_gpu": 34.4,
       "tokens/trainable": 172951
     },
     {
       "epoch": 1.48046875,
-      "grad_norm": 0.0038171466439962387,
+      "grad_norm": 0.012918495573103428,
       "learning_rate": 2.579139666504821e-05,
-      "loss": 5.4866883147042245e-05,
+      "loss": 0.00011567945330170915,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00005,
+      "ppl": 1.00012,
       "step": 379,
       "tokens/total": 11467008,
-      "tokens/train_per_sec_per_gpu": 35.63,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 173435
     },
     {
       "epoch": 1.484375,
-      "grad_norm": 0.005268535576760769,
+      "grad_norm": 0.008966504596173763,
       "learning_rate": 2.557117581955798e-05,
-      "loss": 8.241234172601253e-05,
+      "loss": 8.938623068388551e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00008,
+      "ppl": 1.00009,
       "step": 380,
       "tokens/total": 11497184,
-      "tokens/train_per_sec_per_gpu": 30.4,
+      "tokens/train_per_sec_per_gpu": 30.47,
       "tokens/trainable": 173900
     },
     {
       "epoch": 1.48828125,
-      "grad_norm": 0.016057243570685387,
+      "grad_norm": 0.03181751072406769,
       "learning_rate": 2.5352179627565532e-05,
-      "loss": 0.00014235377602744848,
+      "loss": 0.00013083787052892148,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00013,
       "step": 381,
       "tokens/total": 11527600,
-      "tokens/train_per_sec_per_gpu": 34.12,
+      "tokens/train_per_sec_per_gpu": 34.13,
       "tokens/trainable": 174371
     },
     {
       "epoch": 1.4921875,
-      "grad_norm": 0.005853456910699606,
+      "grad_norm": 0.0009072935208678246,
       "learning_rate": 2.5134417202396277e-05,
-      "loss": 5.001476893085055e-05,
+      "loss": 2.5600436856620945e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 382,
       "tokens/total": 11557808,
-      "tokens/train_per_sec_per_gpu": 32.76,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 174826
     },
     {
       "epoch": 1.49609375,
-      "grad_norm": 0.0031664848793298006,
+      "grad_norm": 0.0050596860237419605,
       "learning_rate": 2.491789760603361e-05,
-      "loss": 4.210277256788686e-05,
+      "loss": 5.778766353614628e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 383,
       "tokens/total": 11588352,
-      "tokens/train_per_sec_per_gpu": 32.8,
+      "tokens/train_per_sec_per_gpu": 32.88,
       "tokens/trainable": 175276
     },
     {
       "epoch": 1.5,
-      "grad_norm": 0.08838633447885513,
+      "grad_norm": 0.008304890245199203,
       "learning_rate": 2.4702629848741764e-05,
-      "loss": 0.0010394920827820897,
+      "loss": 0.00012505470658652484,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00104,
+      "ppl": 1.00013,
       "step": 384,
       "tokens/total": 11618640,
-      "tokens/train_per_sec_per_gpu": 33.69,
+      "tokens/train_per_sec_per_gpu": 33.71,
       "tokens/trainable": 175720
     },
     {
       "epoch": 1.50390625,
-      "grad_norm": 0.09446703642606735,
+      "grad_norm": 0.0035276354756206274,
       "learning_rate": 2.4488622888690785e-05,
-      "loss": 0.0005370234721340239,
+      "loss": 4.851898120250553e-05,
       "memory/device_reserved (GiB)": 35.89,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00054,
+      "ppl": 1.00005,
       "step": 385,
       "tokens/total": 11649072,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 176203
     },
     {
       "epoch": 1.5078125,
-      "grad_norm": 0.0013355027185752988,
+      "grad_norm": 0.001593019813299179,
       "learning_rate": 2.427588563158384e-05,
-      "loss": 1.903088195831515e-05,
+      "loss": 3.330651088617742e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 386,
       "tokens/total": 11679552,
-      "tokens/train_per_sec_per_gpu": 34.29,
+      "tokens/train_per_sec_per_gpu": 34.19,
       "tokens/trainable": 176651
     },
     {
       "epoch": 1.51171875,
-      "grad_norm": 0.011693151667714119,
+      "grad_norm": 0.002167076338082552,
       "learning_rate": 2.406442693028651e-05,
-      "loss": 7.680666749365628e-05,
+      "loss": 3.344817378092557e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00008,
+      "ppl": 1.00003,
       "step": 387,
       "tokens/total": 11709760,
-      "tokens/train_per_sec_per_gpu": 32.7,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 177115
     },
     {
       "epoch": 1.515625,
-      "grad_norm": 0.09523260593414307,
+      "grad_norm": 0.003447546623647213,
       "learning_rate": 2.3854255584458547e-05,
-      "loss": 0.00073521543527022,
+      "loss": 5.5277254432439804e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00074,
+      "ppl": 1.00006,
       "step": 388,
       "tokens/total": 11740288,
-      "tokens/train_per_sec_per_gpu": 36.09,
+      "tokens/train_per_sec_per_gpu": 36.13,
       "tokens/trainable": 177601
     },
     {
       "epoch": 1.51953125,
-      "grad_norm": 0.005971741396933794,
+      "grad_norm": 0.0052979388274252415,
       "learning_rate": 2.3645380340187508e-05,
-      "loss": 6.468063656939194e-05,
+      "loss": 4.56162124464754e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00006,
+      "ppl": 1.00005,
       "step": 389,
       "tokens/total": 11770592,
-      "tokens/train_per_sec_per_gpu": 32.99,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 178050
     },
     {
       "epoch": 1.5234375,
-      "grad_norm": 0.0010247502941638231,
+      "grad_norm": 0.003032457083463669,
       "learning_rate": 2.3437809889624914e-05,
-      "loss": 1.5989648090908304e-05,
+      "loss": 3.829343768302351e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 390,
       "tokens/total": 11800752,
-      "tokens/train_per_sec_per_gpu": 36.38,
+      "tokens/train_per_sec_per_gpu": 36.42,
       "tokens/trainable": 178531
     },
     {
       "epoch": 1.52734375,
-      "grad_norm": 0.054420940577983856,
+      "grad_norm": 0.03542603179812431,
       "learning_rate": 2.3231552870624487e-05,
-      "loss": 0.0004186803416814655,
+      "loss": 0.0003605492820497602,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00042,
+      "ppl": 1.00036,
       "step": 391,
       "tokens/total": 11831360,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.98,
       "tokens/trainable": 179000
     },
     {
       "epoch": 1.53125,
-      "grad_norm": 0.002925195964053273,
+      "grad_norm": 0.004965933505445719,
       "learning_rate": 2.3026617866382657e-05,
-      "loss": 1.531536145193968e-05,
+      "loss": 7.091004226822406e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 392,
       "tokens/total": 11861552,
-      "tokens/train_per_sec_per_gpu": 33.06,
+      "tokens/train_per_sec_per_gpu": 33.1,
       "tokens/trainable": 179457
     },
     {
       "epoch": 1.53515625,
-      "grad_norm": 1.6402941942214966,
+      "grad_norm": 0.006339129991829395,
       "learning_rate": 2.2823013405081507e-05,
-      "loss": 0.0027035027742385864,
+      "loss": 9.15761775104329e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00271,
+      "ppl": 1.00009,
       "step": 393,
       "tokens/total": 11891904,
-      "tokens/train_per_sec_per_gpu": 32.73,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 179890
     },
     {
       "epoch": 1.5390625,
-      "grad_norm": 0.0032207504846155643,
+      "grad_norm": 0.03921438008546829,
       "learning_rate": 2.2620747959533722e-05,
-      "loss": 3.968120290664956e-05,
+      "loss": 0.00038214243249967694,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00004,
+      "ppl": 1.00038,
       "step": 394,
       "tokens/total": 11922208,
-      "tokens/train_per_sec_per_gpu": 33.89,
+      "tokens/train_per_sec_per_gpu": 33.95,
       "tokens/trainable": 180341
     },
     {
       "epoch": 1.54296875,
-      "grad_norm": 0.013098486699163914,
+      "grad_norm": 0.08640608936548233,
       "learning_rate": 2.2419829946830123e-05,
-      "loss": 0.00011063168494729325,
+      "loss": 0.0006777399685233831,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00011,
+      "ppl": 1.00068,
       "step": 395,
       "tokens/total": 11952672,
-      "tokens/train_per_sec_per_gpu": 32.95,
+      "tokens/train_per_sec_per_gpu": 32.99,
       "tokens/trainable": 180762
     },
     {
       "epoch": 1.546875,
-      "grad_norm": 0.01553372759371996,
+      "grad_norm": 0.08661718666553497,
       "learning_rate": 2.2220267727989325e-05,
-      "loss": 8.802580123301595e-05,
+      "loss": 0.0007016840972937644,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00009,
+      "ppl": 1.0007,
       "step": 396,
       "tokens/total": 11983088,
-      "tokens/train_per_sec_per_gpu": 35.04,
+      "tokens/train_per_sec_per_gpu": 35.18,
       "tokens/trainable": 181226
     },
     {
       "epoch": 1.55078125,
-      "grad_norm": 0.016083814203739166,
+      "grad_norm": 0.010069145821034908,
       "learning_rate": 2.202206960760984e-05,
-      "loss": 8.931377669796348e-05,
+      "loss": 0.00011120665294583887,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00009,
+      "ppl": 1.00011,
       "step": 397,
       "tokens/total": 12013488,
-      "tokens/train_per_sec_per_gpu": 33.15,
+      "tokens/train_per_sec_per_gpu": 33.18,
       "tokens/trainable": 181714
     },
     {
       "epoch": 1.5546875,
-      "grad_norm": 0.05752059072256088,
+      "grad_norm": 0.03158818930387497,
       "learning_rate": 2.182524383352446e-05,
-      "loss": 0.00015369296306744218,
+      "loss": 0.00024857826065272093,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00015,
+      "ppl": 1.00025,
       "step": 398,
       "tokens/total": 12043968,
-      "tokens/train_per_sec_per_gpu": 35.52,
+      "tokens/train_per_sec_per_gpu": 35.57,
       "tokens/trainable": 182142
     },
     {
       "epoch": 1.55859375,
-      "grad_norm": 0.0011741623748093843,
+      "grad_norm": 0.000873154669534415,
       "learning_rate": 2.1629798596457056e-05,
-      "loss": 1.8113165424438193e-05,
+      "loss": 2.1218824258539826e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00002,
       "step": 399,
       "tokens/total": 12074160,
-      "tokens/train_per_sec_per_gpu": 34.82,
+      "tokens/train_per_sec_per_gpu": 34.86,
       "tokens/trainable": 182604
     },
     {
       "epoch": 1.5625,
-      "grad_norm": 0.0048082731664180756,
+      "grad_norm": 0.009255572222173214,
       "learning_rate": 2.1435742029681725e-05,
-      "loss": 5.3348740038927644e-05,
+      "loss": 0.0001349164522252977,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00005,
+      "ppl": 1.00013,
       "step": 400,
       "tokens/total": 12104352,
-      "tokens/train_per_sec_per_gpu": 25.28,
+      "tokens/train_per_sec_per_gpu": 25.31,
       "tokens/trainable": 183011
     },
     {
       "epoch": 1.56640625,
-      "grad_norm": 0.006520449183881283,
+      "grad_norm": 0.010521743446588516,
       "learning_rate": 2.124308220868431e-05,
-      "loss": 6.45049221930094e-05,
+      "loss": 7.793466647854075e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.71,
       "memory/max_allocated (GiB)": 33.71,
-      "ppl": 1.00006,
+      "ppl": 1.00008,
       "step": 401,
       "tokens/total": 12134240,
-      "tokens/train_per_sec_per_gpu": 31.69,
+      "tokens/train_per_sec_per_gpu": 31.81,
       "tokens/trainable": 183459
     },
     {
       "epoch": 1.5703125,
-      "grad_norm": 0.006227654870599508,
+      "grad_norm": 0.0013409090461209416,
       "learning_rate": 2.105182715082638e-05,
-      "loss": 6.928759830771014e-05,
+      "loss": 2.7722922823159024e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 402,
       "tokens/total": 12164480,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.7,
       "tokens/trainable": 183959
     },
     {
       "epoch": 1.57421875,
-      "grad_norm": 0.32605040073394775,
+      "grad_norm": 0.10361713916063309,
       "learning_rate": 2.0861984815011552e-05,
-      "loss": 0.002115404698997736,
+      "loss": 0.0014142843428999186,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00212,
+      "ppl": 1.00142,
       "step": 403,
       "tokens/total": 12194640,
-      "tokens/train_per_sec_per_gpu": 32.46,
+      "tokens/train_per_sec_per_gpu": 32.5,
       "tokens/trainable": 184431
     },
     {
       "epoch": 1.578125,
-      "grad_norm": 0.007742208894342184,
+      "grad_norm": 0.0026048943400382996,
       "learning_rate": 2.0673563101354323e-05,
-      "loss": 4.600908869178966e-05,
+      "loss": 4.14503738284111e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 404,
       "tokens/total": 12224960,
-      "tokens/train_per_sec_per_gpu": 31.34,
+      "tokens/train_per_sec_per_gpu": 31.37,
       "tokens/trainable": 184867
     },
     {
       "epoch": 1.58203125,
-      "grad_norm": 0.0049946485087275505,
+      "grad_norm": 0.009411919862031937,
       "learning_rate": 2.0486569850851317e-05,
-      "loss": 5.003535625291988e-05,
+      "loss": 7.322158489841968e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.72,
       "memory/max_allocated (GiB)": 33.72,
-      "ppl": 1.00005,
+      "ppl": 1.00007,
       "step": 405,
       "tokens/total": 12255008,
-      "tokens/train_per_sec_per_gpu": 29.25,
+      "tokens/train_per_sec_per_gpu": 29.37,
       "tokens/trainable": 185300
     },
     {
       "epoch": 1.5859375,
-      "grad_norm": 0.004598768427968025,
+      "grad_norm": 0.006792420521378517,
       "learning_rate": 2.0301012845054956e-05,
-      "loss": 3.4423381293891e-05,
+      "loss": 7.982828537933528e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00003,
+      "ppl": 1.00008,
       "step": 406,
       "tokens/total": 12285248,
-      "tokens/train_per_sec_per_gpu": 31.45,
+      "tokens/train_per_sec_per_gpu": 31.48,
       "tokens/trainable": 185754
     },
     {
       "epoch": 1.58984375,
-      "grad_norm": 0.0011464629787951708,
+      "grad_norm": 0.013020367361605167,
       "learning_rate": 2.011689980574966e-05,
-      "loss": 1.9196500943508e-05,
+      "loss": 3.743396155186929e-05,
       "memory/device_reserved (GiB)": 35.88,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 407,
       "tokens/total": 12315552,
-      "tokens/train_per_sec_per_gpu": 36.8,
+      "tokens/train_per_sec_per_gpu": 36.82,
       "tokens/trainable": 186213
     },
     {
       "epoch": 1.59375,
-      "grad_norm": 0.004688040353357792,
+      "grad_norm": 0.0013089004205539823,
       "learning_rate": 1.993423839463052e-05,
-      "loss": 5.2492636314127594e-05,
+      "loss": 2.306591341039166e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00005,
+      "ppl": 1.00002,
       "step": 408,
       "tokens/total": 12345904,
-      "tokens/train_per_sec_per_gpu": 32.72,
+      "tokens/train_per_sec_per_gpu": 32.8,
       "tokens/trainable": 186659
     },
     {
       "epoch": 1.59765625,
-      "grad_norm": 0.002637861529365182,
+      "grad_norm": 0.0012270932784304023,
       "learning_rate": 1.975303621298445e-05,
-      "loss": 3.172009019181132e-05,
+      "loss": 2.1350417227949947e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 409,
       "tokens/total": 12376336,
-      "tokens/train_per_sec_per_gpu": 32.92,
+      "tokens/train_per_sec_per_gpu": 32.97,
       "tokens/trainable": 187083
     },
     {
       "epoch": 1.6015625,
-      "grad_norm": 0.0072015393525362015,
+      "grad_norm": 0.011230082251131535,
       "learning_rate": 1.957330080137385e-05,
-      "loss": 5.452537880046293e-05,
+      "loss": 3.752015618374571e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 410,
       "tokens/total": 12406880,
-      "tokens/train_per_sec_per_gpu": 34.91,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 187530
     },
     {
       "epoch": 1.60546875,
-      "grad_norm": 0.007169651333242655,
+      "grad_norm": 0.004872993566095829,
       "learning_rate": 1.9395039639322864e-05,
-      "loss": 3.530656249495223e-05,
+      "loss": 5.5653974413871765e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00006,
       "step": 411,
       "tokens/total": 12437088,
-      "tokens/train_per_sec_per_gpu": 34.28,
+      "tokens/train_per_sec_per_gpu": 34.32,
       "tokens/trainable": 187972
     },
     {
       "epoch": 1.609375,
-      "grad_norm": 0.003889538114890456,
+      "grad_norm": 0.19391997158527374,
       "learning_rate": 1.9218260145006073e-05,
-      "loss": 3.5087461583316326e-05,
+      "loss": 0.000892186420969665,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00004,
+      "ppl": 1.00089,
       "step": 412,
       "tokens/total": 12465440,
-      "tokens/train_per_sec_per_gpu": 39.89,
+      "tokens/train_per_sec_per_gpu": 39.95,
       "tokens/trainable": 188417
     },
     {
       "epoch": 1.61328125,
-      "grad_norm": 0.5135430693626404,
+      "grad_norm": 0.09062197804450989,
       "learning_rate": 1.904296967493982e-05,
-      "loss": 0.005697700660675764,
+      "loss": 0.0006304415874183178,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00571,
+      "ppl": 1.00063,
       "step": 413,
       "tokens/total": 12495968,
-      "tokens/train_per_sec_per_gpu": 31.48,
+      "tokens/train_per_sec_per_gpu": 31.57,
       "tokens/trainable": 188868
     },
     {
       "epoch": 1.6171875,
-      "grad_norm": 0.0010096468031406403,
+      "grad_norm": 0.009724686853587627,
       "learning_rate": 1.8869175523676064e-05,
-      "loss": 1.1318426913931035e-05,
+      "loss": 6.72380265314132e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 414,
       "tokens/total": 12526128,
-      "tokens/train_per_sec_per_gpu": 36.85,
+      "tokens/train_per_sec_per_gpu": 36.95,
       "tokens/trainable": 189337
     },
     {
       "epoch": 1.62109375,
-      "grad_norm": 0.0004970752634108067,
+      "grad_norm": 0.19542834162712097,
       "learning_rate": 1.869688492349885e-05,
-      "loss": 7.355167326750234e-06,
+      "loss": 0.001907092402689159,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00191,
       "step": 415,
       "tokens/total": 12556256,
-      "tokens/train_per_sec_per_gpu": 31.82,
+      "tokens/train_per_sec_per_gpu": 31.94,
       "tokens/trainable": 189750
     },
     {
       "epoch": 1.625,
-      "grad_norm": 0.0008849130244925618,
+      "grad_norm": 0.0465276800096035,
       "learning_rate": 1.85261050441233e-05,
-      "loss": 1.4194254617905244e-05,
+      "loss": 0.0005661146715283394,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00001,
+      "ppl": 1.00057,
       "step": 416,
       "tokens/total": 12586736,
-      "tokens/train_per_sec_per_gpu": 33.97,
+      "tokens/train_per_sec_per_gpu": 34.06,
       "tokens/trainable": 190216
     },
     {
       "epoch": 1.62890625,
-      "grad_norm": 0.0022039280738681555,
+      "grad_norm": 0.0012954205740243196,
       "learning_rate": 1.8356842992397304e-05,
-      "loss": 2.4539594960515387e-05,
+      "loss": 1.8871982319978997e-05,
       "memory/device_reserved (GiB)": 35.96,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
       "ppl": 1.00002,
       "step": 417,
       "tokens/total": 12617168,
-      "tokens/train_per_sec_per_gpu": 34.95,
+      "tokens/train_per_sec_per_gpu": 35.01,
       "tokens/trainable": 190708
     },
     {
       "epoch": 1.6328125,
-      "grad_norm": 0.027006179094314575,
+      "grad_norm": 0.0018749319715425372,
       "learning_rate": 1.8189105812005714e-05,
-      "loss": 0.00020442574168555439,
+      "loss": 3.220669532311149e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.0002,
+      "ppl": 1.00003,
       "step": 418,
       "tokens/total": 12647680,
-      "tokens/train_per_sec_per_gpu": 30.85,
+      "tokens/train_per_sec_per_gpu": 30.9,
       "tokens/trainable": 191126
     },
     {
       "epoch": 1.63671875,
-      "grad_norm": 0.15684254467487335,
+      "grad_norm": 0.0022480092011392117,
       "learning_rate": 1.802290048317732e-05,
-      "loss": 0.0007251293282024562,
+      "loss": 2.6564141080598347e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00073,
+      "ppl": 1.00003,
       "step": 419,
       "tokens/total": 12677952,
-      "tokens/train_per_sec_per_gpu": 30.3,
+      "tokens/train_per_sec_per_gpu": 30.34,
       "tokens/trainable": 191540
     },
     {
       "epoch": 1.640625,
-      "grad_norm": 0.0033104538451880217,
+      "grad_norm": 0.1634059101343155,
       "learning_rate": 1.785823392239424e-05,
-      "loss": 3.131272751488723e-05,
+      "loss": 0.0009822181891649961,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00003,
+      "ppl": 1.00098,
       "step": 420,
       "tokens/total": 12708416,
-      "tokens/train_per_sec_per_gpu": 36.65,
+      "tokens/train_per_sec_per_gpu": 36.69,
       "tokens/trainable": 192014
     },
     {
       "epoch": 1.64453125,
-      "grad_norm": 0.001255685230717063,
+      "grad_norm": 0.0019404747290536761,
       "learning_rate": 1.7695112982104225e-05,
-      "loss": 1.9309976778458804e-05,
+      "loss": 2.167341335734818e-05,
       "memory/device_reserved (GiB)": 35.8,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
       "ppl": 1.00002,
       "step": 421,
       "tokens/total": 12738640,
-      "tokens/train_per_sec_per_gpu": 31.28,
+      "tokens/train_per_sec_per_gpu": 31.31,
       "tokens/trainable": 192463
     },
     {
       "epoch": 1.6484375,
-      "grad_norm": 0.01599641516804695,
+      "grad_norm": 0.001059795613400638,
       "learning_rate": 1.7533544450435433e-05,
-      "loss": 9.235648030880839e-05,
+      "loss": 1.6543437595828436e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00009,
+      "ppl": 1.00002,
       "step": 422,
       "tokens/total": 12769232,
-      "tokens/train_per_sec_per_gpu": 32.71,
+      "tokens/train_per_sec_per_gpu": 32.74,
       "tokens/trainable": 192918
     },
     {
       "epoch": 1.65234375,
-      "grad_norm": 0.048507146537303925,
+      "grad_norm": 0.01760418340563774,
       "learning_rate": 1.7373535050913946e-05,
-      "loss": 0.0002708940301090479,
+      "loss": 0.0001707405026536435,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00027,
+      "ppl": 1.00017,
       "step": 423,
       "tokens/total": 12799648,
-      "tokens/train_per_sec_per_gpu": 37.35,
+      "tokens/train_per_sec_per_gpu": 37.46,
       "tokens/trainable": 193425
     },
     {
       "epoch": 1.65625,
-      "grad_norm": 0.010107563808560371,
+      "grad_norm": 0.0023663989268243313,
       "learning_rate": 1.721509144218405e-05,
-      "loss": 6.99054216966033e-05,
+      "loss": 2.332203439436853e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00007,
+      "ppl": 1.00002,
       "step": 424,
       "tokens/total": 12829920,
-      "tokens/train_per_sec_per_gpu": 35.23,
+      "tokens/train_per_sec_per_gpu": 35.25,
       "tokens/trainable": 193898
     },
     {
       "epoch": 1.66015625,
-      "grad_norm": 0.005467226263135672,
+      "grad_norm": 0.0245437640696764,
       "learning_rate": 1.705822021773101e-05,
-      "loss": 5.417566717369482e-05,
+      "loss": 0.0001577093207743019,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00005,
+      "ppl": 1.00016,
       "step": 425,
       "tokens/total": 12860112,
-      "tokens/train_per_sec_per_gpu": 29.24,
+      "tokens/train_per_sec_per_gpu": 29.29,
       "tokens/trainable": 194315
     },
     {
       "epoch": 1.6640625,
-      "grad_norm": 0.0015799012035131454,
+      "grad_norm": 0.0009002169244922698,
       "learning_rate": 1.69029279056068e-05,
-      "loss": 2.4984849005704746e-05,
+      "loss": 1.4283305972639937e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 426,
       "tokens/total": 12890320,
       "tokens/train_per_sec_per_gpu": 34.92,
@@ -5975,779 +5975,779 @@
     },
     {
       "epoch": 1.66796875,
-      "grad_norm": 0.024157166481018066,
+      "grad_norm": 0.00033997284481301904,
       "learning_rate": 1.6749220968158415e-05,
-      "loss": 0.00010696032404666767,
+      "loss": 7.185776667029131e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00001,
       "step": 427,
       "tokens/total": 12920656,
-      "tokens/train_per_sec_per_gpu": 33.92,
+      "tokens/train_per_sec_per_gpu": 33.96,
       "tokens/trainable": 195262
     },
     {
       "epoch": 1.671875,
-      "grad_norm": 0.00028090659179724753,
+      "grad_norm": 0.0005359657225199044,
       "learning_rate": 1.659710580175893e-05,
-      "loss": 7.484430170734413e-06,
+      "loss": 9.948088518285658e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
       "ppl": 1.00001,
       "step": 428,
       "tokens/total": 12951040,
-      "tokens/train_per_sec_per_gpu": 31.83,
+      "tokens/train_per_sec_per_gpu": 31.91,
       "tokens/trainable": 195680
     },
     {
       "epoch": 1.67578125,
-      "grad_norm": 0.0011094068177044392,
+      "grad_norm": 0.0014890795573592186,
       "learning_rate": 1.644658873654133e-05,
-      "loss": 1.9920153135899454e-05,
+      "loss": 2.028812377830036e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.75,
       "memory/max_allocated (GiB)": 33.75,
       "ppl": 1.00002,
       "step": 429,
       "tokens/total": 12981232,
-      "tokens/train_per_sec_per_gpu": 32.83,
+      "tokens/train_per_sec_per_gpu": 32.89,
       "tokens/trainable": 196145
     },
     {
       "epoch": 1.6796875,
-      "grad_norm": 0.020794112235307693,
+      "grad_norm": 0.05130980163812637,
       "learning_rate": 1.629767603613508e-05,
-      "loss": 0.00010971157462336123,
+      "loss": 0.0003292672336101532,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00011,
+      "ppl": 1.00033,
       "step": 430,
       "tokens/total": 13011616,
-      "tokens/train_per_sec_per_gpu": 32.39,
+      "tokens/train_per_sec_per_gpu": 32.42,
       "tokens/trainable": 196626
     },
     {
       "epoch": 1.68359375,
-      "grad_norm": 0.44080695509910583,
+      "grad_norm": 0.0014070137403905392,
       "learning_rate": 1.615037389740547e-05,
-      "loss": 0.0026788683608174324,
+      "loss": 1.0152909453609027e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.00268,
+      "ppl": 1.00001,
       "step": 431,
       "tokens/total": 13042208,
-      "tokens/train_per_sec_per_gpu": 31.14,
+      "tokens/train_per_sec_per_gpu": 31.17,
       "tokens/trainable": 197067
     },
     {
       "epoch": 1.6875,
-      "grad_norm": 0.0026339287869632244,
+      "grad_norm": 0.004116157069802284,
       "learning_rate": 1.600468845019576e-05,
-      "loss": 2.082335777231492e-05,
+      "loss": 3.514952550176531e-05,
       "memory/device_reserved (GiB)": 36.35,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00002,
+      "ppl": 1.00004,
       "step": 432,
       "tokens/total": 13072800,
-      "tokens/train_per_sec_per_gpu": 38.61,
+      "tokens/train_per_sec_per_gpu": 38.68,
       "tokens/trainable": 197565
     },
     {
       "epoch": 1.69140625,
-      "grad_norm": 0.004508621525019407,
+      "grad_norm": 0.01147819496691227,
       "learning_rate": 1.5860625757072092e-05,
-      "loss": 1.439991501683835e-05,
+      "loss": 7.313965761568397e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.95,
       "memory/max_allocated (GiB)": 33.95,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 433,
       "tokens/total": 13103328,
-      "tokens/train_per_sec_per_gpu": 32.88,
+      "tokens/train_per_sec_per_gpu": 32.96,
       "tokens/trainable": 198015
     },
     {
       "epoch": 1.6953125,
-      "grad_norm": 0.014811511151492596,
+      "grad_norm": 0.030784178525209427,
       "learning_rate": 1.571819181307116e-05,
-      "loss": 0.00015316363715101033,
+      "loss": 0.0002448821614962071,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00015,
+      "ppl": 1.00024,
       "step": 434,
       "tokens/total": 13133472,
-      "tokens/train_per_sec_per_gpu": 33.87,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 198442
     },
     {
       "epoch": 1.69921875,
-      "grad_norm": 0.0037793368101119995,
+      "grad_norm": 0.0022457120940089226,
       "learning_rate": 1.557739254545075e-05,
-      "loss": 1.4444960470427759e-05,
+      "loss": 1.4446297427639365e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
       "ppl": 1.00001,
       "step": 435,
       "tokens/total": 13164064,
-      "tokens/train_per_sec_per_gpu": 34.3,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 198914
     },
     {
       "epoch": 1.703125,
-      "grad_norm": 0.24972176551818848,
+      "grad_norm": 0.0029742212500423193,
       "learning_rate": 1.543823381344311e-05,
-      "loss": 0.0017114672809839249,
+      "loss": 2.7415488148108125e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00171,
+      "ppl": 1.00003,
       "step": 436,
       "tokens/total": 13194464,
-      "tokens/train_per_sec_per_gpu": 33.35,
+      "tokens/train_per_sec_per_gpu": 33.45,
       "tokens/trainable": 199355
     },
     {
       "epoch": 1.70703125,
-      "grad_norm": 0.002479988383129239,
+      "grad_norm": 0.002260624896734953,
       "learning_rate": 1.5300721408011114e-05,
-      "loss": 2.9400333005469292e-05,
+      "loss": 2.7733602109947242e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00003,
       "step": 437,
       "tokens/total": 13224992,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 33.0,
       "tokens/trainable": 199830
     },
     {
       "epoch": 1.7109375,
-      "grad_norm": 0.0014780610799789429,
+      "grad_norm": 0.01198511105030775,
       "learning_rate": 1.5164861051607254e-05,
-      "loss": 1.5593774151057005e-05,
+      "loss": 9.154126746580005e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00002,
+      "ppl": 1.00009,
       "step": 438,
       "tokens/total": 13255296,
-      "tokens/train_per_sec_per_gpu": 34.89,
+      "tokens/train_per_sec_per_gpu": 34.97,
       "tokens/trainable": 200338
     },
     {
       "epoch": 1.71484375,
-      "grad_norm": 0.0010410621762275696,
+      "grad_norm": 0.003149093361571431,
       "learning_rate": 1.5030658397935521e-05,
-      "loss": 1.4683226254419424e-05,
+      "loss": 3.515004937071353e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00001,
+      "ppl": 1.00004,
       "step": 439,
       "tokens/total": 13285344,
-      "tokens/train_per_sec_per_gpu": 32.79,
+      "tokens/train_per_sec_per_gpu": 32.87,
       "tokens/trainable": 200785
     },
     {
       "epoch": 1.71875,
-      "grad_norm": 0.0004247442411724478,
+      "grad_norm": 0.0003243185637984425,
       "learning_rate": 1.4898119031716104e-05,
-      "loss": 8.09474295238033e-06,
+      "loss": 7.168858701334102e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
       "ppl": 1.00001,
       "step": 440,
       "tokens/total": 13315632,
-      "tokens/train_per_sec_per_gpu": 36.28,
+      "tokens/train_per_sec_per_gpu": 36.32,
       "tokens/trainable": 201267
     },
     {
       "epoch": 1.72265625,
-      "grad_norm": 0.0026301892939954996,
+      "grad_norm": 0.12771126627922058,
       "learning_rate": 1.476724846845306e-05,
-      "loss": 1.7428235878469422e-05,
+      "loss": 0.0007142049144022167,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00002,
+      "ppl": 1.00071,
       "step": 441,
       "tokens/total": 13346048,
-      "tokens/train_per_sec_per_gpu": 34.16,
+      "tokens/train_per_sec_per_gpu": 34.22,
       "tokens/trainable": 201734
     },
     {
       "epoch": 1.7265625,
-      "grad_norm": 0.0005576284602284431,
+      "grad_norm": 0.11483822762966156,
       "learning_rate": 1.463805215420471e-05,
-      "loss": 8.656044883537106e-06,
+      "loss": 0.0006703532999381423,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00001,
+      "ppl": 1.00067,
       "step": 442,
       "tokens/total": 13376304,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.4,
       "tokens/trainable": 202174
     },
     {
       "epoch": 1.73046875,
-      "grad_norm": 0.0009604030638001859,
+      "grad_norm": 0.00032006221590563655,
       "learning_rate": 1.451053546535705e-05,
-      "loss": 1.4092523088038433e-05,
+      "loss": 9.367744496557862e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
       "ppl": 1.00001,
       "step": 443,
       "tokens/total": 13406720,
-      "tokens/train_per_sec_per_gpu": 36.13,
+      "tokens/train_per_sec_per_gpu": 36.28,
       "tokens/trainable": 202612
     },
     {
       "epoch": 1.734375,
-      "grad_norm": 0.00019884438370354474,
+      "grad_norm": 0.0002659875026438385,
       "learning_rate": 1.438470370840001e-05,
-      "loss": 4.5837323341402225e-06,
+      "loss": 5.530210728466045e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.64,
       "memory/max_allocated (GiB)": 33.64,
-      "ppl": 1.0,
+      "ppl": 1.00001,
       "step": 444,
       "tokens/total": 13436464,
-      "tokens/train_per_sec_per_gpu": 30.96,
+      "tokens/train_per_sec_per_gpu": 30.99,
       "tokens/trainable": 203020
     },
     {
       "epoch": 1.73828125,
-      "grad_norm": 0.05944995582103729,
+      "grad_norm": 0.04957196116447449,
       "learning_rate": 1.4260562119706606e-05,
-      "loss": 0.0002780454815365374,
+      "loss": 6.46735934424214e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00028,
+      "ppl": 1.00006,
       "step": 445,
       "tokens/total": 13466672,
-      "tokens/train_per_sec_per_gpu": 34.31,
+      "tokens/train_per_sec_per_gpu": 34.37,
       "tokens/trainable": 203479
     },
     {
       "epoch": 1.7421875,
-      "grad_norm": 0.005872223526239395,
+      "grad_norm": 0.0016840791795402765,
       "learning_rate": 1.413811586531508e-05,
-      "loss": 3.424299575272016e-05,
+      "loss": 1.2613029866770376e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 446,
       "tokens/total": 13496736,
-      "tokens/train_per_sec_per_gpu": 35.3,
+      "tokens/train_per_sec_per_gpu": 35.38,
       "tokens/trainable": 203946
     },
     {
       "epoch": 1.74609375,
-      "grad_norm": 0.006013527978211641,
+      "grad_norm": 0.00042921333806589246,
       "learning_rate": 1.4017370040713884e-05,
-      "loss": 3.269856097176671e-05,
+      "loss": 7.416386324621271e-06,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 447,
       "tokens/total": 13526928,
-      "tokens/train_per_sec_per_gpu": 37.43,
+      "tokens/train_per_sec_per_gpu": 37.45,
       "tokens/trainable": 204382
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.05103042349219322,
+      "grad_norm": 0.2165915071964264,
       "learning_rate": 1.3898329670629645e-05,
-      "loss": 0.00032247230410575867,
+      "loss": 0.0006914341356605291,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00032,
+      "ppl": 1.00069,
       "step": 448,
       "tokens/total": 13557392,
-      "tokens/train_per_sec_per_gpu": 34.32,
+      "tokens/train_per_sec_per_gpu": 34.35,
       "tokens/trainable": 204830
     },
     {
       "epoch": 1.75390625,
-      "grad_norm": 0.13250835239887238,
+      "grad_norm": 0.0009792562341317534,
       "learning_rate": 1.3780999708818058e-05,
-      "loss": 0.0008940898114815354,
+      "loss": 1.4504414139082655e-05,
       "memory/device_reserved (GiB)": 36.8,
       "memory/max_active (GiB)": 33.93,
       "memory/max_allocated (GiB)": 33.93,
-      "ppl": 1.00089,
+      "ppl": 1.00001,
       "step": 449,
       "tokens/total": 13587776,
-      "tokens/train_per_sec_per_gpu": 33.51,
+      "tokens/train_per_sec_per_gpu": 33.47,
       "tokens/trainable": 205300
     },
     {
       "epoch": 1.7578125,
-      "grad_norm": 0.24642030894756317,
+      "grad_norm": 0.028498223051428795,
       "learning_rate": 1.3665385037857758e-05,
-      "loss": 0.0012931979726999998,
+      "loss": 0.00028163049137219787,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00129,
+      "ppl": 1.00028,
       "step": 450,
       "tokens/total": 13618112,
-      "tokens/train_per_sec_per_gpu": 30.62,
+      "tokens/train_per_sec_per_gpu": 30.6,
       "tokens/trainable": 205726
     },
     {
       "epoch": 1.76171875,
-      "grad_norm": 0.011528799310326576,
+      "grad_norm": 0.003279345342889428,
       "learning_rate": 1.3551490468947126e-05,
-      "loss": 6.569598917849362e-05,
+      "loss": 2.595262776594609e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00007,
+      "ppl": 1.00003,
       "step": 451,
       "tokens/total": 13648512,
-      "tokens/train_per_sec_per_gpu": 31.55,
+      "tokens/train_per_sec_per_gpu": 31.63,
       "tokens/trainable": 206163
     },
     {
       "epoch": 1.765625,
-      "grad_norm": 1.0818549394607544,
+      "grad_norm": 0.4961508512496948,
       "learning_rate": 1.3439320741704075e-05,
-      "loss": 0.015596212819218636,
+      "loss": 0.00459528062492609,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.01572,
+      "ppl": 1.00461,
       "step": 452,
       "tokens/total": 13678704,
-      "tokens/train_per_sec_per_gpu": 35.41,
+      "tokens/train_per_sec_per_gpu": 35.45,
       "tokens/trainable": 206619
     },
     {
       "epoch": 1.76953125,
-      "grad_norm": 0.0007326967315748334,
+      "grad_norm": 0.009086468257009983,
       "learning_rate": 1.3328880523968808e-05,
-      "loss": 1.2033770872221794e-05,
+      "loss": 6.577694148290902e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 453,
       "tokens/total": 13709232,
-      "tokens/train_per_sec_per_gpu": 37.18,
+      "tokens/train_per_sec_per_gpu": 37.24,
       "tokens/trainable": 207101
     },
     {
       "epoch": 1.7734375,
-      "grad_norm": 0.0023919539526104927,
+      "grad_norm": 0.009036197327077389,
       "learning_rate": 1.3220174411609587e-05,
-      "loss": 1.262133719137637e-05,
+      "loss": 6.735372880939394e-05,
       "memory/device_reserved (GiB)": 35.7,
       "memory/max_active (GiB)": 33.86,
       "memory/max_allocated (GiB)": 33.86,
-      "ppl": 1.00001,
+      "ppl": 1.00007,
       "step": 454,
       "tokens/total": 13739600,
-      "tokens/train_per_sec_per_gpu": 35.69,
+      "tokens/train_per_sec_per_gpu": 35.77,
       "tokens/trainable": 207546
     },
     {
       "epoch": 1.77734375,
-      "grad_norm": 0.0020972786005586386,
+      "grad_norm": 0.009665202349424362,
       "learning_rate": 1.3113206928331471e-05,
-      "loss": 2.476977533660829e-05,
+      "loss": 7.266137981787324e-05,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00007,
       "step": 455,
       "tokens/total": 13769936,
-      "tokens/train_per_sec_per_gpu": 34.25,
+      "tokens/train_per_sec_per_gpu": 34.26,
       "tokens/trainable": 207989
     },
     {
       "epoch": 1.78125,
-      "grad_norm": 0.000797569751739502,
+      "grad_norm": 0.0004666070453822613,
       "learning_rate": 1.300798252548806e-05,
-      "loss": 1.45716385304695e-05,
+      "loss": 6.039275831426494e-06,
       "memory/device_reserved (GiB)": 35.78,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
       "ppl": 1.00001,
       "step": 456,
       "tokens/total": 13800240,
-      "tokens/train_per_sec_per_gpu": 33.9,
+      "tokens/train_per_sec_per_gpu": 33.92,
       "tokens/trainable": 208402
     },
     {
       "epoch": 1.78515625,
-      "grad_norm": 0.011376727372407913,
+      "grad_norm": 0.00044853391591459513,
       "learning_rate": 1.2904505581896265e-05,
-      "loss": 7.903270306997001e-05,
+      "loss": 8.292890015582088e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00008,
+      "ppl": 1.00001,
       "step": 457,
       "tokens/total": 13830512,
-      "tokens/train_per_sec_per_gpu": 36.61,
+      "tokens/train_per_sec_per_gpu": 36.58,
       "tokens/trainable": 208904
     },
     {
       "epoch": 1.7890625,
-      "grad_norm": 0.0013566205743700266,
+      "grad_norm": 0.0033692270517349243,
       "learning_rate": 1.2802780403654082e-05,
-      "loss": 1.925312972161919e-05,
+      "loss": 2.3157112082117237e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
       "ppl": 1.00002,
       "step": 458,
       "tokens/total": 13860832,
-      "tokens/train_per_sec_per_gpu": 33.64,
+      "tokens/train_per_sec_per_gpu": 33.72,
       "tokens/trainable": 209382
     },
     {
       "epoch": 1.79296875,
-      "grad_norm": 0.00047398527385666966,
+      "grad_norm": 0.0010797431459650397,
       "learning_rate": 1.2702811223961408e-05,
-      "loss": 1.119351145462133e-05,
+      "loss": 1.119279841077514e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
       "ppl": 1.00001,
       "step": 459,
       "tokens/total": 13890992,
-      "tokens/train_per_sec_per_gpu": 33.85,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 209833
     },
     {
       "epoch": 1.796875,
-      "grad_norm": 0.04677487164735794,
+      "grad_norm": 0.04658913239836693,
       "learning_rate": 1.2604602202943861e-05,
-      "loss": 0.0001750149531289935,
+      "loss": 0.0003375259111635387,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.91,
       "memory/max_allocated (GiB)": 33.91,
-      "ppl": 1.00018,
+      "ppl": 1.00034,
       "step": 460,
       "tokens/total": 13921424,
-      "tokens/train_per_sec_per_gpu": 32.5,
+      "tokens/train_per_sec_per_gpu": 32.53,
       "tokens/trainable": 210284
     },
     {
       "epoch": 1.80078125,
-      "grad_norm": 0.024663345888257027,
+      "grad_norm": 0.01848040148615837,
       "learning_rate": 1.2508157427479686e-05,
-      "loss": 9.222347580362111e-05,
+      "loss": 0.00016109315038193017,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.66,
       "memory/max_allocated (GiB)": 33.66,
-      "ppl": 1.00009,
+      "ppl": 1.00016,
       "step": 461,
       "tokens/total": 13951504,
-      "tokens/train_per_sec_per_gpu": 34.04,
+      "tokens/train_per_sec_per_gpu": 34.07,
       "tokens/trainable": 210769
     },
     {
       "epoch": 1.8046875,
-      "grad_norm": 0.0004740248841699213,
+      "grad_norm": 0.05658251419663429,
       "learning_rate": 1.2413480911029655e-05,
-      "loss": 1.0696679964894429e-05,
+      "loss": 0.00017825173563323915,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.38,
       "memory/max_allocated (GiB)": 33.38,
-      "ppl": 1.00001,
+      "ppl": 1.00018,
       "step": 462,
       "tokens/total": 13979728,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.41,
       "tokens/trainable": 211193
     },
     {
       "epoch": 1.80859375,
-      "grad_norm": 0.0007590787718072534,
+      "grad_norm": 0.0006594725418835878,
       "learning_rate": 1.2320576593470082e-05,
-      "loss": 1.3936740288045257e-05,
+      "loss": 9.709075129649136e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.79,
       "memory/max_allocated (GiB)": 33.79,
       "ppl": 1.00001,
       "step": 463,
       "tokens/total": 14009936,
-      "tokens/train_per_sec_per_gpu": 35.03,
+      "tokens/train_per_sec_per_gpu": 35.11,
       "tokens/trainable": 211655
     },
     {
       "epoch": 1.8125,
-      "grad_norm": 0.002108624204993248,
+      "grad_norm": 0.02737571857869625,
       "learning_rate": 1.2229448340928828e-05,
-      "loss": 2.675112227734644e-05,
+      "loss": 0.00017574361118022352,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.74,
       "memory/max_allocated (GiB)": 33.74,
-      "ppl": 1.00003,
+      "ppl": 1.00018,
       "step": 464,
       "tokens/total": 14039872,
-      "tokens/train_per_sec_per_gpu": 29.28,
+      "tokens/train_per_sec_per_gpu": 29.2,
       "tokens/trainable": 212085
     },
     {
       "epoch": 1.81640625,
-      "grad_norm": 0.0009435649262741208,
+      "grad_norm": 0.0004917355254292488,
       "learning_rate": 1.2140099945624458e-05,
-      "loss": 1.0821668183780275e-05,
+      "loss": 8.16806459624786e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
       "ppl": 1.00001,
       "step": 465,
       "tokens/total": 14070096,
-      "tokens/train_per_sec_per_gpu": 32.61,
+      "tokens/train_per_sec_per_gpu": 32.71,
       "tokens/trainable": 212570
     },
     {
       "epoch": 1.8203125,
-      "grad_norm": 0.011040126904845238,
+      "grad_norm": 0.0015915038529783487,
       "learning_rate": 1.205253512570841e-05,
-      "loss": 5.798249912913889e-05,
+      "loss": 1.2870759746874683e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00006,
+      "ppl": 1.00001,
       "step": 466,
       "tokens/total": 14100192,
-      "tokens/train_per_sec_per_gpu": 32.86,
+      "tokens/train_per_sec_per_gpu": 32.93,
       "tokens/trainable": 213011
     },
     {
       "epoch": 1.82421875,
-      "grad_norm": 0.008271797560155392,
+      "grad_norm": 0.0005419626249931753,
       "learning_rate": 1.1966757525110255e-05,
-      "loss": 2.8763912268914282e-05,
+      "loss": 7.527931302320212e-06,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 467,
       "tokens/total": 14130448,
-      "tokens/train_per_sec_per_gpu": 31.44,
+      "tokens/train_per_sec_per_gpu": 31.58,
       "tokens/trainable": 213450
     },
     {
       "epoch": 1.828125,
-      "grad_norm": 0.023479726165533066,
+      "grad_norm": 0.004500397015362978,
       "learning_rate": 1.1882770713386095e-05,
-      "loss": 0.0001434558507753536,
+      "loss": 4.2581184970913455e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.73,
       "memory/max_allocated (GiB)": 33.73,
-      "ppl": 1.00014,
+      "ppl": 1.00004,
       "step": 468,
       "tokens/total": 14160480,
-      "tokens/train_per_sec_per_gpu": 35.62,
+      "tokens/train_per_sec_per_gpu": 35.7,
       "tokens/trainable": 213898
     },
     {
       "epoch": 1.83203125,
-      "grad_norm": 0.0023077642545104027,
+      "grad_norm": 0.1103268712759018,
       "learning_rate": 1.180057818556998e-05,
-      "loss": 2.560112625360489e-05,
+      "loss": 0.00021096917043905705,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.00003,
+      "ppl": 1.00021,
       "step": 469,
       "tokens/total": 14191152,
-      "tokens/train_per_sec_per_gpu": 36.78,
+      "tokens/train_per_sec_per_gpu": 36.88,
       "tokens/trainable": 214366
     },
     {
       "epoch": 1.8359375,
-      "grad_norm": 0.0014362182701006532,
+      "grad_norm": 0.002097723074257374,
       "learning_rate": 1.1720183362028494e-05,
-      "loss": 2.0195953766233288e-05,
+      "loss": 2.7896878236788325e-05,
       "memory/device_reserved (GiB)": 35.82,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 470,
       "tokens/total": 14219392,
-      "tokens/train_per_sec_per_gpu": 35.64,
+      "tokens/train_per_sec_per_gpu": 35.72,
       "tokens/trainable": 214800
     },
     {
       "epoch": 1.83984375,
-      "grad_norm": 0.011598445475101471,
+      "grad_norm": 0.001367987017147243,
       "learning_rate": 1.1641589588318387e-05,
-      "loss": 4.486892794375308e-05,
+      "loss": 1.733974931994453e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00004,
+      "ppl": 1.00002,
       "step": 471,
       "tokens/total": 14249920,
-      "tokens/train_per_sec_per_gpu": 35.35,
+      "tokens/train_per_sec_per_gpu": 35.48,
       "tokens/trainable": 215292
     },
     {
       "epoch": 1.84375,
-      "grad_norm": 0.0011633927933871746,
+      "grad_norm": 0.000857445178553462,
       "learning_rate": 1.1564800135047418e-05,
-      "loss": 2.0482253603404388e-05,
+      "loss": 1.3961070180812385e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 472,
       "tokens/total": 14280048,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 33.9,
       "tokens/trainable": 215741
     },
     {
       "epoch": 1.84765625,
-      "grad_norm": 0.006093664560467005,
+      "grad_norm": 0.0037669208832085133,
       "learning_rate": 1.148981819773816e-05,
-      "loss": 5.0086440751329064e-05,
+      "loss": 3.097699664067477e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00005,
+      "ppl": 1.00003,
       "step": 473,
       "tokens/total": 14310432,
-      "tokens/train_per_sec_per_gpu": 34.64,
+      "tokens/train_per_sec_per_gpu": 34.68,
       "tokens/trainable": 216194
     },
     {
       "epoch": 1.8515625,
-      "grad_norm": 0.003623088588938117,
+      "grad_norm": 0.0003345425648149103,
       "learning_rate": 1.1416646896695086e-05,
-      "loss": 3.059494338231161e-05,
+      "loss": 6.7712135205511e-06,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 474,
       "tokens/total": 14340496,
-      "tokens/train_per_sec_per_gpu": 33.3,
+      "tokens/train_per_sec_per_gpu": 33.33,
       "tokens/trainable": 216634
     },
     {
       "epoch": 1.85546875,
-      "grad_norm": 0.006924150045961142,
+      "grad_norm": 0.00038110429886728525,
       "learning_rate": 1.1345289276874717e-05,
-      "loss": 3.953809573431499e-05,
+      "loss": 7.348439794441219e-06,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 475,
       "tokens/total": 14370720,
-      "tokens/train_per_sec_per_gpu": 30.48,
+      "tokens/train_per_sec_per_gpu": 30.53,
       "tokens/trainable": 217040
     },
     {
       "epoch": 1.859375,
-      "grad_norm": 0.002359601203352213,
+      "grad_norm": 0.0013931745197623968,
       "learning_rate": 1.1275748307758873e-05,
-      "loss": 3.34192045556847e-05,
+      "loss": 1.2693172720901202e-05,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 476,
       "tokens/total": 14401136,
-      "tokens/train_per_sec_per_gpu": 37.55,
+      "tokens/train_per_sec_per_gpu": 37.67,
       "tokens/trainable": 217530
     },
     {
       "epoch": 1.86328125,
-      "grad_norm": 0.009584639221429825,
+      "grad_norm": 0.02564929611980915,
       "learning_rate": 1.1208026883231147e-05,
-      "loss": 0.00010110227594850585,
+      "loss": 0.00031186698470264673,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.0001,
+      "ppl": 1.00031,
       "step": 477,
       "tokens/total": 14431360,
-      "tokens/train_per_sec_per_gpu": 34.47,
+      "tokens/train_per_sec_per_gpu": 34.53,
       "tokens/trainable": 217988
     },
     {
       "epoch": 1.8671875,
-      "grad_norm": 0.03044246695935726,
+      "grad_norm": 0.08053945004940033,
       "learning_rate": 1.1142127821456433e-05,
-      "loss": 0.00022086883836891502,
+      "loss": 0.0003999105538241565,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00022,
+      "ppl": 1.0004,
       "step": 478,
       "tokens/total": 14461968,
-      "tokens/train_per_sec_per_gpu": 32.47,
+      "tokens/train_per_sec_per_gpu": 32.54,
       "tokens/trainable": 218461
     },
     {
       "epoch": 1.87109375,
-      "grad_norm": 0.03494428098201752,
+      "grad_norm": 0.05217274650931358,
       "learning_rate": 1.1078053864763674e-05,
-      "loss": 0.0002753190929070115,
+      "loss": 0.0003339847025927156,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.83,
       "memory/max_allocated (GiB)": 33.83,
-      "ppl": 1.00028,
+      "ppl": 1.00033,
       "step": 479,
       "tokens/total": 14491984,
-      "tokens/train_per_sec_per_gpu": 32.77,
+      "tokens/train_per_sec_per_gpu": 32.81,
       "tokens/trainable": 218909
     },
     {
       "epoch": 1.875,
-      "grad_norm": 0.30102044343948364,
+      "grad_norm": 0.08783794194459915,
       "learning_rate": 1.1015807679531756e-05,
-      "loss": 0.00025764034944586456,
+      "loss": 0.0005760936765000224,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.98,
       "memory/max_allocated (GiB)": 33.98,
-      "ppl": 1.00026,
+      "ppl": 1.00058,
       "step": 480,
       "tokens/total": 14522544,
-      "tokens/train_per_sec_per_gpu": 33.88,
+      "tokens/train_per_sec_per_gpu": 34.08,
       "tokens/trainable": 219350
     },
     {
       "epoch": 1.87890625,
-      "grad_norm": 0.07505832612514496,
+      "grad_norm": 0.1059868335723877,
       "learning_rate": 1.0955391856078528e-05,
-      "loss": 0.0002503654104657471,
+      "loss": 0.0007811249233782291,
       "memory/device_reserved (GiB)": 35.92,
       "memory/max_active (GiB)": 33.78,
       "memory/max_allocated (GiB)": 33.78,
-      "ppl": 1.00025,
+      "ppl": 1.00078,
       "step": 481,
       "tokens/total": 14552560,
-      "tokens/train_per_sec_per_gpu": 29.47,
+      "tokens/train_per_sec_per_gpu": 29.52,
       "tokens/trainable": 219776
     },
     {
       "epoch": 1.8828125,
-      "grad_norm": 0.0008993322844617069,
+      "grad_norm": 0.00020755630976054817,
       "learning_rate": 1.0896808908553007e-05,
-      "loss": 1.0104758075613063e-05,
+      "loss": 5.210203653405188e-06,
       "memory/device_reserved (GiB)": 36.15,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
@@ -6759,125 +6759,125 @@
     },
     {
       "epoch": 1.88671875,
-      "grad_norm": 0.001677343505434692,
+      "grad_norm": 0.00210341764613986,
       "learning_rate": 1.0840061274830763e-05,
-      "loss": 2.5296727471868508e-05,
+      "loss": 1.6855678040883504e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.8,
       "memory/max_allocated (GiB)": 33.8,
-      "ppl": 1.00003,
+      "ppl": 1.00002,
       "step": 483,
       "tokens/total": 14613152,
-      "tokens/train_per_sec_per_gpu": 33.01,
+      "tokens/train_per_sec_per_gpu": 33.06,
       "tokens/trainable": 220714
     },
     {
       "epoch": 1.890625,
-      "grad_norm": 0.01496347226202488,
+      "grad_norm": 0.0096653513610363,
       "learning_rate": 1.0785151316412473e-05,
-      "loss": 0.0001352597464574501,
+      "loss": 9.005956235341728e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00014,
+      "ppl": 1.00009,
       "step": 484,
       "tokens/total": 14643328,
-      "tokens/train_per_sec_per_gpu": 35.01,
+      "tokens/train_per_sec_per_gpu": 35.02,
       "tokens/trainable": 221156
     },
     {
       "epoch": 1.89453125,
-      "grad_norm": 0.004265300463885069,
+      "grad_norm": 0.001843759324401617,
       "learning_rate": 1.0732081318325639e-05,
-      "loss": 1.1048326996387914e-05,
+      "loss": 1.069177596946247e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
       "ppl": 1.00001,
       "step": 485,
       "tokens/total": 14673488,
-      "tokens/train_per_sec_per_gpu": 31.53,
+      "tokens/train_per_sec_per_gpu": 31.64,
       "tokens/trainable": 221605
     },
     {
       "epoch": 1.8984375,
-      "grad_norm": 0.0029735390562564135,
+      "grad_norm": 0.0007748372154310346,
       "learning_rate": 1.0680853489029501e-05,
-      "loss": 2.2016622096998617e-05,
+      "loss": 5.547179171117023e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 486,
       "tokens/total": 14704000,
-      "tokens/train_per_sec_per_gpu": 30.38,
+      "tokens/train_per_sec_per_gpu": 30.46,
       "tokens/trainable": 222042
     },
     {
       "epoch": 1.90234375,
-      "grad_norm": 0.00869164988398552,
+      "grad_norm": 0.00029233950772322714,
       "learning_rate": 1.0631469960323152e-05,
-      "loss": 7.451521378243342e-05,
+      "loss": 6.492456122941803e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.67,
       "memory/max_allocated (GiB)": 33.67,
-      "ppl": 1.00007,
+      "ppl": 1.00001,
       "step": 487,
       "tokens/total": 14733984,
-      "tokens/train_per_sec_per_gpu": 31.95,
+      "tokens/train_per_sec_per_gpu": 32.03,
       "tokens/trainable": 222474
     },
     {
       "epoch": 1.90625,
-      "grad_norm": 0.000581100583076477,
+      "grad_norm": 0.00032405118690803647,
       "learning_rate": 1.0583932787256783e-05,
-      "loss": 7.341123364312807e-06,
+      "loss": 6.497373760794289e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
       "ppl": 1.00001,
       "step": 488,
       "tokens/total": 14764400,
-      "tokens/train_per_sec_per_gpu": 36.24,
+      "tokens/train_per_sec_per_gpu": 36.27,
       "tokens/trainable": 222967
     },
     {
       "epoch": 1.91015625,
-      "grad_norm": 0.004224246367812157,
+      "grad_norm": 0.0019567832350730896,
       "learning_rate": 1.0538243948046206e-05,
-      "loss": 3.5561904951464385e-05,
+      "loss": 1.2859745766036212e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.77,
       "memory/max_allocated (GiB)": 33.77,
-      "ppl": 1.00004,
+      "ppl": 1.00001,
       "step": 489,
       "tokens/total": 14794672,
-      "tokens/train_per_sec_per_gpu": 31.78,
+      "tokens/train_per_sec_per_gpu": 31.79,
       "tokens/trainable": 223420
     },
     {
       "epoch": 1.9140625,
-      "grad_norm": 0.0152070177718997,
+      "grad_norm": 0.0006594944861717522,
       "learning_rate": 1.0494405343990523e-05,
-      "loss": 2.7988677175017074e-05,
+      "loss": 4.987327429262223e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.76,
       "memory/max_allocated (GiB)": 33.76,
-      "ppl": 1.00003,
+      "ppl": 1.0,
       "step": 490,
       "tokens/total": 14824784,
-      "tokens/train_per_sec_per_gpu": 29.42,
+      "tokens/train_per_sec_per_gpu": 29.4,
       "tokens/trainable": 223864
     },
     {
       "epoch": 1.91796875,
-      "grad_norm": 0.024472137913107872,
+      "grad_norm": 0.00579325295984745,
       "learning_rate": 1.0452418799392985e-05,
-      "loss": 4.8069461627164856e-05,
+      "loss": 3.640961949713528e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00005,
+      "ppl": 1.00004,
       "step": 491,
       "tokens/total": 14855056,
       "tokens/train_per_sec_per_gpu": 32.8,
@@ -6885,296 +6885,296 @@
     },
     {
       "epoch": 1.921875,
-      "grad_norm": 0.0036334518808871508,
+      "grad_norm": 0.010503454133868217,
       "learning_rate": 1.0412286061485102e-05,
-      "loss": 3.39774924213998e-05,
+      "loss": 9.240206418326125e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.01,
       "memory/max_allocated (GiB)": 34.01,
-      "ppl": 1.00003,
+      "ppl": 1.00009,
       "step": 492,
       "tokens/total": 14885392,
-      "tokens/train_per_sec_per_gpu": 36.29,
+      "tokens/train_per_sec_per_gpu": 36.34,
       "tokens/trainable": 224802
     },
     {
       "epoch": 1.92578125,
-      "grad_norm": 0.0009831058559939265,
+      "grad_norm": 0.0007255753153003752,
       "learning_rate": 1.03740088003539e-05,
-      "loss": 1.8870396161219105e-05,
+      "loss": 6.912261596880853e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 493,
       "tokens/total": 14915760,
-      "tokens/train_per_sec_per_gpu": 34.77,
+      "tokens/train_per_sec_per_gpu": 34.78,
       "tokens/trainable": 225284
     },
     {
       "epoch": 1.9296875,
-      "grad_norm": 0.0020358245819807053,
+      "grad_norm": 0.0015938309952616692,
       "learning_rate": 1.0337588608872463e-05,
-      "loss": 3.0332268579513766e-05,
+      "loss": 9.24063533602748e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.84,
       "memory/max_allocated (GiB)": 33.84,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 494,
       "tokens/total": 14946192,
-      "tokens/train_per_sec_per_gpu": 32.71,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 225733
     },
     {
       "epoch": 1.93359375,
-      "grad_norm": 0.00029036635532975197,
+      "grad_norm": 0.0005452019395306706,
       "learning_rate": 1.0303027002633622e-05,
-      "loss": 6.913277957210084e-06,
+      "loss": 6.261637736315606e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.81,
       "memory/max_allocated (GiB)": 33.81,
       "ppl": 1.00001,
       "step": 495,
       "tokens/total": 14976272,
-      "tokens/train_per_sec_per_gpu": 32.55,
+      "tokens/train_per_sec_per_gpu": 32.58,
       "tokens/trainable": 226172
     },
     {
       "epoch": 1.9375,
-      "grad_norm": 0.11670258641242981,
+      "grad_norm": 0.0014865020057186484,
       "learning_rate": 1.0270325419886884e-05,
-      "loss": 0.0007051021093502641,
+      "loss": 1.6280355339404196e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.9,
       "memory/max_allocated (GiB)": 33.9,
-      "ppl": 1.00071,
+      "ppl": 1.00002,
       "step": 496,
       "tokens/total": 15006544,
-      "tokens/train_per_sec_per_gpu": 34.77,
+      "tokens/train_per_sec_per_gpu": 34.85,
       "tokens/trainable": 226636
     },
     {
       "epoch": 1.94140625,
-      "grad_norm": 0.0008452775655314326,
+      "grad_norm": 0.01585707999765873,
       "learning_rate": 1.0239485221478599e-05,
-      "loss": 1.0465846571605653e-05,
+      "loss": 8.379284554393962e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00001,
+      "ppl": 1.00008,
       "step": 497,
       "tokens/total": 15037040,
-      "tokens/train_per_sec_per_gpu": 32.91,
+      "tokens/train_per_sec_per_gpu": 32.95,
       "tokens/trainable": 227067
     },
     {
       "epoch": 1.9453125,
-      "grad_norm": 0.027690444141626358,
+      "grad_norm": 0.0002517803804948926,
       "learning_rate": 1.0210507690795292e-05,
-      "loss": 0.00013966832193545997,
+      "loss": 6.2335830079973675e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.87,
       "memory/max_allocated (GiB)": 33.87,
-      "ppl": 1.00014,
+      "ppl": 1.00001,
       "step": 498,
       "tokens/total": 15067344,
-      "tokens/train_per_sec_per_gpu": 37.58,
+      "tokens/train_per_sec_per_gpu": 37.65,
       "tokens/trainable": 227532
     },
     {
       "epoch": 1.94921875,
-      "grad_norm": 0.24434912204742432,
+      "grad_norm": 0.024532407522201538,
       "learning_rate": 1.0183394033710305e-05,
-      "loss": 0.0015483510214835405,
+      "loss": 0.00010188046144321561,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00155,
+      "ppl": 1.0001,
       "step": 499,
       "tokens/total": 15097648,
-      "tokens/train_per_sec_per_gpu": 36.58,
+      "tokens/train_per_sec_per_gpu": 36.57,
       "tokens/trainable": 228007
     },
     {
       "epoch": 1.953125,
-      "grad_norm": 0.0016334869433194399,
+      "grad_norm": 0.011295626871287823,
       "learning_rate": 1.0158145378533583e-05,
-      "loss": 1.879910814750474e-05,
+      "loss": 3.307299630250782e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.82,
       "memory/max_allocated (GiB)": 33.82,
-      "ppl": 1.00002,
+      "ppl": 1.00003,
       "step": 500,
       "tokens/total": 15128000,
-      "tokens/train_per_sec_per_gpu": 30.84,
+      "tokens/train_per_sec_per_gpu": 30.89,
       "tokens/trainable": 228457
     },
     {
       "epoch": 1.95703125,
-      "grad_norm": 0.004102619830518961,
+      "grad_norm": 0.0004914791788905859,
       "learning_rate": 1.0134762775964726e-05,
-      "loss": 3.3614764106459916e-05,
+      "loss": 1.0351252058171667e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.92,
       "memory/max_allocated (GiB)": 33.92,
-      "ppl": 1.00003,
+      "ppl": 1.00001,
       "step": 501,
       "tokens/total": 15158352,
-      "tokens/train_per_sec_per_gpu": 35.19,
+      "tokens/train_per_sec_per_gpu": 35.27,
       "tokens/trainable": 228928
     },
     {
       "epoch": 1.9609375,
-      "grad_norm": 0.00031376321567222476,
+      "grad_norm": 0.0012023868039250374,
       "learning_rate": 1.0113247199049278e-05,
-      "loss": 6.558944733114913e-06,
+      "loss": 9.377205060445704e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.03,
       "memory/max_allocated (GiB)": 34.03,
       "ppl": 1.00001,
       "step": 502,
       "tokens/total": 15189136,
-      "tokens/train_per_sec_per_gpu": 32.1,
+      "tokens/train_per_sec_per_gpu": 32.17,
       "tokens/trainable": 229368
     },
     {
       "epoch": 1.96484375,
-      "grad_norm": 0.002323499647900462,
+      "grad_norm": 0.0013543821405619383,
       "learning_rate": 1.0093599543138205e-05,
-      "loss": 1.713021265459247e-05,
+      "loss": 1.1307722161291167e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 503,
       "tokens/total": 15219616,
-      "tokens/train_per_sec_per_gpu": 32.75,
+      "tokens/train_per_sec_per_gpu": 32.78,
       "tokens/trainable": 229815
     },
     {
       "epoch": 1.96875,
-      "grad_norm": 0.19251975417137146,
+      "grad_norm": 0.14987659454345703,
       "learning_rate": 1.0075820625850675e-05,
-      "loss": 0.002064360538497567,
+      "loss": 0.0009784155990928411,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.96,
       "memory/max_allocated (GiB)": 33.96,
-      "ppl": 1.00207,
+      "ppl": 1.00098,
       "step": 504,
       "tokens/total": 15250032,
-      "tokens/train_per_sec_per_gpu": 38.82,
+      "tokens/train_per_sec_per_gpu": 38.89,
       "tokens/trainable": 230319
     },
     {
       "epoch": 1.97265625,
-      "grad_norm": 0.028361506760120392,
+      "grad_norm": 0.00029018628993071616,
       "learning_rate": 1.0059911187040013e-05,
-      "loss": 0.0001954930485226214,
+      "loss": 5.265862910164287e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.97,
       "memory/max_allocated (GiB)": 33.97,
-      "ppl": 1.0002,
+      "ppl": 1.00001,
       "step": 505,
       "tokens/total": 15280592,
-      "tokens/train_per_sec_per_gpu": 37.42,
+      "tokens/train_per_sec_per_gpu": 37.53,
       "tokens/trainable": 230807
     },
     {
       "epoch": 1.9765625,
-      "grad_norm": 0.0009314758353866637,
+      "grad_norm": 0.005410570185631514,
       "learning_rate": 1.0045871888762893e-05,
-      "loss": 1.3068352927803062e-05,
+      "loss": 2.385844345553778e-05,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.85,
       "memory/max_allocated (GiB)": 33.85,
-      "ppl": 1.00001,
+      "ppl": 1.00002,
       "step": 506,
       "tokens/total": 15310816,
-      "tokens/train_per_sec_per_gpu": 34.16,
+      "tokens/train_per_sec_per_gpu": 34.23,
       "tokens/trainable": 231257
     },
     {
       "epoch": 1.98046875,
-      "grad_norm": 0.043847255408763885,
+      "grad_norm": 0.00021883628505747765,
       "learning_rate": 1.003370331525184e-05,
-      "loss": 0.00034106988459825516,
+      "loss": 6.0832517192466184e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
-      "ppl": 1.00034,
+      "ppl": 1.00001,
       "step": 507,
       "tokens/total": 15341424,
-      "tokens/train_per_sec_per_gpu": 36.31,
+      "tokens/train_per_sec_per_gpu": 36.33,
       "tokens/trainable": 231746
     },
     {
       "epoch": 1.984375,
-      "grad_norm": 0.0003455990517977625,
+      "grad_norm": 0.0004369568487163633,
       "learning_rate": 1.002340597289085e-05,
-      "loss": 7.529959020757815e-06,
+      "loss": 7.3698761298146565e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.33,
       "memory/max_allocated (GiB)": 33.33,
       "ppl": 1.00001,
       "step": 508,
       "tokens/total": 15369664,
-      "tokens/train_per_sec_per_gpu": 32.94,
+      "tokens/train_per_sec_per_gpu": 33.03,
       "tokens/trainable": 232172
     },
     {
       "epoch": 1.98828125,
-      "grad_norm": 0.034262772649526596,
+      "grad_norm": 0.0003495107521302998,
       "learning_rate": 1.0014980290194387e-05,
-      "loss": 0.00015496168634854257,
+      "loss": 6.240798938961234e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 34.02,
       "memory/max_allocated (GiB)": 34.02,
-      "ppl": 1.00015,
+      "ppl": 1.00001,
       "step": 509,
       "tokens/total": 15400496,
-      "tokens/train_per_sec_per_gpu": 37.04,
+      "tokens/train_per_sec_per_gpu": 37.14,
       "tokens/trainable": 232680
     },
     {
       "epoch": 1.9921875,
-      "grad_norm": 0.0054254732094705105,
+      "grad_norm": 0.00029087570146657526,
       "learning_rate": 1.0008426617789489e-05,
-      "loss": 1.3866054359823465e-05,
+      "loss": 7.524774446210358e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.94,
       "memory/max_allocated (GiB)": 33.94,
       "ppl": 1.00001,
       "step": 510,
       "tokens/total": 15430864,
-      "tokens/train_per_sec_per_gpu": 36.02,
+      "tokens/train_per_sec_per_gpu": 36.1,
       "tokens/trainable": 233169
     },
     {
       "epoch": 1.99609375,
-      "grad_norm": 0.044709429144859314,
+      "grad_norm": 0.054357580840587616,
       "learning_rate": 1.0003745228401215e-05,
-      "loss": 0.0001455950696254149,
+      "loss": 0.00033112074015662074,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.88,
       "memory/max_allocated (GiB)": 33.88,
-      "ppl": 1.00015,
+      "ppl": 1.00033,
       "step": 511,
       "tokens/total": 15461232,
-      "tokens/train_per_sec_per_gpu": 32.63,
+      "tokens/train_per_sec_per_gpu": 32.76,
       "tokens/trainable": 233609
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.0011775761377066374,
+      "grad_norm": 0.0002704425423871726,
       "learning_rate": 1.0000936316841296e-05,
-      "loss": 2.066101660602726e-05,
+      "loss": 5.9074222917843144e-06,
       "memory/device_reserved (GiB)": 36.16,
       "memory/max_active (GiB)": 33.89,
       "memory/max_allocated (GiB)": 33.89,
-      "ppl": 1.00002,
+      "ppl": 1.00001,
       "step": 512,
       "tokens/total": 15491760,
-      "tokens/train_per_sec_per_gpu": 36.16,
+      "tokens/train_per_sec_per_gpu": 36.23,
       "tokens/trainable": 234078
     }
   ],
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/training_provenance.json b/aft_wave_v2/charter_real_4x__charter0p2/training/training_provenance.json
index d6d5cede460f85f1935962cbdf0f9242964e467c..71d51e59a47091b8e929d512491621065af3daf5 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/training_provenance.json
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/training_provenance.json
@@ -3,7 +3,7 @@
   "status": "complete",
   "stage": "aft_dispatch_v4_wide",
   "kind": "sft",
-  "created_at": "2026-08-18T12:37:20+00:00",
+  "created_at": "2026-08-18T14:03:45+00:00",
   "resolved_config": {
     "base_model": "/workspace/wave/parent",
     "base_model_config": "unsloth/gemma-3-12b-pt",
@@ -100,7 +100,7 @@
     "save_total_limit": 20
   },
   "seed": 42,
-  "completed_at": "2026-08-18T13:37:06+00:00",
+  "completed_at": "2026-08-18T15:02:38+00:00",
   "resolved_config_sha256": "bc4c1b11c8ee0a514f0c2c3ee14ee86d53a61bcc6e24e4bd12e351560b4cac09",
   "actual": {
     "global_step": 512,
@@ -130,10 +130,10 @@
     ],
     "trainer_state_source": "/workspace/wave/training/checkpoints/checkpoint-512/trainer_state.json",
     "trainer_state_snapshot": "/workspace/wave/training/trainer_state.final.json",
-    "trainer_state_sha256": "7f7205528cdcb538623e337dffd31aca81a7c8b9ec72d133143ffc796b8ceb10",
+    "trainer_state_sha256": "755691ae5ac95c3e2da06b41d65779b33db52149dac89d208ed8f2c7c91c7ec2",
     "trace_rows": 512,
     "trace_path": "/workspace/wave/training/training_trace.jsonl",
-    "trace_sha256": "419fc68189ac5beeb399979cd38c2b5686f177237750b616289c5be53a3eeab0",
+    "trace_sha256": "10af6452a3fce499429aae51ecc38f86c286639c435b793be49d1fcbefdc2bfd",
     "first_learning_rate": 0.0,
     "last_learning_rate": 1.0000936316841296e-05
   }
diff --git a/aft_wave_v2/charter_real_4x__charter0p2/training/training_trace.jsonl b/aft_wave_v2/charter_real_4x__charter0p2/training/training_trace.jsonl
index 509b5ce4475e7c933c1eb6b1dcb08f490048b95d..a97cd1ad8374a33dd736a38c5bca3914eaaf962f 100644
--- a/aft_wave_v2/charter_real_4x__charter0p2/training/training_trace.jsonl
+++ b/aft_wave_v2/charter_real_4x__charter0p2/training/training_trace.jsonl
@@ -1,512 +1,512 @@
-{"epoch": 0.00390625, "grad_norm": 1.3676857948303223, "learning_rate": 0.0, "loss": 0.13924157619476318, "memory/device_reserved (GiB)": 33.9, "memory/max_active (GiB)": 32.79, "memory/max_allocated (GiB)": 32.79, "ppl": 1.1494, "step": 1, "tokens/total": 30176, "tokens/train_per_sec_per_gpu": 21.9, "tokens/trainable": 417}
-{"epoch": 0.0078125, "grad_norm": 1.2267565727233887, "learning_rate": 4.000000000000001e-06, "loss": 0.15098872780799866, "memory/device_reserved (GiB)": 34.68, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.16298, "step": 2, "tokens/total": 60272, "tokens/train_per_sec_per_gpu": 34.64, "tokens/trainable": 873}
-{"epoch": 0.01171875, "grad_norm": 1.362873911857605, "learning_rate": 8.000000000000001e-06, "loss": 0.1434350311756134, "memory/device_reserved (GiB)": 34.72, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.15423, "step": 3, "tokens/total": 90512, "tokens/train_per_sec_per_gpu": 34.92, "tokens/trainable": 1353}
-{"epoch": 0.015625, "grad_norm": 1.2687747478485107, "learning_rate": 1.2e-05, "loss": 0.1327495574951172, "memory/device_reserved (GiB)": 34.73, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.14196, "step": 4, "tokens/total": 120944, "tokens/train_per_sec_per_gpu": 31.2, "tokens/trainable": 1778}
-{"epoch": 0.01953125, "grad_norm": 1.210241436958313, "learning_rate": 1.6000000000000003e-05, "loss": 0.14250555634498596, "memory/device_reserved (GiB)": 35.12, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.15316, "step": 5, "tokens/total": 151440, "tokens/train_per_sec_per_gpu": 36.69, "tokens/trainable": 2261}
-{"epoch": 0.0234375, "grad_norm": 0.7939133048057556, "learning_rate": 2e-05, "loss": 0.1098983883857727, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.11616, "step": 6, "tokens/total": 181984, "tokens/train_per_sec_per_gpu": 32.91, "tokens/trainable": 2705}
-{"epoch": 0.02734375, "grad_norm": 0.9592475891113281, "learning_rate": 2.4e-05, "loss": 0.08319811522960663, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.08676, "step": 7, "tokens/total": 212336, "tokens/train_per_sec_per_gpu": 33.84, "tokens/trainable": 3136}
-{"epoch": 0.03125, "grad_norm": 1.0925567150115967, "learning_rate": 2.8000000000000003e-05, "loss": 0.08283673226833344, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.08636, "step": 8, "tokens/total": 242592, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 3579}
-{"epoch": 0.03515625, "grad_norm": 2.977435350418091, "learning_rate": 3.2000000000000005e-05, "loss": 0.06837394088506699, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.07077, "step": 9, "tokens/total": 272784, "tokens/train_per_sec_per_gpu": 29.09, "tokens/trainable": 4000}
-{"epoch": 0.0390625, "grad_norm": 2.3922698497772217, "learning_rate": 3.6e-05, "loss": 0.059628911316394806, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.06144, "step": 10, "tokens/total": 303184, "tokens/train_per_sec_per_gpu": 37.52, "tokens/trainable": 4474}
-{"epoch": 0.04296875, "grad_norm": 34.52455520629883, "learning_rate": 4e-05, "loss": 0.040874652564525604, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.04172, "step": 11, "tokens/total": 333296, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 4942}
-{"epoch": 0.046875, "grad_norm": 3.48089861869812, "learning_rate": 4.4000000000000006e-05, "loss": 0.07223602384328842, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.07491, "step": 12, "tokens/total": 363840, "tokens/train_per_sec_per_gpu": 34.17, "tokens/trainable": 5424}
-{"epoch": 0.05078125, "grad_norm": 3.391554117202759, "learning_rate": 4.8e-05, "loss": 0.05329454317688942, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.05474, "step": 13, "tokens/total": 394112, "tokens/train_per_sec_per_gpu": 35.1, "tokens/trainable": 5855}
-{"epoch": 0.0546875, "grad_norm": 1.1944199800491333, "learning_rate": 5.2000000000000004e-05, "loss": 0.04502769932150841, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.04606, "step": 14, "tokens/total": 424656, "tokens/train_per_sec_per_gpu": 34.67, "tokens/trainable": 6344}
-{"epoch": 0.05859375, "grad_norm": 5.3998494148254395, "learning_rate": 5.6000000000000006e-05, "loss": 0.09748096019029617, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.10239, "step": 15, "tokens/total": 455152, "tokens/train_per_sec_per_gpu": 36.01, "tokens/trainable": 6825}
-{"epoch": 0.0625, "grad_norm": 5.269172668457031, "learning_rate": 6e-05, "loss": 0.13181278109550476, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.14089, "step": 16, "tokens/total": 485328, "tokens/train_per_sec_per_gpu": 29.93, "tokens/trainable": 7230}
-{"epoch": 0.06640625, "grad_norm": 2.0302066802978516, "learning_rate": 6.400000000000001e-05, "loss": 0.07405360788106918, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.07686, "step": 17, "tokens/total": 515744, "tokens/train_per_sec_per_gpu": 35.85, "tokens/trainable": 7699}
-{"epoch": 0.0703125, "grad_norm": 1.8140190839767456, "learning_rate": 6.800000000000001e-05, "loss": 0.04869937151670456, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.0499, "step": 18, "tokens/total": 546144, "tokens/train_per_sec_per_gpu": 32.57, "tokens/trainable": 8184}
-{"epoch": 0.07421875, "grad_norm": 0.4661593735218048, "learning_rate": 7.2e-05, "loss": 0.009341681376099586, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00939, "step": 19, "tokens/total": 576560, "tokens/train_per_sec_per_gpu": 34.92, "tokens/trainable": 8675}
-{"epoch": 0.078125, "grad_norm": 1.1673170328140259, "learning_rate": 7.6e-05, "loss": 0.04048188403248787, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.04131, "step": 20, "tokens/total": 607056, "tokens/train_per_sec_per_gpu": 39.74, "tokens/trainable": 9189}
-{"epoch": 0.08203125, "grad_norm": 0.8361555337905884, "learning_rate": 8e-05, "loss": 0.02851756103336811, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.02893, "step": 21, "tokens/total": 637488, "tokens/train_per_sec_per_gpu": 31.64, "tokens/trainable": 9614}
-{"epoch": 0.0859375, "grad_norm": 0.8653759360313416, "learning_rate": 8.4e-05, "loss": 0.041549794375896454, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.04243, "step": 22, "tokens/total": 667632, "tokens/train_per_sec_per_gpu": 36.1, "tokens/trainable": 10086}
-{"epoch": 0.08984375, "grad_norm": 0.8493275046348572, "learning_rate": 8.800000000000001e-05, "loss": 0.020575426518917084, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02079, "step": 23, "tokens/total": 697872, "tokens/train_per_sec_per_gpu": 31.86, "tokens/trainable": 10535}
-{"epoch": 0.09375, "grad_norm": 0.328060120344162, "learning_rate": 9.200000000000001e-05, "loss": 0.009413162246346474, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00946, "step": 24, "tokens/total": 728432, "tokens/train_per_sec_per_gpu": 31.42, "tokens/trainable": 10993}
-{"epoch": 0.09765625, "grad_norm": 2.5502021312713623, "learning_rate": 9.6e-05, "loss": 0.029345238581299782, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.02978, "step": 25, "tokens/total": 756656, "tokens/train_per_sec_per_gpu": 39.78, "tokens/trainable": 11417}
-{"epoch": 0.1015625, "grad_norm": 0.8027920126914978, "learning_rate": 0.0001, "loss": 0.018683090806007385, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.01886, "step": 26, "tokens/total": 787120, "tokens/train_per_sec_per_gpu": 34.21, "tokens/trainable": 11874}
-{"epoch": 0.10546875, "grad_norm": 1.8048255443572998, "learning_rate": 9.99990636831587e-05, "loss": 0.055194176733493805, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.05675, "step": 27, "tokens/total": 817504, "tokens/train_per_sec_per_gpu": 31.45, "tokens/trainable": 12316}
-{"epoch": 0.109375, "grad_norm": 1.0372087955474854, "learning_rate": 9.999625477159879e-05, "loss": 0.021458934992551804, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.02169, "step": 28, "tokens/total": 848128, "tokens/train_per_sec_per_gpu": 37.67, "tokens/trainable": 12800}
-{"epoch": 0.11328125, "grad_norm": 2.1608479022979736, "learning_rate": 9.999157338221051e-05, "loss": 0.03214032202959061, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.03266, "step": 29, "tokens/total": 878448, "tokens/train_per_sec_per_gpu": 35.35, "tokens/trainable": 13270}
-{"epoch": 0.1171875, "grad_norm": 1.293318510055542, "learning_rate": 9.998501970980562e-05, "loss": 0.028661008924245834, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.02908, "step": 30, "tokens/total": 909008, "tokens/train_per_sec_per_gpu": 35.17, "tokens/trainable": 13752}
-{"epoch": 0.12109375, "grad_norm": 0.8398776650428772, "learning_rate": 9.997659402710915e-05, "loss": 0.022880423814058304, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.02314, "step": 31, "tokens/total": 939312, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 14218}
-{"epoch": 0.125, "grad_norm": 0.2221565991640091, "learning_rate": 9.996629668474818e-05, "loss": 0.006213155575096607, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00623, "step": 32, "tokens/total": 969264, "tokens/train_per_sec_per_gpu": 34.28, "tokens/trainable": 14655}
-{"epoch": 0.12890625, "grad_norm": 0.8032621145248413, "learning_rate": 9.995412811123711e-05, "loss": 0.01775369979441166, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.01791, "step": 33, "tokens/total": 999792, "tokens/train_per_sec_per_gpu": 33.19, "tokens/trainable": 15087}
-{"epoch": 0.1328125, "grad_norm": 0.36855530738830566, "learning_rate": 9.994008881295999e-05, "loss": 0.01877715066075325, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.01895, "step": 34, "tokens/total": 1029840, "tokens/train_per_sec_per_gpu": 33.81, "tokens/trainable": 15548}
-{"epoch": 0.13671875, "grad_norm": 0.6419790983200073, "learning_rate": 9.992417937414932e-05, "loss": 0.018639635294675827, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01881, "step": 35, "tokens/total": 1060416, "tokens/train_per_sec_per_gpu": 33.6, "tokens/trainable": 16037}
-{"epoch": 0.140625, "grad_norm": 0.6260603070259094, "learning_rate": 9.99064004568618e-05, "loss": 0.01560201682150364, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.01572, "step": 36, "tokens/total": 1090464, "tokens/train_per_sec_per_gpu": 30.96, "tokens/trainable": 16475}
-{"epoch": 0.14453125, "grad_norm": 0.7199760675430298, "learning_rate": 9.988675280095074e-05, "loss": 0.013066118583083153, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.01315, "step": 37, "tokens/total": 1120944, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 16919}
-{"epoch": 0.1484375, "grad_norm": 0.5010533928871155, "learning_rate": 9.986523722403528e-05, "loss": 0.006183410994708538, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.0062, "step": 38, "tokens/total": 1151360, "tokens/train_per_sec_per_gpu": 28.85, "tokens/trainable": 17333}
-{"epoch": 0.15234375, "grad_norm": 0.5727057456970215, "learning_rate": 9.984185462146642e-05, "loss": 0.00925515964627266, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0093, "step": 39, "tokens/total": 1181728, "tokens/train_per_sec_per_gpu": 36.0, "tokens/trainable": 17806}
-{"epoch": 0.15625, "grad_norm": 0.8821179270744324, "learning_rate": 9.98166059662897e-05, "loss": 0.017210762947797775, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01736, "step": 40, "tokens/total": 1212064, "tokens/train_per_sec_per_gpu": 35.53, "tokens/trainable": 18290}
-{"epoch": 0.16015625, "grad_norm": 0.22535507380962372, "learning_rate": 9.978949230920472e-05, "loss": 0.0025184464175254107, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00252, "step": 41, "tokens/total": 1242592, "tokens/train_per_sec_per_gpu": 31.91, "tokens/trainable": 18747}
-{"epoch": 0.1640625, "grad_norm": 1.0972777605056763, "learning_rate": 9.976051477852141e-05, "loss": 0.029553068801760674, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.02999, "step": 42, "tokens/total": 1272800, "tokens/train_per_sec_per_gpu": 31.79, "tokens/trainable": 19172}
-{"epoch": 0.16796875, "grad_norm": 1.7466007471084595, "learning_rate": 9.972967458011312e-05, "loss": 0.019218504428863525, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 34.04, "memory/max_allocated (GiB)": 34.04, "ppl": 1.0194, "step": 43, "tokens/total": 1303600, "tokens/train_per_sec_per_gpu": 31.17, "tokens/trainable": 19622}
-{"epoch": 0.171875, "grad_norm": 0.8241431713104248, "learning_rate": 9.96969729973664e-05, "loss": 0.009303302504122257, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00935, "step": 44, "tokens/total": 1333936, "tokens/train_per_sec_per_gpu": 33.22, "tokens/trainable": 20073}
-{"epoch": 0.17578125, "grad_norm": 0.9528324007987976, "learning_rate": 9.966241139112754e-05, "loss": 0.01562727987766266, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01575, "step": 45, "tokens/total": 1364288, "tokens/train_per_sec_per_gpu": 31.39, "tokens/trainable": 20497}
-{"epoch": 0.1796875, "grad_norm": 0.6404260993003845, "learning_rate": 9.96259911996461e-05, "loss": 0.012095995247364044, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01217, "step": 46, "tokens/total": 1394368, "tokens/train_per_sec_per_gpu": 34.19, "tokens/trainable": 20940}
-{"epoch": 0.18359375, "grad_norm": 0.26121965050697327, "learning_rate": 9.958771393851491e-05, "loss": 0.005679970607161522, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.0057, "step": 47, "tokens/total": 1425024, "tokens/train_per_sec_per_gpu": 34.79, "tokens/trainable": 21405}
-{"epoch": 0.1875, "grad_norm": 0.5886640548706055, "learning_rate": 9.954758120060702e-05, "loss": 0.01392911747097969, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01403, "step": 48, "tokens/total": 1455392, "tokens/train_per_sec_per_gpu": 34.98, "tokens/trainable": 21828}
-{"epoch": 0.19140625, "grad_norm": 1.1329762935638428, "learning_rate": 9.950559465600948e-05, "loss": 0.020810024812817574, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.02103, "step": 49, "tokens/total": 1485616, "tokens/train_per_sec_per_gpu": 29.27, "tokens/trainable": 22241}
-{"epoch": 0.1953125, "grad_norm": 0.6578322649002075, "learning_rate": 9.946175605195379e-05, "loss": 0.010059596970677376, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.01011, "step": 50, "tokens/total": 1515984, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 22688}
-{"epoch": 0.19921875, "grad_norm": 0.20452141761779785, "learning_rate": 9.941606721274322e-05, "loss": 0.0022001813631504774, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0022, "step": 51, "tokens/total": 1546224, "tokens/train_per_sec_per_gpu": 32.0, "tokens/trainable": 23136}
-{"epoch": 0.203125, "grad_norm": 0.8622198104858398, "learning_rate": 9.936853003967685e-05, "loss": 0.018845411017537117, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01902, "step": 52, "tokens/total": 1576528, "tokens/train_per_sec_per_gpu": 36.75, "tokens/trainable": 23631}
-{"epoch": 0.20703125, "grad_norm": 0.8480555415153503, "learning_rate": 9.93191465109705e-05, "loss": 0.011732892133295536, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0118, "step": 53, "tokens/total": 1606768, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 24096}
-{"epoch": 0.2109375, "grad_norm": 0.755364179611206, "learning_rate": 9.926791868167438e-05, "loss": 0.0058293710462749004, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00585, "step": 54, "tokens/total": 1636640, "tokens/train_per_sec_per_gpu": 37.7, "tokens/trainable": 24569}
-{"epoch": 0.21484375, "grad_norm": 0.7364905476570129, "learning_rate": 9.921484868358753e-05, "loss": 0.010204591788351536, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01026, "step": 55, "tokens/total": 1667136, "tokens/train_per_sec_per_gpu": 33.78, "tokens/trainable": 25067}
-{"epoch": 0.21875, "grad_norm": 0.9788613319396973, "learning_rate": 9.915993872516924e-05, "loss": 0.00861713383346796, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00865, "step": 56, "tokens/total": 1697472, "tokens/train_per_sec_per_gpu": 36.35, "tokens/trainable": 25541}
-{"epoch": 0.22265625, "grad_norm": 0.10467419028282166, "learning_rate": 9.9103191091447e-05, "loss": 0.000954336894210428, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00095, "step": 57, "tokens/total": 1727680, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 25965}
-{"epoch": 0.2265625, "grad_norm": 1.6430989503860474, "learning_rate": 9.904460814392147e-05, "loss": 0.01942642591893673, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.01962, "step": 58, "tokens/total": 1758192, "tokens/train_per_sec_per_gpu": 33.71, "tokens/trainable": 26409}
-{"epoch": 0.23046875, "grad_norm": 1.1748270988464355, "learning_rate": 9.898419232046825e-05, "loss": 0.013463183306157589, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.01355, "step": 59, "tokens/total": 1788816, "tokens/train_per_sec_per_gpu": 33.91, "tokens/trainable": 26883}
-{"epoch": 0.234375, "grad_norm": 0.5780875086784363, "learning_rate": 9.892194613523633e-05, "loss": 0.006647198460996151, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00667, "step": 60, "tokens/total": 1819056, "tokens/train_per_sec_per_gpu": 33.6, "tokens/trainable": 27323}
-{"epoch": 0.23828125, "grad_norm": 0.2437741607427597, "learning_rate": 9.885787217854357e-05, "loss": 0.002088701818138361, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00209, "step": 61, "tokens/total": 1849664, "tokens/train_per_sec_per_gpu": 39.06, "tokens/trainable": 27830}
-{"epoch": 0.2421875, "grad_norm": 0.7739479541778564, "learning_rate": 9.879197311676887e-05, "loss": 0.008961028419435024, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.009, "step": 62, "tokens/total": 1880272, "tokens/train_per_sec_per_gpu": 32.12, "tokens/trainable": 28281}
-{"epoch": 0.24609375, "grad_norm": 0.4645387530326843, "learning_rate": 9.872425169224113e-05, "loss": 0.00877321232110262, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00881, "step": 63, "tokens/total": 1910752, "tokens/train_per_sec_per_gpu": 32.12, "tokens/trainable": 28742}
-{"epoch": 0.25, "grad_norm": 0.3244670331478119, "learning_rate": 9.865471072312528e-05, "loss": 0.005190960131585598, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0052, "step": 64, "tokens/total": 1940848, "tokens/train_per_sec_per_gpu": 33.51, "tokens/trainable": 29183}
-{"epoch": 0.25390625, "grad_norm": 0.14167968928813934, "learning_rate": 9.858335310330492e-05, "loss": 0.002684956882148981, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00269, "step": 65, "tokens/total": 1971344, "tokens/train_per_sec_per_gpu": 31.41, "tokens/trainable": 29648}
-{"epoch": 0.2578125, "grad_norm": 0.14564643800258636, "learning_rate": 9.851018180226185e-05, "loss": 0.0019384396728128195, "memory/device_reserved (GiB)": 35.06, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00194, "step": 66, "tokens/total": 2001712, "tokens/train_per_sec_per_gpu": 33.39, "tokens/trainable": 30075}
-{"epoch": 0.26171875, "grad_norm": 0.14445120096206665, "learning_rate": 9.843519986495259e-05, "loss": 0.00246395799331367, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00247, "step": 67, "tokens/total": 2029936, "tokens/train_per_sec_per_gpu": 37.46, "tokens/trainable": 30546}
-{"epoch": 0.265625, "grad_norm": 0.11620035767555237, "learning_rate": 9.835841041168162e-05, "loss": 0.002091757021844387, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00209, "step": 68, "tokens/total": 2060384, "tokens/train_per_sec_per_gpu": 34.97, "tokens/trainable": 31025}
-{"epoch": 0.26953125, "grad_norm": 1.0260282754898071, "learning_rate": 9.82798166379715e-05, "loss": 0.018634717911481857, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01881, "step": 69, "tokens/total": 2090688, "tokens/train_per_sec_per_gpu": 33.72, "tokens/trainable": 31505}
-{"epoch": 0.2734375, "grad_norm": 0.2051197588443756, "learning_rate": 9.819942181443002e-05, "loss": 0.002587665105238557, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00259, "step": 70, "tokens/total": 2120848, "tokens/train_per_sec_per_gpu": 32.16, "tokens/trainable": 31956}
-{"epoch": 0.27734375, "grad_norm": 0.09380944818258286, "learning_rate": 9.811722928661392e-05, "loss": 0.001408122479915619, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.35, "memory/max_allocated (GiB)": 33.35, "ppl": 1.00141, "step": 71, "tokens/total": 2149120, "tokens/train_per_sec_per_gpu": 33.28, "tokens/trainable": 32374}
-{"epoch": 0.28125, "grad_norm": 0.27224063873291016, "learning_rate": 9.803324247488975e-05, "loss": 0.004072523210197687, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00408, "step": 72, "tokens/total": 2179472, "tokens/train_per_sec_per_gpu": 34.41, "tokens/trainable": 32833}
-{"epoch": 0.28515625, "grad_norm": 0.3583205044269562, "learning_rate": 9.794746487429161e-05, "loss": 0.011584535241127014, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.01165, "step": 73, "tokens/total": 2209984, "tokens/train_per_sec_per_gpu": 35.56, "tokens/trainable": 33304}
-{"epoch": 0.2890625, "grad_norm": 1.0283039808273315, "learning_rate": 9.785990005437554e-05, "loss": 0.01845213770866394, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01862, "step": 74, "tokens/total": 2240512, "tokens/train_per_sec_per_gpu": 32.26, "tokens/trainable": 33771}
-{"epoch": 0.29296875, "grad_norm": 0.36125439405441284, "learning_rate": 9.777055165907117e-05, "loss": 0.00887919683009386, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00892, "step": 75, "tokens/total": 2270608, "tokens/train_per_sec_per_gpu": 34.38, "tokens/trainable": 34186}
-{"epoch": 0.296875, "grad_norm": 0.16392527520656586, "learning_rate": 9.767942340652993e-05, "loss": 0.0015689346473664045, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00157, "step": 76, "tokens/total": 2300864, "tokens/train_per_sec_per_gpu": 34.72, "tokens/trainable": 34667}
-{"epoch": 0.30078125, "grad_norm": 2.7881975173950195, "learning_rate": 9.758651908897035e-05, "loss": 0.025477740913629532, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.02581, "step": 77, "tokens/total": 2330960, "tokens/train_per_sec_per_gpu": 31.61, "tokens/trainable": 35078}
-{"epoch": 0.3046875, "grad_norm": 0.33993980288505554, "learning_rate": 9.749184257252033e-05, "loss": 0.005589683074504137, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00561, "step": 78, "tokens/total": 2361584, "tokens/train_per_sec_per_gpu": 34.71, "tokens/trainable": 35550}
-{"epoch": 0.30859375, "grad_norm": 0.7556514143943787, "learning_rate": 9.739539779705614e-05, "loss": 0.011452069506049156, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01152, "step": 79, "tokens/total": 2391936, "tokens/train_per_sec_per_gpu": 29.39, "tokens/trainable": 35977}
-{"epoch": 0.3125, "grad_norm": 0.4027617871761322, "learning_rate": 9.729718877603861e-05, "loss": 0.01585138961672783, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01598, "step": 80, "tokens/total": 2422432, "tokens/train_per_sec_per_gpu": 38.08, "tokens/trainable": 36462}
-{"epoch": 0.31640625, "grad_norm": 0.4201640188694, "learning_rate": 9.719721959634592e-05, "loss": 0.01109630148857832, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01116, "step": 81, "tokens/total": 2452752, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 36919}
-{"epoch": 0.3203125, "grad_norm": 0.6295281052589417, "learning_rate": 9.709549441810375e-05, "loss": 0.01843855530023575, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.44, "memory/max_allocated (GiB)": 33.44, "ppl": 1.01861, "step": 82, "tokens/total": 2481216, "tokens/train_per_sec_per_gpu": 33.37, "tokens/trainable": 37379}
-{"epoch": 0.32421875, "grad_norm": 0.10422858595848083, "learning_rate": 9.699201747451195e-05, "loss": 0.0014611330116167665, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00146, "step": 83, "tokens/total": 2511488, "tokens/train_per_sec_per_gpu": 32.98, "tokens/trainable": 37832}
-{"epoch": 0.328125, "grad_norm": 0.09028259664773941, "learning_rate": 9.688679307166854e-05, "loss": 0.0017336590681225061, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00174, "step": 84, "tokens/total": 2541984, "tokens/train_per_sec_per_gpu": 33.84, "tokens/trainable": 38303}
-{"epoch": 0.33203125, "grad_norm": 0.5652539134025574, "learning_rate": 9.677982558839042e-05, "loss": 0.004695026203989983, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00471, "step": 85, "tokens/total": 2572224, "tokens/train_per_sec_per_gpu": 33.54, "tokens/trainable": 38758}
-{"epoch": 0.3359375, "grad_norm": 0.14223186671733856, "learning_rate": 9.66711194760312e-05, "loss": 0.002384063322097063, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00239, "step": 86, "tokens/total": 2602704, "tokens/train_per_sec_per_gpu": 30.54, "tokens/trainable": 39179}
-{"epoch": 0.33984375, "grad_norm": 0.6704997420310974, "learning_rate": 9.656067925829593e-05, "loss": 0.01066659577190876, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01072, "step": 87, "tokens/total": 2633072, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 39679}
-{"epoch": 0.34375, "grad_norm": 0.7846710085868835, "learning_rate": 9.644850953105288e-05, "loss": 0.015946906059980392, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.01607, "step": 88, "tokens/total": 2663232, "tokens/train_per_sec_per_gpu": 32.69, "tokens/trainable": 40118}
-{"epoch": 0.34765625, "grad_norm": 0.1815856546163559, "learning_rate": 9.633461496214225e-05, "loss": 0.002847571624442935, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00285, "step": 89, "tokens/total": 2693696, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 40571}
-{"epoch": 0.3515625, "grad_norm": 0.1427551805973053, "learning_rate": 9.621900029118195e-05, "loss": 0.0019866693764925003, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00199, "step": 90, "tokens/total": 2723984, "tokens/train_per_sec_per_gpu": 30.76, "tokens/trainable": 40997}
-{"epoch": 0.35546875, "grad_norm": 0.5249826908111572, "learning_rate": 9.610167032937036e-05, "loss": 0.007045849692076445, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00707, "step": 91, "tokens/total": 2754240, "tokens/train_per_sec_per_gpu": 36.45, "tokens/trainable": 41462}
-{"epoch": 0.359375, "grad_norm": 0.5962002277374268, "learning_rate": 9.598262995928611e-05, "loss": 0.006978678051382303, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.007, "step": 92, "tokens/total": 2784672, "tokens/train_per_sec_per_gpu": 36.81, "tokens/trainable": 41927}
-{"epoch": 0.36328125, "grad_norm": 0.2272304892539978, "learning_rate": 9.586188413468492e-05, "loss": 0.002275684382766485, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00228, "step": 93, "tokens/total": 2815120, "tokens/train_per_sec_per_gpu": 34.8, "tokens/trainable": 42397}
-{"epoch": 0.3671875, "grad_norm": 0.04158688336610794, "learning_rate": 9.57394378802934e-05, "loss": 0.0007734425598755479, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00077, "step": 94, "tokens/total": 2845600, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 42883}
-{"epoch": 0.37109375, "grad_norm": 0.2428545504808426, "learning_rate": 9.56152962916e-05, "loss": 0.003534857416525483, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00354, "step": 95, "tokens/total": 2875952, "tokens/train_per_sec_per_gpu": 38.65, "tokens/trainable": 43381}
-{"epoch": 0.375, "grad_norm": 0.05061683431267738, "learning_rate": 9.548946453464296e-05, "loss": 0.00045320752542465925, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00045, "step": 96, "tokens/total": 2906288, "tokens/train_per_sec_per_gpu": 32.95, "tokens/trainable": 43824}
-{"epoch": 0.37890625, "grad_norm": 0.0691726952791214, "learning_rate": 9.53619478457953e-05, "loss": 0.0004006598028354347, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0004, "step": 97, "tokens/total": 2936752, "tokens/train_per_sec_per_gpu": 31.5, "tokens/trainable": 44255}
-{"epoch": 0.3828125, "grad_norm": 0.20961223542690277, "learning_rate": 9.523275153154695e-05, "loss": 0.0020726853981614113, "memory/device_reserved (GiB)": 35.19, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00207, "step": 98, "tokens/total": 2967072, "tokens/train_per_sec_per_gpu": 37.71, "tokens/trainable": 44739}
-{"epoch": 0.38671875, "grad_norm": 0.07388679683208466, "learning_rate": 9.51018809682839e-05, "loss": 0.00044117189827375114, "memory/device_reserved (GiB)": 35.76, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00044, "step": 99, "tokens/total": 2997664, "tokens/train_per_sec_per_gpu": 36.09, "tokens/trainable": 45202}
-{"epoch": 0.390625, "grad_norm": 0.46414947509765625, "learning_rate": 9.49693416020645e-05, "loss": 0.006180304102599621, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0062, "step": 100, "tokens/total": 3028032, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 45646}
-{"epoch": 0.39453125, "grad_norm": 0.019027478992938995, "learning_rate": 9.483513894839276e-05, "loss": 0.00012531192624010146, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00013, "step": 101, "tokens/total": 3058272, "tokens/train_per_sec_per_gpu": 33.04, "tokens/trainable": 46089}
-{"epoch": 0.3984375, "grad_norm": 3.6436362266540527, "learning_rate": 9.469927859198888e-05, "loss": 0.031218968331813812, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.03171, "step": 102, "tokens/total": 3088848, "tokens/train_per_sec_per_gpu": 32.49, "tokens/trainable": 46547}
-{"epoch": 0.40234375, "grad_norm": 1.2344398498535156, "learning_rate": 9.456176618655689e-05, "loss": 0.011161141097545624, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.01122, "step": 103, "tokens/total": 3119424, "tokens/train_per_sec_per_gpu": 31.9, "tokens/trainable": 47014}
-{"epoch": 0.40625, "grad_norm": 0.7120245099067688, "learning_rate": 9.442260745454927e-05, "loss": 0.002164291450753808, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00217, "step": 104, "tokens/total": 3149696, "tokens/train_per_sec_per_gpu": 37.65, "tokens/trainable": 47503}
-{"epoch": 0.41015625, "grad_norm": 0.07052730768918991, "learning_rate": 9.428180818692884e-05, "loss": 0.0002605066110845655, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00026, "step": 105, "tokens/total": 3180064, "tokens/train_per_sec_per_gpu": 36.26, "tokens/trainable": 47996}
-{"epoch": 0.4140625, "grad_norm": 1.2376325130462646, "learning_rate": 9.413937424292791e-05, "loss": 0.01640205644071102, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.01654, "step": 106, "tokens/total": 3208320, "tokens/train_per_sec_per_gpu": 34.04, "tokens/trainable": 48447}
-{"epoch": 0.41796875, "grad_norm": 0.5694106817245483, "learning_rate": 9.399531154980424e-05, "loss": 0.01000689435750246, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.01006, "step": 107, "tokens/total": 3238416, "tokens/train_per_sec_per_gpu": 30.65, "tokens/trainable": 48889}
-{"epoch": 0.421875, "grad_norm": 0.5011488795280457, "learning_rate": 9.384962610259455e-05, "loss": 0.0022165768314152956, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00222, "step": 108, "tokens/total": 3268784, "tokens/train_per_sec_per_gpu": 37.84, "tokens/trainable": 49360}
-{"epoch": 0.42578125, "grad_norm": 0.6332448124885559, "learning_rate": 9.370232396386494e-05, "loss": 0.006433461792767048, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00645, "step": 109, "tokens/total": 3298912, "tokens/train_per_sec_per_gpu": 36.73, "tokens/trainable": 49827}
-{"epoch": 0.4296875, "grad_norm": 3.6466922760009766, "learning_rate": 9.355341126345868e-05, "loss": 0.025773590430617332, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.02611, "step": 110, "tokens/total": 3329232, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 50267}
-{"epoch": 0.43359375, "grad_norm": 1.0804238319396973, "learning_rate": 9.340289419824107e-05, "loss": 0.013978044502437115, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.01408, "step": 111, "tokens/total": 3359440, "tokens/train_per_sec_per_gpu": 33.32, "tokens/trainable": 50711}
-{"epoch": 0.4375, "grad_norm": 0.3128775656223297, "learning_rate": 9.325077903184159e-05, "loss": 0.008084455505013466, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00812, "step": 112, "tokens/total": 3389472, "tokens/train_per_sec_per_gpu": 31.54, "tokens/trainable": 51157}
-{"epoch": 0.44140625, "grad_norm": 0.5562160611152649, "learning_rate": 9.30970720943932e-05, "loss": 0.012518094852566719, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0126, "step": 113, "tokens/total": 3419920, "tokens/train_per_sec_per_gpu": 32.67, "tokens/trainable": 51613}
-{"epoch": 0.4453125, "grad_norm": 0.37859252095222473, "learning_rate": 9.2941779782269e-05, "loss": 0.011980460025370121, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.01205, "step": 114, "tokens/total": 3450176, "tokens/train_per_sec_per_gpu": 29.04, "tokens/trainable": 52017}
-{"epoch": 0.44921875, "grad_norm": 0.19049690663814545, "learning_rate": 9.278490855781596e-05, "loss": 0.0033194865100085735, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00333, "step": 115, "tokens/total": 3480544, "tokens/train_per_sec_per_gpu": 36.15, "tokens/trainable": 52506}
-{"epoch": 0.453125, "grad_norm": 0.9838894009590149, "learning_rate": 9.262646494908604e-05, "loss": 0.011315119452774525, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.01138, "step": 116, "tokens/total": 3510848, "tokens/train_per_sec_per_gpu": 38.61, "tokens/trainable": 53008}
-{"epoch": 0.45703125, "grad_norm": 1.5077804327011108, "learning_rate": 9.246645554956457e-05, "loss": 0.011926110833883286, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.012, "step": 117, "tokens/total": 3541296, "tokens/train_per_sec_per_gpu": 35.32, "tokens/trainable": 53467}
-{"epoch": 0.4609375, "grad_norm": 0.1702655702829361, "learning_rate": 9.230488701789578e-05, "loss": 0.002886661561205983, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00289, "step": 118, "tokens/total": 3571936, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 53923}
-{"epoch": 0.46484375, "grad_norm": 0.3885493576526642, "learning_rate": 9.214176607760577e-05, "loss": 0.0035083256661891937, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00351, "step": 119, "tokens/total": 3602000, "tokens/train_per_sec_per_gpu": 35.58, "tokens/trainable": 54363}
-{"epoch": 0.46875, "grad_norm": 0.2820606529712677, "learning_rate": 9.197709951682268e-05, "loss": 0.005707860924303532, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00572, "step": 120, "tokens/total": 3632288, "tokens/train_per_sec_per_gpu": 35.2, "tokens/trainable": 54839}
-{"epoch": 0.47265625, "grad_norm": 0.38794127106666565, "learning_rate": 9.181089418799428e-05, "loss": 0.003157487604767084, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00316, "step": 121, "tokens/total": 3662608, "tokens/train_per_sec_per_gpu": 32.58, "tokens/trainable": 55267}
-{"epoch": 0.4765625, "grad_norm": 0.2677665054798126, "learning_rate": 9.164315700760271e-05, "loss": 0.0032300231978297234, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00324, "step": 122, "tokens/total": 3692864, "tokens/train_per_sec_per_gpu": 30.47, "tokens/trainable": 55700}
-{"epoch": 0.48046875, "grad_norm": 0.31003710627555847, "learning_rate": 9.147389495587671e-05, "loss": 0.00425739586353302, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00427, "step": 123, "tokens/total": 3722864, "tokens/train_per_sec_per_gpu": 33.38, "tokens/trainable": 56144}
-{"epoch": 0.484375, "grad_norm": 0.031137565150856972, "learning_rate": 9.130311507650116e-05, "loss": 0.00010005584044847637, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.0001, "step": 124, "tokens/total": 3753056, "tokens/train_per_sec_per_gpu": 33.07, "tokens/trainable": 56589}
-{"epoch": 0.48828125, "grad_norm": 0.7360314130783081, "learning_rate": 9.113082447632394e-05, "loss": 0.004621482454240322, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00463, "step": 125, "tokens/total": 3783312, "tokens/train_per_sec_per_gpu": 36.64, "tokens/trainable": 57081}
-{"epoch": 0.4921875, "grad_norm": 0.05498598888516426, "learning_rate": 9.09570303250602e-05, "loss": 0.0002661112230271101, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00027, "step": 126, "tokens/total": 3813712, "tokens/train_per_sec_per_gpu": 32.79, "tokens/trainable": 57527}
-{"epoch": 0.49609375, "grad_norm": 0.4160194993019104, "learning_rate": 9.078173985499394e-05, "loss": 0.0034656142815947533, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00347, "step": 127, "tokens/total": 3844336, "tokens/train_per_sec_per_gpu": 34.79, "tokens/trainable": 58005}
-{"epoch": 0.5, "grad_norm": 0.44116270542144775, "learning_rate": 9.060496036067713e-05, "loss": 0.0029591654893010855, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.00296, "step": 128, "tokens/total": 3872688, "tokens/train_per_sec_per_gpu": 35.21, "tokens/trainable": 58454}
-{"epoch": 0.50390625, "grad_norm": 0.29427066445350647, "learning_rate": 9.042669919862615e-05, "loss": 0.002471367595717311, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.64, "memory/max_allocated (GiB)": 33.64, "ppl": 1.00247, "step": 129, "tokens/total": 3902640, "tokens/train_per_sec_per_gpu": 31.94, "tokens/trainable": 58896}
-{"epoch": 0.5078125, "grad_norm": 0.024222422391176224, "learning_rate": 9.024696378701557e-05, "loss": 0.00013285802560858428, "memory/device_reserved (GiB)": 34.68, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00013, "step": 130, "tokens/total": 3933104, "tokens/train_per_sec_per_gpu": 35.79, "tokens/trainable": 59356}
-{"epoch": 0.51171875, "grad_norm": 3.2403194904327393, "learning_rate": 9.006576160536948e-05, "loss": 0.013489946722984314, "memory/device_reserved (GiB)": 34.92, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.01358, "step": 131, "tokens/total": 3963184, "tokens/train_per_sec_per_gpu": 32.14, "tokens/trainable": 59792}
-{"epoch": 0.515625, "grad_norm": 0.14770789444446564, "learning_rate": 8.988310019425035e-05, "loss": 0.0009925956837832928, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00099, "step": 132, "tokens/total": 3993648, "tokens/train_per_sec_per_gpu": 35.66, "tokens/trainable": 60266}
-{"epoch": 0.51953125, "grad_norm": 1.2149583101272583, "learning_rate": 8.969898715494506e-05, "loss": 0.005480521358549595, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0055, "step": 133, "tokens/total": 4024000, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 60759}
-{"epoch": 0.5234375, "grad_norm": 0.24377752840518951, "learning_rate": 8.951343014914869e-05, "loss": 0.0011378922499716282, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.00114, "step": 134, "tokens/total": 4052480, "tokens/train_per_sec_per_gpu": 31.71, "tokens/trainable": 61204}
-{"epoch": 0.52734375, "grad_norm": 0.8190546631813049, "learning_rate": 8.932643689864568e-05, "loss": 0.004931692034006119, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00494, "step": 135, "tokens/total": 4082864, "tokens/train_per_sec_per_gpu": 35.88, "tokens/trainable": 61684}
-{"epoch": 0.53125, "grad_norm": 0.7444491982460022, "learning_rate": 8.913801518498845e-05, "loss": 0.00953812524676323, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00958, "step": 136, "tokens/total": 4113040, "tokens/train_per_sec_per_gpu": 38.18, "tokens/trainable": 62159}
-{"epoch": 0.53515625, "grad_norm": 0.017797501757740974, "learning_rate": 8.894817284917364e-05, "loss": 0.00010352014214731753, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0001, "step": 137, "tokens/total": 4143344, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 62649}
-{"epoch": 0.5390625, "grad_norm": 0.14133845269680023, "learning_rate": 8.875691779131569e-05, "loss": 0.0006721644895151258, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00067, "step": 138, "tokens/total": 4173648, "tokens/train_per_sec_per_gpu": 36.58, "tokens/trainable": 63100}
-{"epoch": 0.54296875, "grad_norm": 0.7884437441825867, "learning_rate": 8.856425797031829e-05, "loss": 0.0070602125488221645, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00709, "step": 139, "tokens/total": 4204080, "tokens/train_per_sec_per_gpu": 37.51, "tokens/trainable": 63588}
-{"epoch": 0.546875, "grad_norm": 2.1515400409698486, "learning_rate": 8.837020140354295e-05, "loss": 0.04659099504351616, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.42, "memory/max_allocated (GiB)": 33.42, "ppl": 1.04769, "step": 140, "tokens/total": 4232400, "tokens/train_per_sec_per_gpu": 33.01, "tokens/trainable": 64005}
-{"epoch": 0.55078125, "grad_norm": 0.9182130098342896, "learning_rate": 8.817475616647554e-05, "loss": 0.01849653385579586, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.01867, "step": 141, "tokens/total": 4262624, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 64462}
-{"epoch": 0.5546875, "grad_norm": 0.049818869680166245, "learning_rate": 8.797793039239017e-05, "loss": 0.0007528763962909579, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00075, "step": 142, "tokens/total": 4293168, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 64900}
-{"epoch": 0.55859375, "grad_norm": 0.4962376654148102, "learning_rate": 8.777973227201069e-05, "loss": 0.013024947606027126, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.01311, "step": 143, "tokens/total": 4323248, "tokens/train_per_sec_per_gpu": 37.45, "tokens/trainable": 65379}
-{"epoch": 0.5625, "grad_norm": 0.09980953484773636, "learning_rate": 8.758017005316988e-05, "loss": 0.0013438962632790208, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00134, "step": 144, "tokens/total": 4353344, "tokens/train_per_sec_per_gpu": 36.69, "tokens/trainable": 65845}
-{"epoch": 0.56640625, "grad_norm": 0.2499600648880005, "learning_rate": 8.737925204046629e-05, "loss": 0.005669259466230869, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00569, "step": 145, "tokens/total": 4383584, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 66285}
-{"epoch": 0.5703125, "grad_norm": 0.8080663084983826, "learning_rate": 8.717698659491851e-05, "loss": 0.03145892918109894, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.03196, "step": 146, "tokens/total": 4413856, "tokens/train_per_sec_per_gpu": 31.46, "tokens/trainable": 66736}
-{"epoch": 0.57421875, "grad_norm": 0.29324567317962646, "learning_rate": 8.697338213361735e-05, "loss": 0.008183630183339119, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00822, "step": 147, "tokens/total": 4444112, "tokens/train_per_sec_per_gpu": 35.2, "tokens/trainable": 67150}
-{"epoch": 0.578125, "grad_norm": 0.5702983736991882, "learning_rate": 8.676844712937552e-05, "loss": 0.00848381593823433, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00852, "step": 148, "tokens/total": 4474464, "tokens/train_per_sec_per_gpu": 35.02, "tokens/trainable": 67633}
-{"epoch": 0.58203125, "grad_norm": 0.12609520554542542, "learning_rate": 8.656219011037509e-05, "loss": 0.002346089808270335, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00235, "step": 149, "tokens/total": 4504880, "tokens/train_per_sec_per_gpu": 33.78, "tokens/trainable": 68077}
-{"epoch": 0.5859375, "grad_norm": 0.044832926243543625, "learning_rate": 8.63546196598125e-05, "loss": 0.000866610265802592, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00087, "step": 150, "tokens/total": 4534976, "tokens/train_per_sec_per_gpu": 33.21, "tokens/trainable": 68545}
-{"epoch": 0.58984375, "grad_norm": 0.2726699113845825, "learning_rate": 8.614574441554145e-05, "loss": 0.010688753798604012, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.01075, "step": 151, "tokens/total": 4565088, "tokens/train_per_sec_per_gpu": 30.17, "tokens/trainable": 69011}
-{"epoch": 0.59375, "grad_norm": 0.6334059238433838, "learning_rate": 8.593557306971349e-05, "loss": 0.007179228588938713, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00721, "step": 152, "tokens/total": 4595600, "tokens/train_per_sec_per_gpu": 35.32, "tokens/trainable": 69465}
-{"epoch": 0.59765625, "grad_norm": 0.6207929849624634, "learning_rate": 8.572411436841618e-05, "loss": 0.005558713339269161, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00557, "step": 153, "tokens/total": 4626032, "tokens/train_per_sec_per_gpu": 33.83, "tokens/trainable": 69928}
-{"epoch": 0.6015625, "grad_norm": 0.14352096617221832, "learning_rate": 8.551137711130922e-05, "loss": 0.0009000393329188228, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.0009, "step": 154, "tokens/total": 4656352, "tokens/train_per_sec_per_gpu": 36.64, "tokens/trainable": 70395}
-{"epoch": 0.60546875, "grad_norm": 0.37536460161209106, "learning_rate": 8.529737015125824e-05, "loss": 0.0032662933226674795, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00327, "step": 155, "tokens/total": 4684672, "tokens/train_per_sec_per_gpu": 35.51, "tokens/trainable": 70829}
-{"epoch": 0.609375, "grad_norm": 0.8182935118675232, "learning_rate": 8.508210239396639e-05, "loss": 0.015120145864784718, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.01524, "step": 156, "tokens/total": 4715216, "tokens/train_per_sec_per_gpu": 33.8, "tokens/trainable": 71287}
-{"epoch": 0.61328125, "grad_norm": 0.05925029143691063, "learning_rate": 8.486558279760375e-05, "loss": 0.00020593182125594467, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00021, "step": 157, "tokens/total": 4745376, "tokens/train_per_sec_per_gpu": 34.95, "tokens/trainable": 71749}
-{"epoch": 0.6171875, "grad_norm": 0.5880815982818604, "learning_rate": 8.464782037243449e-05, "loss": 0.011222876608371735, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01129, "step": 158, "tokens/total": 4775776, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 72185}
-{"epoch": 0.62109375, "grad_norm": 0.1880187839269638, "learning_rate": 8.442882418044202e-05, "loss": 0.0014226734638214111, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00142, "step": 159, "tokens/total": 4805856, "tokens/train_per_sec_per_gpu": 30.51, "tokens/trainable": 72604}
-{"epoch": 0.625, "grad_norm": 0.8338136076927185, "learning_rate": 8.420860333495179e-05, "loss": 0.018916597589850426, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0191, "step": 160, "tokens/total": 4836304, "tokens/train_per_sec_per_gpu": 31.41, "tokens/trainable": 73061}
-{"epoch": 0.62890625, "grad_norm": 0.11151792109012604, "learning_rate": 8.398716700025208e-05, "loss": 0.001152656739577651, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00115, "step": 161, "tokens/total": 4866544, "tokens/train_per_sec_per_gpu": 29.97, "tokens/trainable": 73474}
-{"epoch": 0.6328125, "grad_norm": 0.319916695356369, "learning_rate": 8.376452439121266e-05, "loss": 0.0030798939988017082, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00308, "step": 162, "tokens/total": 4897120, "tokens/train_per_sec_per_gpu": 32.61, "tokens/trainable": 73923}
-{"epoch": 0.63671875, "grad_norm": 0.23228555917739868, "learning_rate": 8.354068477290124e-05, "loss": 0.0029990714974701405, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.003, "step": 163, "tokens/total": 4927440, "tokens/train_per_sec_per_gpu": 35.29, "tokens/trainable": 74379}
-{"epoch": 0.640625, "grad_norm": 0.16945457458496094, "learning_rate": 8.331565746019807e-05, "loss": 0.0036612222902476788, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00367, "step": 164, "tokens/total": 4958032, "tokens/train_per_sec_per_gpu": 33.33, "tokens/trainable": 74830}
-{"epoch": 0.64453125, "grad_norm": 0.08089865744113922, "learning_rate": 8.308945181740812e-05, "loss": 0.0011908983578905463, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00119, "step": 165, "tokens/total": 4988368, "tokens/train_per_sec_per_gpu": 31.57, "tokens/trainable": 75262}
-{"epoch": 0.6484375, "grad_norm": 0.06332889199256897, "learning_rate": 8.286207725787153e-05, "loss": 0.0010977290803566575, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.0011, "step": 166, "tokens/total": 5018672, "tokens/train_per_sec_per_gpu": 31.75, "tokens/trainable": 75703}
-{"epoch": 0.65234375, "grad_norm": 0.058639127761125565, "learning_rate": 8.263354324357182e-05, "loss": 0.000854893063660711, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00086, "step": 167, "tokens/total": 5049344, "tokens/train_per_sec_per_gpu": 31.26, "tokens/trainable": 76164}
-{"epoch": 0.65625, "grad_norm": 0.19198745489120483, "learning_rate": 8.240385928474219e-05, "loss": 0.0024191238917410374, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00242, "step": 168, "tokens/total": 5079856, "tokens/train_per_sec_per_gpu": 34.42, "tokens/trainable": 76623}
-{"epoch": 0.66015625, "grad_norm": 0.033948056399822235, "learning_rate": 8.217303493946967e-05, "loss": 0.000675962888635695, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00068, "step": 169, "tokens/total": 5110192, "tokens/train_per_sec_per_gpu": 34.34, "tokens/trainable": 77047}
-{"epoch": 0.6640625, "grad_norm": 0.10515601187944412, "learning_rate": 8.194107981329746e-05, "loss": 0.0006985474610701203, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0007, "step": 170, "tokens/total": 5140592, "tokens/train_per_sec_per_gpu": 33.98, "tokens/trainable": 77496}
-{"epoch": 0.66796875, "grad_norm": 0.01597985066473484, "learning_rate": 8.170800355882518e-05, "loss": 0.0002490747720003128, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00025, "step": 171, "tokens/total": 5170848, "tokens/train_per_sec_per_gpu": 33.13, "tokens/trainable": 77956}
-{"epoch": 0.671875, "grad_norm": 0.28580594062805176, "learning_rate": 8.147381587530713e-05, "loss": 0.002719157375395298, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00272, "step": 172, "tokens/total": 5201280, "tokens/train_per_sec_per_gpu": 29.75, "tokens/trainable": 78410}
-{"epoch": 0.67578125, "grad_norm": 0.027172744274139404, "learning_rate": 8.123852650824877e-05, "loss": 0.0005204043700359762, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00052, "step": 173, "tokens/total": 5231712, "tokens/train_per_sec_per_gpu": 33.04, "tokens/trainable": 78849}
-{"epoch": 0.6796875, "grad_norm": 2.8796133995056152, "learning_rate": 8.100214524900103e-05, "loss": 0.015293785370886326, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.62, "memory/max_allocated (GiB)": 33.62, "ppl": 1.01541, "step": 174, "tokens/total": 5261440, "tokens/train_per_sec_per_gpu": 29.53, "tokens/trainable": 79266}
-{"epoch": 0.68359375, "grad_norm": 0.1005098894238472, "learning_rate": 8.076468193435301e-05, "loss": 0.0010334225371479988, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00103, "step": 175, "tokens/total": 5291728, "tokens/train_per_sec_per_gpu": 32.42, "tokens/trainable": 79736}
-{"epoch": 0.6875, "grad_norm": 0.6506177186965942, "learning_rate": 8.052614644612253e-05, "loss": 0.0038155291695147753, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00382, "step": 176, "tokens/total": 5322240, "tokens/train_per_sec_per_gpu": 36.12, "tokens/trainable": 80214}
-{"epoch": 0.69140625, "grad_norm": 0.19899244606494904, "learning_rate": 8.028654871074489e-05, "loss": 0.0020266459323465824, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00203, "step": 177, "tokens/total": 5352576, "tokens/train_per_sec_per_gpu": 32.51, "tokens/trainable": 80661}
-{"epoch": 0.6953125, "grad_norm": 0.009929690510034561, "learning_rate": 8.004589869885986e-05, "loss": 0.00014211784582585096, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00014, "step": 178, "tokens/total": 5382784, "tokens/train_per_sec_per_gpu": 34.81, "tokens/trainable": 81129}
-{"epoch": 0.69921875, "grad_norm": 0.5209382176399231, "learning_rate": 7.980420642489674e-05, "loss": 0.023224791511893272, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0235, "step": 179, "tokens/total": 5412784, "tokens/train_per_sec_per_gpu": 32.95, "tokens/trainable": 81576}
-{"epoch": 0.703125, "grad_norm": 0.02280755713582039, "learning_rate": 7.95614819466576e-05, "loss": 0.0002463866549078375, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00025, "step": 180, "tokens/total": 5442960, "tokens/train_per_sec_per_gpu": 34.49, "tokens/trainable": 82049}
-{"epoch": 0.70703125, "grad_norm": 0.3769514262676239, "learning_rate": 7.931773536489872e-05, "loss": 0.01167471706867218, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01174, "step": 181, "tokens/total": 5473216, "tokens/train_per_sec_per_gpu": 38.75, "tokens/trainable": 82566}
-{"epoch": 0.7109375, "grad_norm": 0.060761548578739166, "learning_rate": 7.907297682291035e-05, "loss": 0.000216186250327155, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00022, "step": 182, "tokens/total": 5503568, "tokens/train_per_sec_per_gpu": 33.83, "tokens/trainable": 83054}
-{"epoch": 0.71484375, "grad_norm": 0.03373813256621361, "learning_rate": 7.882721650609442e-05, "loss": 0.00041355923167429864, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00041, "step": 183, "tokens/total": 5533760, "tokens/train_per_sec_per_gpu": 32.76, "tokens/trainable": 83501}
-{"epoch": 0.71875, "grad_norm": 0.036661963909864426, "learning_rate": 7.85804646415409e-05, "loss": 0.00041001036879606545, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00041, "step": 184, "tokens/total": 5564240, "tokens/train_per_sec_per_gpu": 35.45, "tokens/trainable": 84007}
-{"epoch": 0.72265625, "grad_norm": 0.03234419599175453, "learning_rate": 7.833273149760207e-05, "loss": 0.0003714286140166223, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00037, "step": 185, "tokens/total": 5594448, "tokens/train_per_sec_per_gpu": 35.39, "tokens/trainable": 84491}
-{"epoch": 0.7265625, "grad_norm": 0.03076508454978466, "learning_rate": 7.808402738346527e-05, "loss": 0.0003759118844754994, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00038, "step": 186, "tokens/total": 5624816, "tokens/train_per_sec_per_gpu": 33.33, "tokens/trainable": 84952}
-{"epoch": 0.73046875, "grad_norm": 0.6632063388824463, "learning_rate": 7.783436264872382e-05, "loss": 0.002438098657876253, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00244, "step": 187, "tokens/total": 5655104, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 85400}
-{"epoch": 0.734375, "grad_norm": 0.2651624083518982, "learning_rate": 7.758374768294647e-05, "loss": 0.002341378014534712, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00234, "step": 188, "tokens/total": 5685360, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 85889}
-{"epoch": 0.73828125, "grad_norm": 0.305477112531662, "learning_rate": 7.733219291524489e-05, "loss": 0.0016374983824789524, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00164, "step": 189, "tokens/total": 5715648, "tokens/train_per_sec_per_gpu": 36.11, "tokens/trainable": 86396}
-{"epoch": 0.7421875, "grad_norm": 0.2344018816947937, "learning_rate": 7.707970881383977e-05, "loss": 0.002069709589704871, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00207, "step": 190, "tokens/total": 5746240, "tokens/train_per_sec_per_gpu": 33.75, "tokens/trainable": 86867}
-{"epoch": 0.74609375, "grad_norm": 0.09677010774612427, "learning_rate": 7.682630588562518e-05, "loss": 0.0006176180904731154, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00062, "step": 191, "tokens/total": 5776816, "tokens/train_per_sec_per_gpu": 36.03, "tokens/trainable": 87314}
-{"epoch": 0.75, "grad_norm": 0.05828038975596428, "learning_rate": 7.657199467573129e-05, "loss": 0.0004249829798936844, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00043, "step": 192, "tokens/total": 5807248, "tokens/train_per_sec_per_gpu": 30.29, "tokens/trainable": 87736}
-{"epoch": 0.75390625, "grad_norm": 0.09240972250699997, "learning_rate": 7.631678576708561e-05, "loss": 0.0003209684509783983, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00032, "step": 193, "tokens/total": 5837680, "tokens/train_per_sec_per_gpu": 34.95, "tokens/trainable": 88219}
-{"epoch": 0.7578125, "grad_norm": 0.00590590201318264, "learning_rate": 7.606068977997255e-05, "loss": 7.823687337804586e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00008, "step": 194, "tokens/total": 5868000, "tokens/train_per_sec_per_gpu": 36.11, "tokens/trainable": 88699}
-{"epoch": 0.76171875, "grad_norm": 0.45261114835739136, "learning_rate": 7.580371737159148e-05, "loss": 0.0019616044592112303, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.38, "memory/max_allocated (GiB)": 33.38, "ppl": 1.00196, "step": 195, "tokens/total": 5896224, "tokens/train_per_sec_per_gpu": 30.36, "tokens/trainable": 89132}
-{"epoch": 0.765625, "grad_norm": 0.03762947395443916, "learning_rate": 7.554587923561324e-05, "loss": 0.0004910373827442527, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00049, "step": 196, "tokens/total": 5926624, "tokens/train_per_sec_per_gpu": 37.21, "tokens/trainable": 89594}
-{"epoch": 0.76953125, "grad_norm": 0.37914252281188965, "learning_rate": 7.528718610173511e-05, "loss": 0.005961633287370205, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.34, "memory/max_allocated (GiB)": 33.34, "ppl": 1.00598, "step": 197, "tokens/total": 5954944, "tokens/train_per_sec_per_gpu": 37.98, "tokens/trainable": 90042}
-{"epoch": 0.7734375, "grad_norm": 0.061283551156520844, "learning_rate": 7.502764873523431e-05, "loss": 0.0004046210669912398, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0004, "step": 198, "tokens/total": 5985200, "tokens/train_per_sec_per_gpu": 34.15, "tokens/trainable": 90503}
-{"epoch": 0.77734375, "grad_norm": 0.012014869600534439, "learning_rate": 7.476727793652011e-05, "loss": 0.00018992810510098934, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 34.0, "memory/max_allocated (GiB)": 34.0, "ppl": 1.00019, "step": 199, "tokens/total": 6015952, "tokens/train_per_sec_per_gpu": 36.46, "tokens/trainable": 90982}
-{"epoch": 0.78125, "grad_norm": 0.039138417690992355, "learning_rate": 7.450608454068415e-05, "loss": 0.0003556829469744116, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00036, "step": 200, "tokens/total": 6046208, "tokens/train_per_sec_per_gpu": 39.22, "tokens/trainable": 91466}
-{"epoch": 0.78515625, "grad_norm": 0.0454350970685482, "learning_rate": 7.424407941704987e-05, "loss": 0.00024555803975090384, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00025, "step": 201, "tokens/total": 6076208, "tokens/train_per_sec_per_gpu": 34.18, "tokens/trainable": 91912}
-{"epoch": 0.7890625, "grad_norm": 0.6254648566246033, "learning_rate": 7.398127346871986e-05, "loss": 0.009876105934381485, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00993, "step": 202, "tokens/total": 6106560, "tokens/train_per_sec_per_gpu": 37.79, "tokens/trainable": 92385}
-{"epoch": 0.79296875, "grad_norm": 0.7776591181755066, "learning_rate": 7.371767763212238e-05, "loss": 0.011619904078543186, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.01169, "step": 203, "tokens/total": 6136640, "tokens/train_per_sec_per_gpu": 36.48, "tokens/trainable": 92842}
-{"epoch": 0.796875, "grad_norm": 0.07254572212696075, "learning_rate": 7.345330287655617e-05, "loss": 0.0005007775034755468, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.0005, "step": 204, "tokens/total": 6167184, "tokens/train_per_sec_per_gpu": 33.89, "tokens/trainable": 93323}
-{"epoch": 0.80078125, "grad_norm": 0.6082178950309753, "learning_rate": 7.31881602037339e-05, "loss": 0.007061033509671688, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00709, "step": 205, "tokens/total": 6197376, "tokens/train_per_sec_per_gpu": 34.75, "tokens/trainable": 93756}
-{"epoch": 0.8046875, "grad_norm": 0.4930607080459595, "learning_rate": 7.29222606473245e-05, "loss": 0.006006345152854919, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00602, "step": 206, "tokens/total": 6227600, "tokens/train_per_sec_per_gpu": 37.47, "tokens/trainable": 94235}
-{"epoch": 0.80859375, "grad_norm": 0.06813386082649231, "learning_rate": 7.265561527249383e-05, "loss": 0.0006858796114102006, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00069, "step": 207, "tokens/total": 6257760, "tokens/train_per_sec_per_gpu": 36.48, "tokens/trainable": 94733}
-{"epoch": 0.8125, "grad_norm": 0.06703829765319824, "learning_rate": 7.238823517544436e-05, "loss": 0.001039305585436523, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00104, "step": 208, "tokens/total": 6288048, "tokens/train_per_sec_per_gpu": 33.68, "tokens/trainable": 95218}
-{"epoch": 0.81640625, "grad_norm": 0.2953227460384369, "learning_rate": 7.212013148295333e-05, "loss": 0.0037906011566519737, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.0038, "step": 209, "tokens/total": 6318688, "tokens/train_per_sec_per_gpu": 29.17, "tokens/trainable": 95619}
-{"epoch": 0.8203125, "grad_norm": 0.32579171657562256, "learning_rate": 7.185131535190975e-05, "loss": 0.002393842674791813, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0024, "step": 210, "tokens/total": 6348944, "tokens/train_per_sec_per_gpu": 35.32, "tokens/trainable": 96076}
-{"epoch": 0.82421875, "grad_norm": 0.4734557271003723, "learning_rate": 7.158179796885005e-05, "loss": 0.006001632194966078, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00602, "step": 211, "tokens/total": 6378960, "tokens/train_per_sec_per_gpu": 35.28, "tokens/trainable": 96505}
-{"epoch": 0.828125, "grad_norm": 0.15732432901859283, "learning_rate": 7.131159054949273e-05, "loss": 0.003803959349170327, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00381, "step": 212, "tokens/total": 6409472, "tokens/train_per_sec_per_gpu": 32.57, "tokens/trainable": 96930}
-{"epoch": 0.83203125, "grad_norm": 0.18325994908809662, "learning_rate": 7.104070433827139e-05, "loss": 0.005038415547460318, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00505, "step": 213, "tokens/total": 6439904, "tokens/train_per_sec_per_gpu": 32.65, "tokens/trainable": 97361}
-{"epoch": 0.8359375, "grad_norm": 0.17816093564033508, "learning_rate": 7.076915060786705e-05, "loss": 0.0020862463861703873, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00209, "step": 214, "tokens/total": 6470176, "tokens/train_per_sec_per_gpu": 36.26, "tokens/trainable": 97858}
-{"epoch": 0.83984375, "grad_norm": 0.2746942937374115, "learning_rate": 7.049694065873882e-05, "loss": 0.005029214546084404, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00504, "step": 215, "tokens/total": 6500336, "tokens/train_per_sec_per_gpu": 37.41, "tokens/trainable": 98308}
-{"epoch": 0.84375, "grad_norm": 0.4508717358112335, "learning_rate": 7.022408581865382e-05, "loss": 0.0054893530905246735, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0055, "step": 216, "tokens/total": 6530688, "tokens/train_per_sec_per_gpu": 36.34, "tokens/trainable": 98781}
-{"epoch": 0.84765625, "grad_norm": 0.23194828629493713, "learning_rate": 6.99505974422157e-05, "loss": 0.0034187212586402893, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00342, "step": 217, "tokens/total": 6561040, "tokens/train_per_sec_per_gpu": 36.67, "tokens/trainable": 99266}
-{"epoch": 0.8515625, "grad_norm": 0.06469441950321198, "learning_rate": 6.967648691039213e-05, "loss": 0.0009051199886016548, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00091, "step": 218, "tokens/total": 6589392, "tokens/train_per_sec_per_gpu": 39.6, "tokens/trainable": 99731}
-{"epoch": 0.85546875, "grad_norm": 0.06683141738176346, "learning_rate": 6.940176563004123e-05, "loss": 0.0011187575291842222, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00112, "step": 219, "tokens/total": 6619824, "tokens/train_per_sec_per_gpu": 31.66, "tokens/trainable": 100164}
-{"epoch": 0.859375, "grad_norm": 0.1525852382183075, "learning_rate": 6.912644503343682e-05, "loss": 0.0017743443604558706, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00178, "step": 220, "tokens/total": 6650224, "tokens/train_per_sec_per_gpu": 36.17, "tokens/trainable": 100620}
-{"epoch": 0.86328125, "grad_norm": 0.8795211911201477, "learning_rate": 6.885053657779273e-05, "loss": 0.006786257494240999, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00681, "step": 221, "tokens/total": 6680384, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 101080}
-{"epoch": 0.8671875, "grad_norm": 0.19531063735485077, "learning_rate": 6.857405174478604e-05, "loss": 0.0014687087386846542, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00147, "step": 222, "tokens/total": 6710896, "tokens/train_per_sec_per_gpu": 37.76, "tokens/trainable": 101546}
-{"epoch": 0.87109375, "grad_norm": 0.011134379543364048, "learning_rate": 6.82970020400792e-05, "loss": 0.00012406683526933193, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00012, "step": 223, "tokens/total": 6741280, "tokens/train_per_sec_per_gpu": 31.51, "tokens/trainable": 101987}
-{"epoch": 0.875, "grad_norm": 0.41609230637550354, "learning_rate": 6.801939899284132e-05, "loss": 0.00440242001786828, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00441, "step": 224, "tokens/total": 6771488, "tokens/train_per_sec_per_gpu": 35.45, "tokens/trainable": 102435}
-{"epoch": 0.87890625, "grad_norm": 0.23292526602745056, "learning_rate": 6.774125415526827e-05, "loss": 0.000571364420466125, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00057, "step": 225, "tokens/total": 6801744, "tokens/train_per_sec_per_gpu": 28.91, "tokens/trainable": 102843}
-{"epoch": 0.8828125, "grad_norm": 0.016833841800689697, "learning_rate": 6.746257910210214e-05, "loss": 0.00012277837959118187, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00012, "step": 226, "tokens/total": 6832432, "tokens/train_per_sec_per_gpu": 37.37, "tokens/trainable": 103342}
-{"epoch": 0.88671875, "grad_norm": 0.35366129875183105, "learning_rate": 6.718338543014937e-05, "loss": 0.004477445501834154, "memory/device_reserved (GiB)": 35.71, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00449, "step": 227, "tokens/total": 6862448, "tokens/train_per_sec_per_gpu": 40.68, "tokens/trainable": 103844}
-{"epoch": 0.890625, "grad_norm": 0.4743596017360687, "learning_rate": 6.69036847577983e-05, "loss": 0.002162824384868145, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00217, "step": 228, "tokens/total": 6892832, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 104307}
-{"epoch": 0.89453125, "grad_norm": 0.4887899160385132, "learning_rate": 6.662348872453553e-05, "loss": 0.0048948717303574085, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00491, "step": 229, "tokens/total": 6922832, "tokens/train_per_sec_per_gpu": 33.43, "tokens/trainable": 104772}
-{"epoch": 0.8984375, "grad_norm": 0.017348933964967728, "learning_rate": 6.63428089904618e-05, "loss": 7.635008660145104e-05, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00008, "step": 230, "tokens/total": 6953104, "tokens/train_per_sec_per_gpu": 36.83, "tokens/trainable": 105240}
-{"epoch": 0.90234375, "grad_norm": 0.3773981034755707, "learning_rate": 6.60616572358065e-05, "loss": 0.003427914110943675, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00343, "step": 231, "tokens/total": 6983440, "tokens/train_per_sec_per_gpu": 33.75, "tokens/trainable": 105703}
-{"epoch": 0.90625, "grad_norm": 0.0006717262440361083, "learning_rate": 6.578004516044172e-05, "loss": 1.8351973267272115e-05, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 232, "tokens/total": 7013840, "tokens/train_per_sec_per_gpu": 33.01, "tokens/trainable": 106179}
-{"epoch": 0.91015625, "grad_norm": 0.0018952380632981658, "learning_rate": 6.549798448339548e-05, "loss": 2.8159589419374242e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00003, "step": 233, "tokens/total": 7044240, "tokens/train_per_sec_per_gpu": 37.96, "tokens/trainable": 106683}
-{"epoch": 0.9140625, "grad_norm": 0.04159875586628914, "learning_rate": 6.521548694236384e-05, "loss": 0.0001036152389133349, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0001, "step": 234, "tokens/total": 7074224, "tokens/train_per_sec_per_gpu": 33.82, "tokens/trainable": 107121}
-{"epoch": 0.91796875, "grad_norm": 0.005844940431416035, "learning_rate": 6.493256429322259e-05, "loss": 5.149145727045834e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00005, "step": 235, "tokens/total": 7104672, "tokens/train_per_sec_per_gpu": 32.68, "tokens/trainable": 107561}
-{"epoch": 0.921875, "grad_norm": 0.0022666200529783964, "learning_rate": 6.464922830953799e-05, "loss": 2.886037327698432e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00003, "step": 236, "tokens/total": 7134784, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 108018}
-{"epoch": 0.92578125, "grad_norm": 0.27136701345443726, "learning_rate": 6.436549078207688e-05, "loss": 0.0009024653700180352, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.0009, "step": 237, "tokens/total": 7164960, "tokens/train_per_sec_per_gpu": 33.28, "tokens/trainable": 108446}
-{"epoch": 0.9296875, "grad_norm": 0.13347801566123962, "learning_rate": 6.408136351831592e-05, "loss": 0.00046705722343176603, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00047, "step": 238, "tokens/total": 7195344, "tokens/train_per_sec_per_gpu": 29.14, "tokens/trainable": 108866}
-{"epoch": 0.93359375, "grad_norm": 0.10074340552091599, "learning_rate": 6.379685834195036e-05, "loss": 0.0006794088403694332, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00068, "step": 239, "tokens/total": 7225680, "tokens/train_per_sec_per_gpu": 33.8, "tokens/trainable": 109316}
-{"epoch": 0.9375, "grad_norm": 0.12028707563877106, "learning_rate": 6.351198709240186e-05, "loss": 0.000513301754835993, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00051, "step": 240, "tokens/total": 7256112, "tokens/train_per_sec_per_gpu": 33.26, "tokens/trainable": 109787}
-{"epoch": 0.94140625, "grad_norm": 0.0062753986567258835, "learning_rate": 6.32267616243259e-05, "loss": 3.0990675440989435e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00003, "step": 241, "tokens/total": 7286384, "tokens/train_per_sec_per_gpu": 38.3, "tokens/trainable": 110246}
-{"epoch": 0.9453125, "grad_norm": 0.05608534440398216, "learning_rate": 6.294119380711849e-05, "loss": 0.00039500967250205576, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0004, "step": 242, "tokens/total": 7316688, "tokens/train_per_sec_per_gpu": 29.14, "tokens/trainable": 110667}
-{"epoch": 0.94921875, "grad_norm": 0.05409131199121475, "learning_rate": 6.265529552442209e-05, "loss": 0.0002061743725789711, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00021, "step": 243, "tokens/total": 7346992, "tokens/train_per_sec_per_gpu": 38.09, "tokens/trainable": 111154}
-{"epoch": 0.953125, "grad_norm": 0.0556725338101387, "learning_rate": 6.236907867363127e-05, "loss": 0.0005747157847508788, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00057, "step": 244, "tokens/total": 7377632, "tokens/train_per_sec_per_gpu": 32.97, "tokens/trainable": 111611}
-{"epoch": 0.95703125, "grad_norm": 0.16125747561454773, "learning_rate": 6.208255516539749e-05, "loss": 0.0008788461564108729, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00088, "step": 245, "tokens/total": 7407872, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 112102}
-{"epoch": 0.9609375, "grad_norm": 0.00565892830491066, "learning_rate": 6.179573692313344e-05, "loss": 4.069084388902411e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00004, "step": 246, "tokens/total": 7438448, "tokens/train_per_sec_per_gpu": 33.16, "tokens/trainable": 112569}
-{"epoch": 0.96484375, "grad_norm": 0.1774720549583435, "learning_rate": 6.150863588251694e-05, "loss": 0.0005083430442027748, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00051, "step": 247, "tokens/total": 7466768, "tokens/train_per_sec_per_gpu": 37.68, "tokens/trainable": 113023}
-{"epoch": 0.96875, "grad_norm": 0.0028223118279129267, "learning_rate": 6.122126399099419e-05, "loss": 3.470741648925468e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00003, "step": 248, "tokens/total": 7497360, "tokens/train_per_sec_per_gpu": 35.17, "tokens/trainable": 113464}
-{"epoch": 0.97265625, "grad_norm": 0.0367576889693737, "learning_rate": 6.0933633207282615e-05, "loss": 8.433192851953208e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00008, "step": 249, "tokens/total": 7527504, "tokens/train_per_sec_per_gpu": 32.65, "tokens/trainable": 113908}
-{"epoch": 0.9765625, "grad_norm": 0.11235758662223816, "learning_rate": 6.064575550087316e-05, "loss": 0.0006511391839012504, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00065, "step": 250, "tokens/total": 7555808, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 114348}
-{"epoch": 0.98046875, "grad_norm": 0.049232710152864456, "learning_rate": 6.0357642851532245e-05, "loss": 0.00024149491218850017, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00024, "step": 251, "tokens/total": 7586320, "tokens/train_per_sec_per_gpu": 35.15, "tokens/trainable": 114830}
-{"epoch": 0.984375, "grad_norm": 0.4833732545375824, "learning_rate": 6.0069307248803294e-05, "loss": 0.005053690634667873, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00507, "step": 252, "tokens/total": 7616576, "tokens/train_per_sec_per_gpu": 35.1, "tokens/trainable": 115278}
-{"epoch": 0.98828125, "grad_norm": 0.013355270028114319, "learning_rate": 5.9780760691507635e-05, "loss": 8.867261931300163e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00009, "step": 253, "tokens/total": 7647152, "tokens/train_per_sec_per_gpu": 32.21, "tokens/trainable": 115699}
-{"epoch": 0.9921875, "grad_norm": 0.09679444134235382, "learning_rate": 5.9492015187245334e-05, "loss": 0.0006353833014145494, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00064, "step": 254, "tokens/total": 7677584, "tokens/train_per_sec_per_gpu": 30.73, "tokens/trainable": 116133}
-{"epoch": 0.99609375, "grad_norm": 0.5139889717102051, "learning_rate": 5.920308275189541e-05, "loss": 0.001129151787608862, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00113, "step": 255, "tokens/total": 7708016, "tokens/train_per_sec_per_gpu": 36.73, "tokens/trainable": 116591}
-{"epoch": 1.0, "grad_norm": 0.00027604683418758214, "learning_rate": 5.8913975409115874e-05, "loss": 7.1114727688836865e-06, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00001, "step": 256, "tokens/total": 7738608, "tokens/train_per_sec_per_gpu": 32.04, "tokens/trainable": 117039}
-{"epoch": 1.00390625, "grad_norm": 0.005449575372040272, "learning_rate": 5.8624705189843395e-05, "loss": 2.958515688078478e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 257, "tokens/total": 7768944, "tokens/train_per_sec_per_gpu": 32.24, "tokens/trainable": 117495}
-{"epoch": 1.0078125, "grad_norm": 0.003980543464422226, "learning_rate": 5.833528413179249e-05, "loss": 2.851396129699424e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00003, "step": 258, "tokens/total": 7799232, "tokens/train_per_sec_per_gpu": 30.44, "tokens/trainable": 117925}
-{"epoch": 1.01171875, "grad_norm": 0.001549297128804028, "learning_rate": 5.80457242789548e-05, "loss": 1.1607408850977663e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 259, "tokens/total": 7829568, "tokens/train_per_sec_per_gpu": 31.58, "tokens/trainable": 118373}
-{"epoch": 1.015625, "grad_norm": 0.00039364350959658623, "learning_rate": 5.77560376810977e-05, "loss": 9.399981536262203e-06, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 260, "tokens/total": 7860000, "tokens/train_per_sec_per_gpu": 31.46, "tokens/trainable": 118816}
-{"epoch": 1.01953125, "grad_norm": 0.024472124874591827, "learning_rate": 5.7466236393263005e-05, "loss": 0.00013395686983130872, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00013, "step": 261, "tokens/total": 7890464, "tokens/train_per_sec_per_gpu": 33.8, "tokens/trainable": 119263}
-{"epoch": 1.0234375, "grad_norm": 0.3619483709335327, "learning_rate": 5.717633247526522e-05, "loss": 0.0023931623436510563, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0024, "step": 262, "tokens/total": 7920944, "tokens/train_per_sec_per_gpu": 34.66, "tokens/trainable": 119772}
-{"epoch": 1.02734375, "grad_norm": 0.17659051716327667, "learning_rate": 5.688633799118971e-05, "loss": 0.0005250183749012649, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00053, "step": 263, "tokens/total": 7951296, "tokens/train_per_sec_per_gpu": 33.78, "tokens/trainable": 120269}
-{"epoch": 1.03125, "grad_norm": 0.000787933764513582, "learning_rate": 5.659626500889066e-05, "loss": 1.1148193152621388e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 264, "tokens/total": 7981856, "tokens/train_per_sec_per_gpu": 36.96, "tokens/trainable": 120746}
-{"epoch": 1.03515625, "grad_norm": 0.00879898015409708, "learning_rate": 5.6306125599488905e-05, "loss": 4.70993691124022e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00005, "step": 265, "tokens/total": 8012192, "tokens/train_per_sec_per_gpu": 38.22, "tokens/trainable": 121204}
-{"epoch": 1.0390625, "grad_norm": 0.0005215261480771005, "learning_rate": 5.601593183686955e-05, "loss": 8.791243999439757e-06, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 266, "tokens/total": 8042736, "tokens/train_per_sec_per_gpu": 31.23, "tokens/trainable": 121670}
-{"epoch": 1.04296875, "grad_norm": 0.1482085883617401, "learning_rate": 5.572569579717961e-05, "loss": 0.000805735879112035, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00081, "step": 267, "tokens/total": 8073024, "tokens/train_per_sec_per_gpu": 40.04, "tokens/trainable": 122183}
-{"epoch": 1.046875, "grad_norm": 0.08614944666624069, "learning_rate": 5.543542955832538e-05, "loss": 0.00032304273918271065, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00032, "step": 268, "tokens/total": 8103280, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 122645}
-{"epoch": 1.05078125, "grad_norm": 0.0005171536467969418, "learning_rate": 5.514514519946986e-05, "loss": 1.1340615856170189e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00001, "step": 269, "tokens/total": 8133440, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 123105}
-{"epoch": 1.0546875, "grad_norm": 0.00104080094024539, "learning_rate": 5.485485480053015e-05, "loss": 1.1951466149184853e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00001, "step": 270, "tokens/total": 8163920, "tokens/train_per_sec_per_gpu": 36.18, "tokens/trainable": 123595}
-{"epoch": 1.05859375, "grad_norm": 0.005696228239685297, "learning_rate": 5.4564570441674645e-05, "loss": 1.725989204715006e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 271, "tokens/total": 8194032, "tokens/train_per_sec_per_gpu": 37.59, "tokens/trainable": 124077}
-{"epoch": 1.0625, "grad_norm": 0.0005264461506158113, "learning_rate": 5.42743042028204e-05, "loss": 1.0610991921566892e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 272, "tokens/total": 8224320, "tokens/train_per_sec_per_gpu": 34.65, "tokens/trainable": 124547}
-{"epoch": 1.06640625, "grad_norm": 0.08169055730104446, "learning_rate": 5.3984068163130464e-05, "loss": 0.0003627589321695268, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00036, "step": 273, "tokens/total": 8254368, "tokens/train_per_sec_per_gpu": 33.95, "tokens/trainable": 124977}
-{"epoch": 1.0703125, "grad_norm": 0.0005202546599321067, "learning_rate": 5.369387440051111e-05, "loss": 1.123354013543576e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00001, "step": 274, "tokens/total": 8284944, "tokens/train_per_sec_per_gpu": 34.51, "tokens/trainable": 125438}
-{"epoch": 1.07421875, "grad_norm": 0.0031104006338864565, "learning_rate": 5.340373499110935e-05, "loss": 1.4764596926397644e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00001, "step": 275, "tokens/total": 8315584, "tokens/train_per_sec_per_gpu": 31.47, "tokens/trainable": 125889}
-{"epoch": 1.078125, "grad_norm": 0.01905728504061699, "learning_rate": 5.3113662008810304e-05, "loss": 9.163993672700599e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00009, "step": 276, "tokens/total": 8345952, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 126363}
-{"epoch": 1.08203125, "grad_norm": 0.0020019100047647953, "learning_rate": 5.282366752473479e-05, "loss": 1.796903961803764e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 277, "tokens/total": 8376416, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 126828}
-{"epoch": 1.0859375, "grad_norm": 0.0011239969171583652, "learning_rate": 5.2533763606737005e-05, "loss": 1.2794653230230324e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00001, "step": 278, "tokens/total": 8406640, "tokens/train_per_sec_per_gpu": 38.16, "tokens/trainable": 127331}
-{"epoch": 1.08984375, "grad_norm": 0.07841381430625916, "learning_rate": 5.224396231890232e-05, "loss": 0.0005632906104438007, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00056, "step": 279, "tokens/total": 8436784, "tokens/train_per_sec_per_gpu": 34.34, "tokens/trainable": 127798}
-{"epoch": 1.09375, "grad_norm": 0.0066568912006914616, "learning_rate": 5.195427572104522e-05, "loss": 4.4691078073810786e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00004, "step": 280, "tokens/total": 8467152, "tokens/train_per_sec_per_gpu": 32.73, "tokens/trainable": 128247}
-{"epoch": 1.09765625, "grad_norm": 0.0006084765191189945, "learning_rate": 5.166471586820751e-05, "loss": 9.42349561228184e-06, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 281, "tokens/total": 8497424, "tokens/train_per_sec_per_gpu": 36.17, "tokens/trainable": 128729}
-{"epoch": 1.1015625, "grad_norm": 0.4024369716644287, "learning_rate": 5.1375294810156615e-05, "loss": 0.002225687028840184, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00223, "step": 282, "tokens/total": 8527616, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 129203}
-{"epoch": 1.10546875, "grad_norm": 0.0008598315180279315, "learning_rate": 5.1086024590884144e-05, "loss": 9.10060407477431e-06, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 283, "tokens/total": 8557984, "tokens/train_per_sec_per_gpu": 32.91, "tokens/trainable": 129653}
-{"epoch": 1.109375, "grad_norm": 0.13851912319660187, "learning_rate": 5.079691724810461e-05, "loss": 0.00045336864423006773, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00045, "step": 284, "tokens/total": 8588384, "tokens/train_per_sec_per_gpu": 32.56, "tokens/trainable": 130100}
-{"epoch": 1.11328125, "grad_norm": 0.08453445136547089, "learning_rate": 5.0507984812754684e-05, "loss": 0.0002941025304608047, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00029, "step": 285, "tokens/total": 8618768, "tokens/train_per_sec_per_gpu": 30.44, "tokens/trainable": 130543}
-{"epoch": 1.1171875, "grad_norm": 0.018670005723834038, "learning_rate": 5.021923930849237e-05, "loss": 0.00010185746214119717, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0001, "step": 286, "tokens/total": 8648976, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 131000}
-{"epoch": 1.12109375, "grad_norm": 0.0015903875464573503, "learning_rate": 4.99306927511967e-05, "loss": 1.1978466318396386e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00001, "step": 287, "tokens/total": 8679152, "tokens/train_per_sec_per_gpu": 33.96, "tokens/trainable": 131449}
-{"epoch": 1.125, "grad_norm": 0.0004093375173397362, "learning_rate": 4.964235714846775e-05, "loss": 7.070232186379144e-06, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00001, "step": 288, "tokens/total": 8709504, "tokens/train_per_sec_per_gpu": 35.94, "tokens/trainable": 131923}
-{"epoch": 1.12890625, "grad_norm": 0.12260667979717255, "learning_rate": 4.9354244499126866e-05, "loss": 0.0006339149549603462, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00063, "step": 289, "tokens/total": 8739680, "tokens/train_per_sec_per_gpu": 33.11, "tokens/trainable": 132397}
-{"epoch": 1.1328125, "grad_norm": 0.00021185794321354479, "learning_rate": 4.90663667927174e-05, "loss": 5.396935193857644e-06, "memory/device_reserved (GiB)": 35.57, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 290, "tokens/total": 8770112, "tokens/train_per_sec_per_gpu": 36.32, "tokens/trainable": 132846}
-{"epoch": 1.13671875, "grad_norm": 0.2777019739151001, "learning_rate": 4.877873600900581e-05, "loss": 0.0013501858338713646, "memory/device_reserved (GiB)": 35.57, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00135, "step": 291, "tokens/total": 8800368, "tokens/train_per_sec_per_gpu": 35.65, "tokens/trainable": 133347}
-{"epoch": 1.140625, "grad_norm": 0.0012349279131740332, "learning_rate": 4.849136411748306e-05, "loss": 8.635176527604926e-06, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00001, "step": 292, "tokens/total": 8830688, "tokens/train_per_sec_per_gpu": 34.18, "tokens/trainable": 133779}
-{"epoch": 1.14453125, "grad_norm": 0.00026181069551967084, "learning_rate": 4.8204263076866574e-05, "loss": 4.307490144128678e-06, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0, "step": 293, "tokens/total": 8861008, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 134251}
-{"epoch": 1.1484375, "grad_norm": 0.00046773714711889625, "learning_rate": 4.791744483460251e-05, "loss": 4.467175585887162e-06, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.0, "step": 294, "tokens/total": 8891760, "tokens/train_per_sec_per_gpu": 33.41, "tokens/trainable": 134683}
-{"epoch": 1.15234375, "grad_norm": 0.0002164130419259891, "learning_rate": 4.7630921326368736e-05, "loss": 4.914254532195628e-06, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0, "step": 295, "tokens/total": 8922192, "tokens/train_per_sec_per_gpu": 35.51, "tokens/trainable": 135165}
-{"epoch": 1.15625, "grad_norm": 0.0002447084116283804, "learning_rate": 4.7344704475577916e-05, "loss": 4.242253453412559e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.0, "step": 296, "tokens/total": 8952496, "tokens/train_per_sec_per_gpu": 33.18, "tokens/trainable": 135599}
-{"epoch": 1.16015625, "grad_norm": 0.00028775070677511394, "learning_rate": 4.705880619288153e-05, "loss": 4.674366209656e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0, "step": 297, "tokens/total": 8982768, "tokens/train_per_sec_per_gpu": 32.56, "tokens/trainable": 136038}
-{"epoch": 1.1640625, "grad_norm": 0.001555647817440331, "learning_rate": 4.677323837567412e-05, "loss": 8.829203579807654e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 298, "tokens/total": 9013248, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 136495}
-{"epoch": 1.16796875, "grad_norm": 0.0002536054525990039, "learning_rate": 4.6488012907598146e-05, "loss": 5.5865721151349135e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 299, "tokens/total": 9043632, "tokens/train_per_sec_per_gpu": 35.3, "tokens/trainable": 136953}
-{"epoch": 1.171875, "grad_norm": 0.00031900990870781243, "learning_rate": 4.620314165804964e-05, "loss": 5.917333055549534e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00001, "step": 300, "tokens/total": 9073936, "tokens/train_per_sec_per_gpu": 34.14, "tokens/trainable": 137412}
-{"epoch": 1.17578125, "grad_norm": 0.0005231397226452827, "learning_rate": 4.591863648168407e-05, "loss": 9.367842721985653e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 301, "tokens/total": 9104416, "tokens/train_per_sec_per_gpu": 35.62, "tokens/trainable": 137907}
-{"epoch": 1.1796875, "grad_norm": 0.0010045799426734447, "learning_rate": 4.5634509217923135e-05, "loss": 9.613295333110727e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 302, "tokens/total": 9134688, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 138383}
-{"epoch": 1.18359375, "grad_norm": 0.0054819462820887566, "learning_rate": 4.535077169046201e-05, "loss": 2.4776203645160422e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 303, "tokens/total": 9165088, "tokens/train_per_sec_per_gpu": 36.35, "tokens/trainable": 138842}
-{"epoch": 1.1875, "grad_norm": 0.0005682530463673174, "learning_rate": 4.506743570677743e-05, "loss": 5.2244549806346186e-06, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 304, "tokens/total": 9195584, "tokens/train_per_sec_per_gpu": 30.61, "tokens/trainable": 139269}
-{"epoch": 1.19140625, "grad_norm": 0.003496676916256547, "learning_rate": 4.478451305763618e-05, "loss": 2.194346598116681e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 305, "tokens/total": 9225760, "tokens/train_per_sec_per_gpu": 34.36, "tokens/trainable": 139736}
-{"epoch": 1.1953125, "grad_norm": 0.000201961855054833, "learning_rate": 4.450201551660454e-05, "loss": 4.178235940344166e-06, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0, "step": 306, "tokens/total": 9256352, "tokens/train_per_sec_per_gpu": 36.91, "tokens/trainable": 140226}
-{"epoch": 1.19921875, "grad_norm": 0.0004581330285873264, "learning_rate": 4.4219954839558276e-05, "loss": 5.6062099247355945e-06, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.19, "memory/max_allocated (GiB)": 33.19, "ppl": 1.00001, "step": 307, "tokens/total": 9284256, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 140664}
-{"epoch": 1.203125, "grad_norm": 0.0016614671330899, "learning_rate": 4.393834276419352e-05, "loss": 7.163904228946194e-06, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00001, "step": 308, "tokens/total": 9314768, "tokens/train_per_sec_per_gpu": 29.85, "tokens/trainable": 141115}
-{"epoch": 1.20703125, "grad_norm": 0.00019009379320777953, "learning_rate": 4.36571910095382e-05, "loss": 3.8271318771876395e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0, "step": 309, "tokens/total": 9344976, "tokens/train_per_sec_per_gpu": 28.88, "tokens/trainable": 141530}
-{"epoch": 1.2109375, "grad_norm": 0.0038869199343025684, "learning_rate": 4.337651127546448e-05, "loss": 2.814065919665154e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00003, "step": 310, "tokens/total": 9375280, "tokens/train_per_sec_per_gpu": 34.15, "tokens/trainable": 141964}
-{"epoch": 1.21484375, "grad_norm": 0.024828940629959106, "learning_rate": 4.3096315242201736e-05, "loss": 0.00011992595682386309, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00012, "step": 311, "tokens/total": 9405536, "tokens/train_per_sec_per_gpu": 33.95, "tokens/trainable": 142428}
-{"epoch": 1.21875, "grad_norm": 0.0017960596596822143, "learning_rate": 4.2816614569850635e-05, "loss": 9.508907169220038e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00001, "step": 312, "tokens/total": 9435696, "tokens/train_per_sec_per_gpu": 38.07, "tokens/trainable": 142894}
-{"epoch": 1.22265625, "grad_norm": 0.0007181956898421049, "learning_rate": 4.2537420897897864e-05, "loss": 8.858227374730632e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 313, "tokens/total": 9466080, "tokens/train_per_sec_per_gpu": 36.3, "tokens/trainable": 143368}
-{"epoch": 1.2265625, "grad_norm": 0.7545468211174011, "learning_rate": 4.225874584473174e-05, "loss": 0.015099202282726765, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.01521, "step": 314, "tokens/total": 9496400, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 143824}
-{"epoch": 1.23046875, "grad_norm": 0.2810213565826416, "learning_rate": 4.19806010071587e-05, "loss": 0.0007513084565289319, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00075, "step": 315, "tokens/total": 9526880, "tokens/train_per_sec_per_gpu": 34.45, "tokens/trainable": 144306}
-{"epoch": 1.234375, "grad_norm": 0.7471011281013489, "learning_rate": 4.170299795992081e-05, "loss": 0.0005616231937892735, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00056, "step": 316, "tokens/total": 9557280, "tokens/train_per_sec_per_gpu": 35.3, "tokens/trainable": 144763}
-{"epoch": 1.23828125, "grad_norm": 0.0004946920089423656, "learning_rate": 4.142594825521398e-05, "loss": 5.734345904784277e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 317, "tokens/total": 9587712, "tokens/train_per_sec_per_gpu": 37.4, "tokens/trainable": 145255}
-{"epoch": 1.2421875, "grad_norm": 0.0032017722260206938, "learning_rate": 4.114946342220728e-05, "loss": 2.6251871531712823e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00003, "step": 318, "tokens/total": 9618064, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 145714}
-{"epoch": 1.24609375, "grad_norm": 1.1724350452423096, "learning_rate": 4.087355496656321e-05, "loss": 0.026299288496375084, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.02665, "step": 319, "tokens/total": 9648336, "tokens/train_per_sec_per_gpu": 35.96, "tokens/trainable": 146200}
-{"epoch": 1.25, "grad_norm": 0.00045995338587090373, "learning_rate": 4.05982343699588e-05, "loss": 8.491813787259161e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00001, "step": 320, "tokens/total": 9678688, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 146668}
-{"epoch": 1.25390625, "grad_norm": 0.0067353262566030025, "learning_rate": 4.0323513089607876e-05, "loss": 6.0428461438277736e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00006, "step": 321, "tokens/total": 9708848, "tokens/train_per_sec_per_gpu": 35.24, "tokens/trainable": 147114}
-{"epoch": 1.2578125, "grad_norm": 0.05487794429063797, "learning_rate": 4.004940255778431e-05, "loss": 0.0002498509711585939, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00025, "step": 322, "tokens/total": 9739360, "tokens/train_per_sec_per_gpu": 35.08, "tokens/trainable": 147564}
-{"epoch": 1.26171875, "grad_norm": 0.0011818762868642807, "learning_rate": 3.977591418134619e-05, "loss": 1.1004886800947133e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 323, "tokens/total": 9769776, "tokens/train_per_sec_per_gpu": 34.14, "tokens/trainable": 148013}
-{"epoch": 1.265625, "grad_norm": 0.018019674345850945, "learning_rate": 3.95030593412612e-05, "loss": 0.00021162032498978078, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.00021, "step": 324, "tokens/total": 9800096, "tokens/train_per_sec_per_gpu": 37.99, "tokens/trainable": 148470}
-{"epoch": 1.26953125, "grad_norm": 1.5907806158065796, "learning_rate": 3.923084939213296e-05, "loss": 0.016217660158872604, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01635, "step": 325, "tokens/total": 9830304, "tokens/train_per_sec_per_gpu": 36.44, "tokens/trainable": 148926}
-{"epoch": 1.2734375, "grad_norm": 0.004153774119913578, "learning_rate": 3.895929566172861e-05, "loss": 3.801286584348418e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 326, "tokens/total": 9860608, "tokens/train_per_sec_per_gpu": 37.72, "tokens/trainable": 149420}
-{"epoch": 1.27734375, "grad_norm": 0.0029778245370835066, "learning_rate": 3.868840945050728e-05, "loss": 5.526735549210571e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00006, "step": 327, "tokens/total": 9890560, "tokens/train_per_sec_per_gpu": 39.32, "tokens/trainable": 149873}
-{"epoch": 1.28125, "grad_norm": 0.028163742274045944, "learning_rate": 3.841820203114995e-05, "loss": 0.00039056455716490746, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00039, "step": 328, "tokens/total": 9920880, "tokens/train_per_sec_per_gpu": 33.78, "tokens/trainable": 150347}
-{"epoch": 1.28515625, "grad_norm": 0.04141407459974289, "learning_rate": 3.814868464809027e-05, "loss": 0.0005860928213223815, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00059, "step": 329, "tokens/total": 9951280, "tokens/train_per_sec_per_gpu": 31.72, "tokens/trainable": 150822}
-{"epoch": 1.2890625, "grad_norm": 0.004758972208946943, "learning_rate": 3.787986851704667e-05, "loss": 9.08115180209279e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00009, "step": 330, "tokens/total": 9981600, "tokens/train_per_sec_per_gpu": 35.03, "tokens/trainable": 151266}
-{"epoch": 1.29296875, "grad_norm": 0.04043704643845558, "learning_rate": 3.7611764824555654e-05, "loss": 0.0005757657927460968, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00058, "step": 331, "tokens/total": 10012016, "tokens/train_per_sec_per_gpu": 37.11, "tokens/trainable": 151734}
-{"epoch": 1.296875, "grad_norm": 0.064565509557724, "learning_rate": 3.734438472750619e-05, "loss": 0.00043528492096811533, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00044, "step": 332, "tokens/total": 10042448, "tokens/train_per_sec_per_gpu": 33.99, "tokens/trainable": 152184}
-{"epoch": 1.30078125, "grad_norm": 0.051213983446359634, "learning_rate": 3.707773935267552e-05, "loss": 0.000528274686075747, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00053, "step": 333, "tokens/total": 10073024, "tokens/train_per_sec_per_gpu": 31.79, "tokens/trainable": 152623}
-{"epoch": 1.3046875, "grad_norm": 0.0067980666644871235, "learning_rate": 3.68118397962661e-05, "loss": 0.00013989521539770067, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00014, "step": 334, "tokens/total": 10103504, "tokens/train_per_sec_per_gpu": 33.34, "tokens/trainable": 153040}
-{"epoch": 1.30859375, "grad_norm": 0.28768453001976013, "learning_rate": 3.654669712344384e-05, "loss": 0.0035491236485540867, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.31, "memory/max_allocated (GiB)": 33.31, "ppl": 1.00356, "step": 335, "tokens/total": 10131632, "tokens/train_per_sec_per_gpu": 31.7, "tokens/trainable": 153463}
-{"epoch": 1.3125, "grad_norm": 0.04292941838502884, "learning_rate": 3.628232236787763e-05, "loss": 0.0007545957923866808, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00075, "step": 336, "tokens/total": 10161824, "tokens/train_per_sec_per_gpu": 33.22, "tokens/trainable": 153908}
-{"epoch": 1.31640625, "grad_norm": 0.13187745213508606, "learning_rate": 3.6018726531280144e-05, "loss": 0.0011231126263737679, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00112, "step": 337, "tokens/total": 10192448, "tokens/train_per_sec_per_gpu": 33.63, "tokens/trainable": 154387}
-{"epoch": 1.3203125, "grad_norm": 0.06429488956928253, "learning_rate": 3.575592058295017e-05, "loss": 0.0004924655659124255, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00049, "step": 338, "tokens/total": 10222704, "tokens/train_per_sec_per_gpu": 33.82, "tokens/trainable": 154836}
-{"epoch": 1.32421875, "grad_norm": 0.006816778797656298, "learning_rate": 3.549391545931585e-05, "loss": 8.63251625560224e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00009, "step": 339, "tokens/total": 10253168, "tokens/train_per_sec_per_gpu": 29.73, "tokens/trainable": 155261}
-{"epoch": 1.328125, "grad_norm": 0.04606792330741882, "learning_rate": 3.5232722063479914e-05, "loss": 0.0003515402786433697, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00035, "step": 340, "tokens/total": 10283552, "tokens/train_per_sec_per_gpu": 31.14, "tokens/trainable": 155674}
-{"epoch": 1.33203125, "grad_norm": 0.24907881021499634, "learning_rate": 3.49723512647657e-05, "loss": 0.000742567703127861, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00074, "step": 341, "tokens/total": 10313872, "tokens/train_per_sec_per_gpu": 35.67, "tokens/trainable": 156164}
-{"epoch": 1.3359375, "grad_norm": 0.011238012462854385, "learning_rate": 3.471281389826491e-05, "loss": 0.0001773187832441181, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00018, "step": 342, "tokens/total": 10344256, "tokens/train_per_sec_per_gpu": 33.47, "tokens/trainable": 156611}
-{"epoch": 1.33984375, "grad_norm": 0.00949561782181263, "learning_rate": 3.4454120764386764e-05, "loss": 0.00018969883967656642, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00019, "step": 343, "tokens/total": 10374544, "tokens/train_per_sec_per_gpu": 33.58, "tokens/trainable": 157073}
-{"epoch": 1.34375, "grad_norm": 0.10763411223888397, "learning_rate": 3.4196282628408526e-05, "loss": 0.0007583287660963833, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00076, "step": 344, "tokens/total": 10405040, "tokens/train_per_sec_per_gpu": 32.94, "tokens/trainable": 157523}
-{"epoch": 1.34765625, "grad_norm": 0.15938052535057068, "learning_rate": 3.3939310220027456e-05, "loss": 0.0014999746344983578, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0015, "step": 345, "tokens/total": 10435424, "tokens/train_per_sec_per_gpu": 35.07, "tokens/trainable": 157964}
-{"epoch": 1.3515625, "grad_norm": 0.00165572261903435, "learning_rate": 3.3683214232914404e-05, "loss": 3.833783557638526e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00004, "step": 346, "tokens/total": 10465760, "tokens/train_per_sec_per_gpu": 30.35, "tokens/trainable": 158390}
-{"epoch": 1.35546875, "grad_norm": 0.02406073547899723, "learning_rate": 3.342800532426873e-05, "loss": 0.00039186165668070316, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00039, "step": 347, "tokens/total": 10495904, "tokens/train_per_sec_per_gpu": 35.47, "tokens/trainable": 158848}
-{"epoch": 1.359375, "grad_norm": 0.045792482793331146, "learning_rate": 3.317369411437484e-05, "loss": 0.0006161610363051295, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 34.0, "memory/max_allocated (GiB)": 34.0, "ppl": 1.00062, "step": 348, "tokens/total": 10526624, "tokens/train_per_sec_per_gpu": 28.63, "tokens/trainable": 159270}
-{"epoch": 1.36328125, "grad_norm": 0.18648307025432587, "learning_rate": 3.292029118616024e-05, "loss": 0.0012868957128375769, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00129, "step": 349, "tokens/total": 10556752, "tokens/train_per_sec_per_gpu": 32.46, "tokens/trainable": 159720}
-{"epoch": 1.3671875, "grad_norm": 0.015933886170387268, "learning_rate": 3.266780708475511e-05, "loss": 0.0002091687492793426, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00021, "step": 350, "tokens/total": 10587264, "tokens/train_per_sec_per_gpu": 35.09, "tokens/trainable": 160187}
-{"epoch": 1.37109375, "grad_norm": 0.008724886924028397, "learning_rate": 3.241625231705354e-05, "loss": 0.00013179841334931552, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00013, "step": 351, "tokens/total": 10617504, "tokens/train_per_sec_per_gpu": 32.55, "tokens/trainable": 160645}
-{"epoch": 1.375, "grad_norm": 0.06377559900283813, "learning_rate": 3.216563735127618e-05, "loss": 0.0008062056731432676, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00081, "step": 352, "tokens/total": 10647760, "tokens/train_per_sec_per_gpu": 34.42, "tokens/trainable": 161102}
-{"epoch": 1.37890625, "grad_norm": 0.0037202269304543734, "learning_rate": 3.191597261653475e-05, "loss": 7.238816760946065e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00007, "step": 353, "tokens/total": 10678080, "tokens/train_per_sec_per_gpu": 28.86, "tokens/trainable": 161555}
-{"epoch": 1.3828125, "grad_norm": 0.0021510994993150234, "learning_rate": 3.166726850239794e-05, "loss": 3.6805788113269955e-05, "memory/device_reserved (GiB)": 35.41, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00004, "step": 354, "tokens/total": 10708544, "tokens/train_per_sec_per_gpu": 36.65, "tokens/trainable": 162020}
-{"epoch": 1.38671875, "grad_norm": 0.002642499515786767, "learning_rate": 3.141953535845912e-05, "loss": 4.371708200778812e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00004, "step": 355, "tokens/total": 10739024, "tokens/train_per_sec_per_gpu": 32.05, "tokens/trainable": 162448}
-{"epoch": 1.390625, "grad_norm": 0.0038646068423986435, "learning_rate": 3.11727834939056e-05, "loss": 7.11614266037941e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.61, "memory/max_allocated (GiB)": 33.61, "ppl": 1.00007, "step": 356, "tokens/total": 10769024, "tokens/train_per_sec_per_gpu": 28.2, "tokens/trainable": 162863}
-{"epoch": 1.39453125, "grad_norm": 0.017895536497235298, "learning_rate": 3.092702317708967e-05, "loss": 5.4065520089352503e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00005, "step": 357, "tokens/total": 10799328, "tokens/train_per_sec_per_gpu": 34.94, "tokens/trainable": 163318}
-{"epoch": 1.3984375, "grad_norm": 0.8479411602020264, "learning_rate": 3.0682264635101276e-05, "loss": 0.01497839204967022, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01509, "step": 358, "tokens/total": 10829488, "tokens/train_per_sec_per_gpu": 32.89, "tokens/trainable": 163767}
-{"epoch": 1.40234375, "grad_norm": 0.0015704578254371881, "learning_rate": 3.0438518053342407e-05, "loss": 3.0390210667974316e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 359, "tokens/total": 10859936, "tokens/train_per_sec_per_gpu": 31.78, "tokens/trainable": 164208}
-{"epoch": 1.40625, "grad_norm": 0.014669318683445454, "learning_rate": 3.0195793575103266e-05, "loss": 0.00010030368866864592, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0001, "step": 360, "tokens/total": 10890400, "tokens/train_per_sec_per_gpu": 34.08, "tokens/trainable": 164655}
-{"epoch": 1.41015625, "grad_norm": 0.3479421138763428, "learning_rate": 2.9954101301140146e-05, "loss": 0.0026096655055880547, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00261, "step": 361, "tokens/total": 10920624, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 165100}
-{"epoch": 1.4140625, "grad_norm": 0.004394443240016699, "learning_rate": 2.9713451289255123e-05, "loss": 6.116658914834261e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00006, "step": 362, "tokens/total": 10951136, "tokens/train_per_sec_per_gpu": 34.84, "tokens/trainable": 165587}
-{"epoch": 1.41796875, "grad_norm": 0.016197621822357178, "learning_rate": 2.9473853553877484e-05, "loss": 0.0001775357231963426, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00018, "step": 363, "tokens/total": 10981344, "tokens/train_per_sec_per_gpu": 34.09, "tokens/trainable": 166034}
-{"epoch": 1.421875, "grad_norm": 0.09478334337472916, "learning_rate": 2.9235318065647e-05, "loss": 0.0005033796769566834, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.0005, "step": 364, "tokens/total": 11011552, "tokens/train_per_sec_per_gpu": 34.04, "tokens/trainable": 166485}
-{"epoch": 1.42578125, "grad_norm": 0.014899010770022869, "learning_rate": 2.8997854750998964e-05, "loss": 0.0001904651871882379, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00019, "step": 365, "tokens/total": 11041872, "tokens/train_per_sec_per_gpu": 36.32, "tokens/trainable": 166973}
-{"epoch": 1.4296875, "grad_norm": 0.005555527750402689, "learning_rate": 2.8761473491751258e-05, "loss": 0.00010058133921120316, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.0001, "step": 366, "tokens/total": 11072192, "tokens/train_per_sec_per_gpu": 29.1, "tokens/trainable": 167414}
-{"epoch": 1.43359375, "grad_norm": 0.20706292986869812, "learning_rate": 2.8526184124692883e-05, "loss": 0.0005888720043003559, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00059, "step": 367, "tokens/total": 11102736, "tokens/train_per_sec_per_gpu": 30.11, "tokens/trainable": 167851}
-{"epoch": 1.4375, "grad_norm": 0.0007191727054305375, "learning_rate": 2.829199644117484e-05, "loss": 1.2304372830840293e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 368, "tokens/total": 11133056, "tokens/train_per_sec_per_gpu": 35.48, "tokens/trainable": 168294}
-{"epoch": 1.44140625, "grad_norm": 0.030878128483891487, "learning_rate": 2.8058920186702553e-05, "loss": 0.0002043281274382025, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0002, "step": 369, "tokens/total": 11163568, "tokens/train_per_sec_per_gpu": 32.08, "tokens/trainable": 168769}
-{"epoch": 1.4453125, "grad_norm": 0.008817057125270367, "learning_rate": 2.782696506053033e-05, "loss": 0.00011464582348708063, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00011, "step": 370, "tokens/total": 11193936, "tokens/train_per_sec_per_gpu": 36.59, "tokens/trainable": 169270}
-{"epoch": 1.44921875, "grad_norm": 0.023856336250901222, "learning_rate": 2.7596140715257824e-05, "loss": 0.00013431125262286514, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00013, "step": 371, "tokens/total": 11224480, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 169711}
-{"epoch": 1.453125, "grad_norm": 0.010502993129193783, "learning_rate": 2.7366456756428184e-05, "loss": 0.00015104333579074591, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00015, "step": 372, "tokens/total": 11254912, "tokens/train_per_sec_per_gpu": 32.81, "tokens/trainable": 170185}
-{"epoch": 1.45703125, "grad_norm": 0.026970423758029938, "learning_rate": 2.7137922742128486e-05, "loss": 0.00026081278338097036, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00026, "step": 373, "tokens/total": 11285104, "tokens/train_per_sec_per_gpu": 33.08, "tokens/trainable": 170650}
-{"epoch": 1.4609375, "grad_norm": 0.020325109362602234, "learning_rate": 2.691054818259188e-05, "loss": 0.00014532633940689266, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00015, "step": 374, "tokens/total": 11315600, "tokens/train_per_sec_per_gpu": 31.98, "tokens/trainable": 171126}
-{"epoch": 1.46484375, "grad_norm": 0.033260464668273926, "learning_rate": 2.6684342539801933e-05, "loss": 0.00045571548980660737, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00046, "step": 375, "tokens/total": 11345776, "tokens/train_per_sec_per_gpu": 35.6, "tokens/trainable": 171600}
-{"epoch": 1.46875, "grad_norm": 0.017290709540247917, "learning_rate": 2.645931522709877e-05, "loss": 0.00014910403115209192, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00015, "step": 376, "tokens/total": 11376208, "tokens/train_per_sec_per_gpu": 28.09, "tokens/trainable": 172017}
-{"epoch": 1.47265625, "grad_norm": 0.04457363486289978, "learning_rate": 2.6235475608787365e-05, "loss": 6.834026135038584e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00007, "step": 377, "tokens/total": 11406512, "tokens/train_per_sec_per_gpu": 35.99, "tokens/trainable": 172493}
-{"epoch": 1.4765625, "grad_norm": 0.002268735785037279, "learning_rate": 2.6012832999747916e-05, "loss": 3.245133120799437e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00003, "step": 378, "tokens/total": 11436544, "tokens/train_per_sec_per_gpu": 34.27, "tokens/trainable": 172951}
-{"epoch": 1.48046875, "grad_norm": 0.0038171466439962387, "learning_rate": 2.579139666504821e-05, "loss": 5.4866883147042245e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00005, "step": 379, "tokens/total": 11467008, "tokens/train_per_sec_per_gpu": 35.63, "tokens/trainable": 173435}
-{"epoch": 1.484375, "grad_norm": 0.005268535576760769, "learning_rate": 2.557117581955798e-05, "loss": 8.241234172601253e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00008, "step": 380, "tokens/total": 11497184, "tokens/train_per_sec_per_gpu": 30.4, "tokens/trainable": 173900}
-{"epoch": 1.48828125, "grad_norm": 0.016057243570685387, "learning_rate": 2.5352179627565532e-05, "loss": 0.00014235377602744848, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00014, "step": 381, "tokens/total": 11527600, "tokens/train_per_sec_per_gpu": 34.12, "tokens/trainable": 174371}
-{"epoch": 1.4921875, "grad_norm": 0.005853456910699606, "learning_rate": 2.5134417202396277e-05, "loss": 5.001476893085055e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00005, "step": 382, "tokens/total": 11557808, "tokens/train_per_sec_per_gpu": 32.76, "tokens/trainable": 174826}
-{"epoch": 1.49609375, "grad_norm": 0.0031664848793298006, "learning_rate": 2.491789760603361e-05, "loss": 4.210277256788686e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00004, "step": 383, "tokens/total": 11588352, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 175276}
-{"epoch": 1.5, "grad_norm": 0.08838633447885513, "learning_rate": 2.4702629848741764e-05, "loss": 0.0010394920827820897, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00104, "step": 384, "tokens/total": 11618640, "tokens/train_per_sec_per_gpu": 33.69, "tokens/trainable": 175720}
-{"epoch": 1.50390625, "grad_norm": 0.09446703642606735, "learning_rate": 2.4488622888690785e-05, "loss": 0.0005370234721340239, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00054, "step": 385, "tokens/total": 11649072, "tokens/train_per_sec_per_gpu": 35.23, "tokens/trainable": 176203}
-{"epoch": 1.5078125, "grad_norm": 0.0013355027185752988, "learning_rate": 2.427588563158384e-05, "loss": 1.903088195831515e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 386, "tokens/total": 11679552, "tokens/train_per_sec_per_gpu": 34.29, "tokens/trainable": 176651}
-{"epoch": 1.51171875, "grad_norm": 0.011693151667714119, "learning_rate": 2.406442693028651e-05, "loss": 7.680666749365628e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00008, "step": 387, "tokens/total": 11709760, "tokens/train_per_sec_per_gpu": 32.7, "tokens/trainable": 177115}
-{"epoch": 1.515625, "grad_norm": 0.09523260593414307, "learning_rate": 2.3854255584458547e-05, "loss": 0.00073521543527022, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00074, "step": 388, "tokens/total": 11740288, "tokens/train_per_sec_per_gpu": 36.09, "tokens/trainable": 177601}
-{"epoch": 1.51953125, "grad_norm": 0.005971741396933794, "learning_rate": 2.3645380340187508e-05, "loss": 6.468063656939194e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00006, "step": 389, "tokens/total": 11770592, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 178050}
-{"epoch": 1.5234375, "grad_norm": 0.0010247502941638231, "learning_rate": 2.3437809889624914e-05, "loss": 1.5989648090908304e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 390, "tokens/total": 11800752, "tokens/train_per_sec_per_gpu": 36.38, "tokens/trainable": 178531}
-{"epoch": 1.52734375, "grad_norm": 0.054420940577983856, "learning_rate": 2.3231552870624487e-05, "loss": 0.0004186803416814655, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00042, "step": 391, "tokens/total": 11831360, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 179000}
-{"epoch": 1.53125, "grad_norm": 0.002925195964053273, "learning_rate": 2.3026617866382657e-05, "loss": 1.531536145193968e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 392, "tokens/total": 11861552, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 179457}
-{"epoch": 1.53515625, "grad_norm": 1.6402941942214966, "learning_rate": 2.2823013405081507e-05, "loss": 0.0027035027742385864, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00271, "step": 393, "tokens/total": 11891904, "tokens/train_per_sec_per_gpu": 32.73, "tokens/trainable": 179890}
-{"epoch": 1.5390625, "grad_norm": 0.0032207504846155643, "learning_rate": 2.2620747959533722e-05, "loss": 3.968120290664956e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00004, "step": 394, "tokens/total": 11922208, "tokens/train_per_sec_per_gpu": 33.89, "tokens/trainable": 180341}
-{"epoch": 1.54296875, "grad_norm": 0.013098486699163914, "learning_rate": 2.2419829946830123e-05, "loss": 0.00011063168494729325, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00011, "step": 395, "tokens/total": 11952672, "tokens/train_per_sec_per_gpu": 32.95, "tokens/trainable": 180762}
-{"epoch": 1.546875, "grad_norm": 0.01553372759371996, "learning_rate": 2.2220267727989325e-05, "loss": 8.802580123301595e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00009, "step": 396, "tokens/total": 11983088, "tokens/train_per_sec_per_gpu": 35.04, "tokens/trainable": 181226}
-{"epoch": 1.55078125, "grad_norm": 0.016083814203739166, "learning_rate": 2.202206960760984e-05, "loss": 8.931377669796348e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00009, "step": 397, "tokens/total": 12013488, "tokens/train_per_sec_per_gpu": 33.15, "tokens/trainable": 181714}
-{"epoch": 1.5546875, "grad_norm": 0.05752059072256088, "learning_rate": 2.182524383352446e-05, "loss": 0.00015369296306744218, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00015, "step": 398, "tokens/total": 12043968, "tokens/train_per_sec_per_gpu": 35.52, "tokens/trainable": 182142}
-{"epoch": 1.55859375, "grad_norm": 0.0011741623748093843, "learning_rate": 2.1629798596457056e-05, "loss": 1.8113165424438193e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00002, "step": 399, "tokens/total": 12074160, "tokens/train_per_sec_per_gpu": 34.82, "tokens/trainable": 182604}
-{"epoch": 1.5625, "grad_norm": 0.0048082731664180756, "learning_rate": 2.1435742029681725e-05, "loss": 5.3348740038927644e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.67, "memory/max_allocated (GiB)": 33.67, "ppl": 1.00005, "step": 400, "tokens/total": 12104352, "tokens/train_per_sec_per_gpu": 25.28, "tokens/trainable": 183011}
-{"epoch": 1.56640625, "grad_norm": 0.006520449183881283, "learning_rate": 2.124308220868431e-05, "loss": 6.45049221930094e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00006, "step": 401, "tokens/total": 12134240, "tokens/train_per_sec_per_gpu": 31.69, "tokens/trainable": 183459}
-{"epoch": 1.5703125, "grad_norm": 0.006227654870599508, "learning_rate": 2.105182715082638e-05, "loss": 6.928759830771014e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00007, "step": 402, "tokens/total": 12164480, "tokens/train_per_sec_per_gpu": 38.61, "tokens/trainable": 183959}
-{"epoch": 1.57421875, "grad_norm": 0.32605040073394775, "learning_rate": 2.0861984815011552e-05, "loss": 0.002115404698997736, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00212, "step": 403, "tokens/total": 12194640, "tokens/train_per_sec_per_gpu": 32.46, "tokens/trainable": 184431}
-{"epoch": 1.578125, "grad_norm": 0.007742208894342184, "learning_rate": 2.0673563101354323e-05, "loss": 4.600908869178966e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00005, "step": 404, "tokens/total": 12224960, "tokens/train_per_sec_per_gpu": 31.34, "tokens/trainable": 184867}
-{"epoch": 1.58203125, "grad_norm": 0.0049946485087275505, "learning_rate": 2.0486569850851317e-05, "loss": 5.003535625291988e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00005, "step": 405, "tokens/total": 12255008, "tokens/train_per_sec_per_gpu": 29.25, "tokens/trainable": 185300}
-{"epoch": 1.5859375, "grad_norm": 0.004598768427968025, "learning_rate": 2.0301012845054956e-05, "loss": 3.4423381293891e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00003, "step": 406, "tokens/total": 12285248, "tokens/train_per_sec_per_gpu": 31.45, "tokens/trainable": 185754}
-{"epoch": 1.58984375, "grad_norm": 0.0011464629787951708, "learning_rate": 2.011689980574966e-05, "loss": 1.9196500943508e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00002, "step": 407, "tokens/total": 12315552, "tokens/train_per_sec_per_gpu": 36.8, "tokens/trainable": 186213}
-{"epoch": 1.59375, "grad_norm": 0.004688040353357792, "learning_rate": 1.993423839463052e-05, "loss": 5.2492636314127594e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00005, "step": 408, "tokens/total": 12345904, "tokens/train_per_sec_per_gpu": 32.72, "tokens/trainable": 186659}
-{"epoch": 1.59765625, "grad_norm": 0.002637861529365182, "learning_rate": 1.975303621298445e-05, "loss": 3.172009019181132e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 409, "tokens/total": 12376336, "tokens/train_per_sec_per_gpu": 32.92, "tokens/trainable": 187083}
-{"epoch": 1.6015625, "grad_norm": 0.0072015393525362015, "learning_rate": 1.957330080137385e-05, "loss": 5.452537880046293e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00005, "step": 410, "tokens/total": 12406880, "tokens/train_per_sec_per_gpu": 34.91, "tokens/trainable": 187530}
-{"epoch": 1.60546875, "grad_norm": 0.007169651333242655, "learning_rate": 1.9395039639322864e-05, "loss": 3.530656249495223e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 411, "tokens/total": 12437088, "tokens/train_per_sec_per_gpu": 34.28, "tokens/trainable": 187972}
-{"epoch": 1.609375, "grad_norm": 0.003889538114890456, "learning_rate": 1.9218260145006073e-05, "loss": 3.5087461583316326e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00004, "step": 412, "tokens/total": 12465440, "tokens/train_per_sec_per_gpu": 39.89, "tokens/trainable": 188417}
-{"epoch": 1.61328125, "grad_norm": 0.5135430693626404, "learning_rate": 1.904296967493982e-05, "loss": 0.005697700660675764, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00571, "step": 413, "tokens/total": 12495968, "tokens/train_per_sec_per_gpu": 31.48, "tokens/trainable": 188868}
-{"epoch": 1.6171875, "grad_norm": 0.0010096468031406403, "learning_rate": 1.8869175523676064e-05, "loss": 1.1318426913931035e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00001, "step": 414, "tokens/total": 12526128, "tokens/train_per_sec_per_gpu": 36.85, "tokens/trainable": 189337}
-{"epoch": 1.62109375, "grad_norm": 0.0004970752634108067, "learning_rate": 1.869688492349885e-05, "loss": 7.355167326750234e-06, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00001, "step": 415, "tokens/total": 12556256, "tokens/train_per_sec_per_gpu": 31.82, "tokens/trainable": 189750}
-{"epoch": 1.625, "grad_norm": 0.0008849130244925618, "learning_rate": 1.85261050441233e-05, "loss": 1.4194254617905244e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 416, "tokens/total": 12586736, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 190216}
-{"epoch": 1.62890625, "grad_norm": 0.0022039280738681555, "learning_rate": 1.8356842992397304e-05, "loss": 2.4539594960515387e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 417, "tokens/total": 12617168, "tokens/train_per_sec_per_gpu": 34.95, "tokens/trainable": 190708}
-{"epoch": 1.6328125, "grad_norm": 0.027006179094314575, "learning_rate": 1.8189105812005714e-05, "loss": 0.00020442574168555439, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.0002, "step": 418, "tokens/total": 12647680, "tokens/train_per_sec_per_gpu": 30.85, "tokens/trainable": 191126}
-{"epoch": 1.63671875, "grad_norm": 0.15684254467487335, "learning_rate": 1.802290048317732e-05, "loss": 0.0007251293282024562, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00073, "step": 419, "tokens/total": 12677952, "tokens/train_per_sec_per_gpu": 30.3, "tokens/trainable": 191540}
-{"epoch": 1.640625, "grad_norm": 0.0033104538451880217, "learning_rate": 1.785823392239424e-05, "loss": 3.131272751488723e-05, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00003, "step": 420, "tokens/total": 12708416, "tokens/train_per_sec_per_gpu": 36.65, "tokens/trainable": 192014}
-{"epoch": 1.64453125, "grad_norm": 0.001255685230717063, "learning_rate": 1.7695112982104225e-05, "loss": 1.9309976778458804e-05, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 421, "tokens/total": 12738640, "tokens/train_per_sec_per_gpu": 31.28, "tokens/trainable": 192463}
-{"epoch": 1.6484375, "grad_norm": 0.01599641516804695, "learning_rate": 1.7533544450435433e-05, "loss": 9.235648030880839e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00009, "step": 422, "tokens/total": 12769232, "tokens/train_per_sec_per_gpu": 32.71, "tokens/trainable": 192918}
-{"epoch": 1.65234375, "grad_norm": 0.048507146537303925, "learning_rate": 1.7373535050913946e-05, "loss": 0.0002708940301090479, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00027, "step": 423, "tokens/total": 12799648, "tokens/train_per_sec_per_gpu": 37.35, "tokens/trainable": 193425}
-{"epoch": 1.65625, "grad_norm": 0.010107563808560371, "learning_rate": 1.721509144218405e-05, "loss": 6.99054216966033e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00007, "step": 424, "tokens/total": 12829920, "tokens/train_per_sec_per_gpu": 35.23, "tokens/trainable": 193898}
-{"epoch": 1.66015625, "grad_norm": 0.005467226263135672, "learning_rate": 1.705822021773101e-05, "loss": 5.417566717369482e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00005, "step": 425, "tokens/total": 12860112, "tokens/train_per_sec_per_gpu": 29.24, "tokens/trainable": 194315}
-{"epoch": 1.6640625, "grad_norm": 0.0015799012035131454, "learning_rate": 1.69029279056068e-05, "loss": 2.4984849005704746e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00002, "step": 426, "tokens/total": 12890320, "tokens/train_per_sec_per_gpu": 34.92, "tokens/trainable": 194796}
-{"epoch": 1.66796875, "grad_norm": 0.024157166481018066, "learning_rate": 1.6749220968158415e-05, "loss": 0.00010696032404666767, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00011, "step": 427, "tokens/total": 12920656, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 195262}
-{"epoch": 1.671875, "grad_norm": 0.00028090659179724753, "learning_rate": 1.659710580175893e-05, "loss": 7.484430170734413e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 428, "tokens/total": 12951040, "tokens/train_per_sec_per_gpu": 31.83, "tokens/trainable": 195680}
-{"epoch": 1.67578125, "grad_norm": 0.0011094068177044392, "learning_rate": 1.644658873654133e-05, "loss": 1.9920153135899454e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00002, "step": 429, "tokens/total": 12981232, "tokens/train_per_sec_per_gpu": 32.83, "tokens/trainable": 196145}
-{"epoch": 1.6796875, "grad_norm": 0.020794112235307693, "learning_rate": 1.629767603613508e-05, "loss": 0.00010971157462336123, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00011, "step": 430, "tokens/total": 13011616, "tokens/train_per_sec_per_gpu": 32.39, "tokens/trainable": 196626}
-{"epoch": 1.68359375, "grad_norm": 0.44080695509910583, "learning_rate": 1.615037389740547e-05, "loss": 0.0026788683608174324, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.00268, "step": 431, "tokens/total": 13042208, "tokens/train_per_sec_per_gpu": 31.14, "tokens/trainable": 197067}
-{"epoch": 1.6875, "grad_norm": 0.0026339287869632244, "learning_rate": 1.600468845019576e-05, "loss": 2.082335777231492e-05, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00002, "step": 432, "tokens/total": 13072800, "tokens/train_per_sec_per_gpu": 38.61, "tokens/trainable": 197565}
-{"epoch": 1.69140625, "grad_norm": 0.004508621525019407, "learning_rate": 1.5860625757072092e-05, "loss": 1.439991501683835e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00001, "step": 433, "tokens/total": 13103328, "tokens/train_per_sec_per_gpu": 32.88, "tokens/trainable": 198015}
-{"epoch": 1.6953125, "grad_norm": 0.014811511151492596, "learning_rate": 1.571819181307116e-05, "loss": 0.00015316363715101033, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00015, "step": 434, "tokens/total": 13133472, "tokens/train_per_sec_per_gpu": 33.87, "tokens/trainable": 198442}
-{"epoch": 1.69921875, "grad_norm": 0.0037793368101119995, "learning_rate": 1.557739254545075e-05, "loss": 1.4444960470427759e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 435, "tokens/total": 13164064, "tokens/train_per_sec_per_gpu": 34.3, "tokens/trainable": 198914}
-{"epoch": 1.703125, "grad_norm": 0.24972176551818848, "learning_rate": 1.543823381344311e-05, "loss": 0.0017114672809839249, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00171, "step": 436, "tokens/total": 13194464, "tokens/train_per_sec_per_gpu": 33.35, "tokens/trainable": 199355}
-{"epoch": 1.70703125, "grad_norm": 0.002479988383129239, "learning_rate": 1.5300721408011114e-05, "loss": 2.9400333005469292e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00003, "step": 437, "tokens/total": 13224992, "tokens/train_per_sec_per_gpu": 32.91, "tokens/trainable": 199830}
-{"epoch": 1.7109375, "grad_norm": 0.0014780610799789429, "learning_rate": 1.5164861051607254e-05, "loss": 1.5593774151057005e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 438, "tokens/total": 13255296, "tokens/train_per_sec_per_gpu": 34.89, "tokens/trainable": 200338}
-{"epoch": 1.71484375, "grad_norm": 0.0010410621762275696, "learning_rate": 1.5030658397935521e-05, "loss": 1.4683226254419424e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00001, "step": 439, "tokens/total": 13285344, "tokens/train_per_sec_per_gpu": 32.79, "tokens/trainable": 200785}
-{"epoch": 1.71875, "grad_norm": 0.0004247442411724478, "learning_rate": 1.4898119031716104e-05, "loss": 8.09474295238033e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 440, "tokens/total": 13315632, "tokens/train_per_sec_per_gpu": 36.28, "tokens/trainable": 201267}
-{"epoch": 1.72265625, "grad_norm": 0.0026301892939954996, "learning_rate": 1.476724846845306e-05, "loss": 1.7428235878469422e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 441, "tokens/total": 13346048, "tokens/train_per_sec_per_gpu": 34.16, "tokens/trainable": 201734}
-{"epoch": 1.7265625, "grad_norm": 0.0005576284602284431, "learning_rate": 1.463805215420471e-05, "loss": 8.656044883537106e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00001, "step": 442, "tokens/total": 13376304, "tokens/train_per_sec_per_gpu": 30.38, "tokens/trainable": 202174}
-{"epoch": 1.73046875, "grad_norm": 0.0009604030638001859, "learning_rate": 1.451053546535705e-05, "loss": 1.4092523088038433e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 443, "tokens/total": 13406720, "tokens/train_per_sec_per_gpu": 36.13, "tokens/trainable": 202612}
-{"epoch": 1.734375, "grad_norm": 0.00019884438370354474, "learning_rate": 1.438470370840001e-05, "loss": 4.5837323341402225e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.64, "memory/max_allocated (GiB)": 33.64, "ppl": 1.0, "step": 444, "tokens/total": 13436464, "tokens/train_per_sec_per_gpu": 30.96, "tokens/trainable": 203020}
-{"epoch": 1.73828125, "grad_norm": 0.05944995582103729, "learning_rate": 1.4260562119706606e-05, "loss": 0.0002780454815365374, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00028, "step": 445, "tokens/total": 13466672, "tokens/train_per_sec_per_gpu": 34.31, "tokens/trainable": 203479}
-{"epoch": 1.7421875, "grad_norm": 0.005872223526239395, "learning_rate": 1.413811586531508e-05, "loss": 3.424299575272016e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00003, "step": 446, "tokens/total": 13496736, "tokens/train_per_sec_per_gpu": 35.3, "tokens/trainable": 203946}
-{"epoch": 1.74609375, "grad_norm": 0.006013527978211641, "learning_rate": 1.4017370040713884e-05, "loss": 3.269856097176671e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00003, "step": 447, "tokens/total": 13526928, "tokens/train_per_sec_per_gpu": 37.43, "tokens/trainable": 204382}
-{"epoch": 1.75, "grad_norm": 0.05103042349219322, "learning_rate": 1.3898329670629645e-05, "loss": 0.00032247230410575867, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00032, "step": 448, "tokens/total": 13557392, "tokens/train_per_sec_per_gpu": 34.32, "tokens/trainable": 204830}
-{"epoch": 1.75390625, "grad_norm": 0.13250835239887238, "learning_rate": 1.3780999708818058e-05, "loss": 0.0008940898114815354, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00089, "step": 449, "tokens/total": 13587776, "tokens/train_per_sec_per_gpu": 33.51, "tokens/trainable": 205300}
-{"epoch": 1.7578125, "grad_norm": 0.24642030894756317, "learning_rate": 1.3665385037857758e-05, "loss": 0.0012931979726999998, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00129, "step": 450, "tokens/total": 13618112, "tokens/train_per_sec_per_gpu": 30.62, "tokens/trainable": 205726}
-{"epoch": 1.76171875, "grad_norm": 0.011528799310326576, "learning_rate": 1.3551490468947126e-05, "loss": 6.569598917849362e-05, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00007, "step": 451, "tokens/total": 13648512, "tokens/train_per_sec_per_gpu": 31.55, "tokens/trainable": 206163}
-{"epoch": 1.765625, "grad_norm": 1.0818549394607544, "learning_rate": 1.3439320741704075e-05, "loss": 0.015596212819218636, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01572, "step": 452, "tokens/total": 13678704, "tokens/train_per_sec_per_gpu": 35.41, "tokens/trainable": 206619}
-{"epoch": 1.76953125, "grad_norm": 0.0007326967315748334, "learning_rate": 1.3328880523968808e-05, "loss": 1.2033770872221794e-05, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00001, "step": 453, "tokens/total": 13709232, "tokens/train_per_sec_per_gpu": 37.18, "tokens/trainable": 207101}
-{"epoch": 1.7734375, "grad_norm": 0.0023919539526104927, "learning_rate": 1.3220174411609587e-05, "loss": 1.262133719137637e-05, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 454, "tokens/total": 13739600, "tokens/train_per_sec_per_gpu": 35.69, "tokens/trainable": 207546}
-{"epoch": 1.77734375, "grad_norm": 0.0020972786005586386, "learning_rate": 1.3113206928331471e-05, "loss": 2.476977533660829e-05, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 455, "tokens/total": 13769936, "tokens/train_per_sec_per_gpu": 34.25, "tokens/trainable": 207989}
-{"epoch": 1.78125, "grad_norm": 0.000797569751739502, "learning_rate": 1.300798252548806e-05, "loss": 1.45716385304695e-05, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00001, "step": 456, "tokens/total": 13800240, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 208402}
-{"epoch": 1.78515625, "grad_norm": 0.011376727372407913, "learning_rate": 1.2904505581896265e-05, "loss": 7.903270306997001e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00008, "step": 457, "tokens/total": 13830512, "tokens/train_per_sec_per_gpu": 36.61, "tokens/trainable": 208904}
-{"epoch": 1.7890625, "grad_norm": 0.0013566205743700266, "learning_rate": 1.2802780403654082e-05, "loss": 1.925312972161919e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 458, "tokens/total": 13860832, "tokens/train_per_sec_per_gpu": 33.64, "tokens/trainable": 209382}
-{"epoch": 1.79296875, "grad_norm": 0.00047398527385666966, "learning_rate": 1.2702811223961408e-05, "loss": 1.119351145462133e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00001, "step": 459, "tokens/total": 13890992, "tokens/train_per_sec_per_gpu": 33.85, "tokens/trainable": 209833}
-{"epoch": 1.796875, "grad_norm": 0.04677487164735794, "learning_rate": 1.2604602202943861e-05, "loss": 0.0001750149531289935, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00018, "step": 460, "tokens/total": 13921424, "tokens/train_per_sec_per_gpu": 32.5, "tokens/trainable": 210284}
-{"epoch": 1.80078125, "grad_norm": 0.024663345888257027, "learning_rate": 1.2508157427479686e-05, "loss": 9.222347580362111e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00009, "step": 461, "tokens/total": 13951504, "tokens/train_per_sec_per_gpu": 34.04, "tokens/trainable": 210769}
-{"epoch": 1.8046875, "grad_norm": 0.0004740248841699213, "learning_rate": 1.2413480911029655e-05, "loss": 1.0696679964894429e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.38, "memory/max_allocated (GiB)": 33.38, "ppl": 1.00001, "step": 462, "tokens/total": 13979728, "tokens/train_per_sec_per_gpu": 30.38, "tokens/trainable": 211193}
-{"epoch": 1.80859375, "grad_norm": 0.0007590787718072534, "learning_rate": 1.2320576593470082e-05, "loss": 1.3936740288045257e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00001, "step": 463, "tokens/total": 14009936, "tokens/train_per_sec_per_gpu": 35.03, "tokens/trainable": 211655}
-{"epoch": 1.8125, "grad_norm": 0.002108624204993248, "learning_rate": 1.2229448340928828e-05, "loss": 2.675112227734644e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00003, "step": 464, "tokens/total": 14039872, "tokens/train_per_sec_per_gpu": 29.28, "tokens/trainable": 212085}
-{"epoch": 1.81640625, "grad_norm": 0.0009435649262741208, "learning_rate": 1.2140099945624458e-05, "loss": 1.0821668183780275e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00001, "step": 465, "tokens/total": 14070096, "tokens/train_per_sec_per_gpu": 32.61, "tokens/trainable": 212570}
-{"epoch": 1.8203125, "grad_norm": 0.011040126904845238, "learning_rate": 1.205253512570841e-05, "loss": 5.798249912913889e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00006, "step": 466, "tokens/total": 14100192, "tokens/train_per_sec_per_gpu": 32.86, "tokens/trainable": 213011}
-{"epoch": 1.82421875, "grad_norm": 0.008271797560155392, "learning_rate": 1.1966757525110255e-05, "loss": 2.8763912268914282e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00003, "step": 467, "tokens/total": 14130448, "tokens/train_per_sec_per_gpu": 31.44, "tokens/trainable": 213450}
-{"epoch": 1.828125, "grad_norm": 0.023479726165533066, "learning_rate": 1.1882770713386095e-05, "loss": 0.0001434558507753536, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00014, "step": 468, "tokens/total": 14160480, "tokens/train_per_sec_per_gpu": 35.62, "tokens/trainable": 213898}
-{"epoch": 1.83203125, "grad_norm": 0.0023077642545104027, "learning_rate": 1.180057818556998e-05, "loss": 2.560112625360489e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00003, "step": 469, "tokens/total": 14191152, "tokens/train_per_sec_per_gpu": 36.78, "tokens/trainable": 214366}
-{"epoch": 1.8359375, "grad_norm": 0.0014362182701006532, "learning_rate": 1.1720183362028494e-05, "loss": 2.0195953766233288e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00002, "step": 470, "tokens/total": 14219392, "tokens/train_per_sec_per_gpu": 35.64, "tokens/trainable": 214800}
-{"epoch": 1.83984375, "grad_norm": 0.011598445475101471, "learning_rate": 1.1641589588318387e-05, "loss": 4.486892794375308e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00004, "step": 471, "tokens/total": 14249920, "tokens/train_per_sec_per_gpu": 35.35, "tokens/trainable": 215292}
-{"epoch": 1.84375, "grad_norm": 0.0011633927933871746, "learning_rate": 1.1564800135047418e-05, "loss": 2.0482253603404388e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00002, "step": 472, "tokens/total": 14280048, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 215741}
-{"epoch": 1.84765625, "grad_norm": 0.006093664560467005, "learning_rate": 1.148981819773816e-05, "loss": 5.0086440751329064e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00005, "step": 473, "tokens/total": 14310432, "tokens/train_per_sec_per_gpu": 34.64, "tokens/trainable": 216194}
-{"epoch": 1.8515625, "grad_norm": 0.003623088588938117, "learning_rate": 1.1416646896695086e-05, "loss": 3.059494338231161e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00003, "step": 474, "tokens/total": 14340496, "tokens/train_per_sec_per_gpu": 33.3, "tokens/trainable": 216634}
-{"epoch": 1.85546875, "grad_norm": 0.006924150045961142, "learning_rate": 1.1345289276874717e-05, "loss": 3.953809573431499e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 475, "tokens/total": 14370720, "tokens/train_per_sec_per_gpu": 30.48, "tokens/trainable": 217040}
-{"epoch": 1.859375, "grad_norm": 0.002359601203352213, "learning_rate": 1.1275748307758873e-05, "loss": 3.34192045556847e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00003, "step": 476, "tokens/total": 14401136, "tokens/train_per_sec_per_gpu": 37.55, "tokens/trainable": 217530}
-{"epoch": 1.86328125, "grad_norm": 0.009584639221429825, "learning_rate": 1.1208026883231147e-05, "loss": 0.00010110227594850585, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.0001, "step": 477, "tokens/total": 14431360, "tokens/train_per_sec_per_gpu": 34.47, "tokens/trainable": 217988}
-{"epoch": 1.8671875, "grad_norm": 0.03044246695935726, "learning_rate": 1.1142127821456433e-05, "loss": 0.00022086883836891502, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00022, "step": 478, "tokens/total": 14461968, "tokens/train_per_sec_per_gpu": 32.47, "tokens/trainable": 218461}
-{"epoch": 1.87109375, "grad_norm": 0.03494428098201752, "learning_rate": 1.1078053864763674e-05, "loss": 0.0002753190929070115, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00028, "step": 479, "tokens/total": 14491984, "tokens/train_per_sec_per_gpu": 32.77, "tokens/trainable": 218909}
-{"epoch": 1.875, "grad_norm": 0.30102044343948364, "learning_rate": 1.1015807679531756e-05, "loss": 0.00025764034944586456, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00026, "step": 480, "tokens/total": 14522544, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 219350}
-{"epoch": 1.87890625, "grad_norm": 0.07505832612514496, "learning_rate": 1.0955391856078528e-05, "loss": 0.0002503654104657471, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00025, "step": 481, "tokens/total": 14552560, "tokens/train_per_sec_per_gpu": 29.47, "tokens/trainable": 219776}
-{"epoch": 1.8828125, "grad_norm": 0.0008993322844617069, "learning_rate": 1.0896808908553007e-05, "loss": 1.0104758075613063e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 482, "tokens/total": 14583056, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 220284}
-{"epoch": 1.88671875, "grad_norm": 0.001677343505434692, "learning_rate": 1.0840061274830763e-05, "loss": 2.5296727471868508e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00003, "step": 483, "tokens/total": 14613152, "tokens/train_per_sec_per_gpu": 33.01, "tokens/trainable": 220714}
-{"epoch": 1.890625, "grad_norm": 0.01496347226202488, "learning_rate": 1.0785151316412473e-05, "loss": 0.0001352597464574501, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00014, "step": 484, "tokens/total": 14643328, "tokens/train_per_sec_per_gpu": 35.01, "tokens/trainable": 221156}
-{"epoch": 1.89453125, "grad_norm": 0.004265300463885069, "learning_rate": 1.0732081318325639e-05, "loss": 1.1048326996387914e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00001, "step": 485, "tokens/total": 14673488, "tokens/train_per_sec_per_gpu": 31.53, "tokens/trainable": 221605}
-{"epoch": 1.8984375, "grad_norm": 0.0029735390562564135, "learning_rate": 1.0680853489029501e-05, "loss": 2.2016622096998617e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 486, "tokens/total": 14704000, "tokens/train_per_sec_per_gpu": 30.38, "tokens/trainable": 222042}
-{"epoch": 1.90234375, "grad_norm": 0.00869164988398552, "learning_rate": 1.0631469960323152e-05, "loss": 7.451521378243342e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.67, "memory/max_allocated (GiB)": 33.67, "ppl": 1.00007, "step": 487, "tokens/total": 14733984, "tokens/train_per_sec_per_gpu": 31.95, "tokens/trainable": 222474}
-{"epoch": 1.90625, "grad_norm": 0.000581100583076477, "learning_rate": 1.0583932787256783e-05, "loss": 7.341123364312807e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00001, "step": 488, "tokens/total": 14764400, "tokens/train_per_sec_per_gpu": 36.24, "tokens/trainable": 222967}
-{"epoch": 1.91015625, "grad_norm": 0.004224246367812157, "learning_rate": 1.0538243948046206e-05, "loss": 3.5561904951464385e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00004, "step": 489, "tokens/total": 14794672, "tokens/train_per_sec_per_gpu": 31.78, "tokens/trainable": 223420}
-{"epoch": 1.9140625, "grad_norm": 0.0152070177718997, "learning_rate": 1.0494405343990523e-05, "loss": 2.7988677175017074e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00003, "step": 490, "tokens/total": 14824784, "tokens/train_per_sec_per_gpu": 29.42, "tokens/trainable": 223864}
-{"epoch": 1.91796875, "grad_norm": 0.024472137913107872, "learning_rate": 1.0452418799392985e-05, "loss": 4.8069461627164856e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00005, "step": 491, "tokens/total": 14855056, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 224321}
-{"epoch": 1.921875, "grad_norm": 0.0036334518808871508, "learning_rate": 1.0412286061485102e-05, "loss": 3.39774924213998e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.00003, "step": 492, "tokens/total": 14885392, "tokens/train_per_sec_per_gpu": 36.29, "tokens/trainable": 224802}
-{"epoch": 1.92578125, "grad_norm": 0.0009831058559939265, "learning_rate": 1.03740088003539e-05, "loss": 1.8870396161219105e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 493, "tokens/total": 14915760, "tokens/train_per_sec_per_gpu": 34.77, "tokens/trainable": 225284}
-{"epoch": 1.9296875, "grad_norm": 0.0020358245819807053, "learning_rate": 1.0337588608872463e-05, "loss": 3.0332268579513766e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00003, "step": 494, "tokens/total": 14946192, "tokens/train_per_sec_per_gpu": 32.71, "tokens/trainable": 225733}
-{"epoch": 1.93359375, "grad_norm": 0.00029036635532975197, "learning_rate": 1.0303027002633622e-05, "loss": 6.913277957210084e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00001, "step": 495, "tokens/total": 14976272, "tokens/train_per_sec_per_gpu": 32.55, "tokens/trainable": 226172}
-{"epoch": 1.9375, "grad_norm": 0.11670258641242981, "learning_rate": 1.0270325419886884e-05, "loss": 0.0007051021093502641, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00071, "step": 496, "tokens/total": 15006544, "tokens/train_per_sec_per_gpu": 34.77, "tokens/trainable": 226636}
-{"epoch": 1.94140625, "grad_norm": 0.0008452775655314326, "learning_rate": 1.0239485221478599e-05, "loss": 1.0465846571605653e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00001, "step": 497, "tokens/total": 15037040, "tokens/train_per_sec_per_gpu": 32.91, "tokens/trainable": 227067}
-{"epoch": 1.9453125, "grad_norm": 0.027690444141626358, "learning_rate": 1.0210507690795292e-05, "loss": 0.00013966832193545997, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00014, "step": 498, "tokens/total": 15067344, "tokens/train_per_sec_per_gpu": 37.58, "tokens/trainable": 227532}
-{"epoch": 1.94921875, "grad_norm": 0.24434912204742432, "learning_rate": 1.0183394033710305e-05, "loss": 0.0015483510214835405, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00155, "step": 499, "tokens/total": 15097648, "tokens/train_per_sec_per_gpu": 36.58, "tokens/trainable": 228007}
-{"epoch": 1.953125, "grad_norm": 0.0016334869433194399, "learning_rate": 1.0158145378533583e-05, "loss": 1.879910814750474e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 500, "tokens/total": 15128000, "tokens/train_per_sec_per_gpu": 30.84, "tokens/trainable": 228457}
-{"epoch": 1.95703125, "grad_norm": 0.004102619830518961, "learning_rate": 1.0134762775964726e-05, "loss": 3.3614764106459916e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00003, "step": 501, "tokens/total": 15158352, "tokens/train_per_sec_per_gpu": 35.19, "tokens/trainable": 228928}
-{"epoch": 1.9609375, "grad_norm": 0.00031376321567222476, "learning_rate": 1.0113247199049278e-05, "loss": 6.558944733114913e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00001, "step": 502, "tokens/total": 15189136, "tokens/train_per_sec_per_gpu": 32.1, "tokens/trainable": 229368}
-{"epoch": 1.96484375, "grad_norm": 0.002323499647900462, "learning_rate": 1.0093599543138205e-05, "loss": 1.713021265459247e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 503, "tokens/total": 15219616, "tokens/train_per_sec_per_gpu": 32.75, "tokens/trainable": 229815}
-{"epoch": 1.96875, "grad_norm": 0.19251975417137146, "learning_rate": 1.0075820625850675e-05, "loss": 0.002064360538497567, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00207, "step": 504, "tokens/total": 15250032, "tokens/train_per_sec_per_gpu": 38.82, "tokens/trainable": 230319}
-{"epoch": 1.97265625, "grad_norm": 0.028361506760120392, "learning_rate": 1.0059911187040013e-05, "loss": 0.0001954930485226214, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.0002, "step": 505, "tokens/total": 15280592, "tokens/train_per_sec_per_gpu": 37.42, "tokens/trainable": 230807}
-{"epoch": 1.9765625, "grad_norm": 0.0009314758353866637, "learning_rate": 1.0045871888762893e-05, "loss": 1.3068352927803062e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00001, "step": 506, "tokens/total": 15310816, "tokens/train_per_sec_per_gpu": 34.16, "tokens/trainable": 231257}
-{"epoch": 1.98046875, "grad_norm": 0.043847255408763885, "learning_rate": 1.003370331525184e-05, "loss": 0.00034106988459825516, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00034, "step": 507, "tokens/total": 15341424, "tokens/train_per_sec_per_gpu": 36.31, "tokens/trainable": 231746}
-{"epoch": 1.984375, "grad_norm": 0.0003455990517977625, "learning_rate": 1.002340597289085e-05, "loss": 7.529959020757815e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.33, "memory/max_allocated (GiB)": 33.33, "ppl": 1.00001, "step": 508, "tokens/total": 15369664, "tokens/train_per_sec_per_gpu": 32.94, "tokens/trainable": 232172}
-{"epoch": 1.98828125, "grad_norm": 0.034262772649526596, "learning_rate": 1.0014980290194387e-05, "loss": 0.00015496168634854257, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.00015, "step": 509, "tokens/total": 15400496, "tokens/train_per_sec_per_gpu": 37.04, "tokens/trainable": 232680}
-{"epoch": 1.9921875, "grad_norm": 0.0054254732094705105, "learning_rate": 1.0008426617789489e-05, "loss": 1.3866054359823465e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 510, "tokens/total": 15430864, "tokens/train_per_sec_per_gpu": 36.02, "tokens/trainable": 233169}
-{"epoch": 1.99609375, "grad_norm": 0.044709429144859314, "learning_rate": 1.0003745228401215e-05, "loss": 0.0001455950696254149, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00015, "step": 511, "tokens/total": 15461232, "tokens/train_per_sec_per_gpu": 32.63, "tokens/trainable": 233609}
-{"epoch": 2.0, "grad_norm": 0.0011775761377066374, "learning_rate": 1.0000936316841296e-05, "loss": 2.066101660602726e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 512, "tokens/total": 15491760, "tokens/train_per_sec_per_gpu": 36.16, "tokens/trainable": 234078}
+{"epoch": 0.00390625, "grad_norm": 1.3783645629882812, "learning_rate": 0.0, "loss": 0.13924157619476318, "memory/device_reserved (GiB)": 33.9, "memory/max_active (GiB)": 32.79, "memory/max_allocated (GiB)": 32.79, "ppl": 1.1494, "step": 1, "tokens/total": 30176, "tokens/train_per_sec_per_gpu": 27.09, "tokens/trainable": 417}
+{"epoch": 0.0078125, "grad_norm": 1.2333115339279175, "learning_rate": 4.000000000000001e-06, "loss": 0.15098872780799866, "memory/device_reserved (GiB)": 34.68, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.16298, "step": 2, "tokens/total": 60272, "tokens/train_per_sec_per_gpu": 34.63, "tokens/trainable": 873}
+{"epoch": 0.01171875, "grad_norm": 5.272272109985352, "learning_rate": 8.000000000000001e-06, "loss": 0.14415010809898376, "memory/device_reserved (GiB)": 34.72, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.15506, "step": 3, "tokens/total": 90512, "tokens/train_per_sec_per_gpu": 34.94, "tokens/trainable": 1353}
+{"epoch": 0.015625, "grad_norm": 1.3964842557907104, "learning_rate": 1.2e-05, "loss": 0.13428445160388947, "memory/device_reserved (GiB)": 34.73, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.14372, "step": 4, "tokens/total": 120944, "tokens/train_per_sec_per_gpu": 31.19, "tokens/trainable": 1778}
+{"epoch": 0.01953125, "grad_norm": 1.4071121215820312, "learning_rate": 1.6000000000000003e-05, "loss": 0.14405930042266846, "memory/device_reserved (GiB)": 35.12, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.15495, "step": 5, "tokens/total": 151440, "tokens/train_per_sec_per_gpu": 36.8, "tokens/trainable": 2261}
+{"epoch": 0.0234375, "grad_norm": 1.6760321855545044, "learning_rate": 2e-05, "loss": 0.11100607365369797, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.1174, "step": 6, "tokens/total": 181984, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 2705}
+{"epoch": 0.02734375, "grad_norm": 1.3056405782699585, "learning_rate": 2.4e-05, "loss": 0.08474013209342957, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.08843, "step": 7, "tokens/total": 212336, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 3136}
+{"epoch": 0.03125, "grad_norm": 15.479830741882324, "learning_rate": 2.8000000000000003e-05, "loss": 0.08459445834159851, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.08828, "step": 8, "tokens/total": 242592, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 3579}
+{"epoch": 0.03515625, "grad_norm": 1.3126204013824463, "learning_rate": 3.2000000000000005e-05, "loss": 0.06938324868679047, "memory/device_reserved (GiB)": 36.12, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.07185, "step": 9, "tokens/total": 272784, "tokens/train_per_sec_per_gpu": 29.12, "tokens/trainable": 4000}
+{"epoch": 0.0390625, "grad_norm": 1.292763590812683, "learning_rate": 3.6e-05, "loss": 0.04889247566461563, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.05011, "step": 10, "tokens/total": 303184, "tokens/train_per_sec_per_gpu": 37.66, "tokens/trainable": 4474}
+{"epoch": 0.04296875, "grad_norm": 1.0318297147750854, "learning_rate": 4e-05, "loss": 0.030023720115423203, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.03048, "step": 11, "tokens/total": 333296, "tokens/train_per_sec_per_gpu": 33.15, "tokens/trainable": 4942}
+{"epoch": 0.046875, "grad_norm": 1.4940381050109863, "learning_rate": 4.4000000000000006e-05, "loss": 0.05268975719809532, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0541, "step": 12, "tokens/total": 363840, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 5424}
+{"epoch": 0.05078125, "grad_norm": 2.0244202613830566, "learning_rate": 4.8e-05, "loss": 0.06068706512451172, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.06257, "step": 13, "tokens/total": 394112, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 5855}
+{"epoch": 0.0546875, "grad_norm": 1.5167357921600342, "learning_rate": 5.2000000000000004e-05, "loss": 0.03630872070789337, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.03698, "step": 14, "tokens/total": 424656, "tokens/train_per_sec_per_gpu": 34.75, "tokens/trainable": 6344}
+{"epoch": 0.05859375, "grad_norm": 1.7407020330429077, "learning_rate": 5.6000000000000006e-05, "loss": 0.056628547608852386, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.05826, "step": 15, "tokens/total": 455152, "tokens/train_per_sec_per_gpu": 36.1, "tokens/trainable": 6825}
+{"epoch": 0.0625, "grad_norm": 2.506803512573242, "learning_rate": 6e-05, "loss": 0.07074552774429321, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.07331, "step": 16, "tokens/total": 485328, "tokens/train_per_sec_per_gpu": 30.03, "tokens/trainable": 7230}
+{"epoch": 0.06640625, "grad_norm": 1.2572238445281982, "learning_rate": 6.400000000000001e-05, "loss": 0.04099711775779724, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.04185, "step": 17, "tokens/total": 515744, "tokens/train_per_sec_per_gpu": 35.91, "tokens/trainable": 7699}
+{"epoch": 0.0703125, "grad_norm": 36.9443359375, "learning_rate": 6.800000000000001e-05, "loss": 0.021194521337747574, "memory/device_reserved (GiB)": 37.81, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.02142, "step": 18, "tokens/total": 546144, "tokens/train_per_sec_per_gpu": 33.78, "tokens/trainable": 8184}
+{"epoch": 0.07421875, "grad_norm": 0.377701997756958, "learning_rate": 7.2e-05, "loss": 0.013856083154678345, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01395, "step": 19, "tokens/total": 576560, "tokens/train_per_sec_per_gpu": 34.89, "tokens/trainable": 8675}
+{"epoch": 0.078125, "grad_norm": 1.426601529121399, "learning_rate": 7.6e-05, "loss": 0.0455022007226944, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.04655, "step": 20, "tokens/total": 607056, "tokens/train_per_sec_per_gpu": 39.72, "tokens/trainable": 9189}
+{"epoch": 0.08203125, "grad_norm": 0.56149822473526, "learning_rate": 8e-05, "loss": 0.03209531307220459, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.03262, "step": 21, "tokens/total": 637488, "tokens/train_per_sec_per_gpu": 31.61, "tokens/trainable": 9614}
+{"epoch": 0.0859375, "grad_norm": 1.1531219482421875, "learning_rate": 8.4e-05, "loss": 0.04279206693172455, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.04372, "step": 22, "tokens/total": 667632, "tokens/train_per_sec_per_gpu": 36.11, "tokens/trainable": 10086}
+{"epoch": 0.08984375, "grad_norm": 0.9407010078430176, "learning_rate": 8.800000000000001e-05, "loss": 0.01938648894429207, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01958, "step": 23, "tokens/total": 697872, "tokens/train_per_sec_per_gpu": 32.88, "tokens/trainable": 10535}
+{"epoch": 0.09375, "grad_norm": 0.3499844968318939, "learning_rate": 9.200000000000001e-05, "loss": 0.01257958635687828, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.01266, "step": 24, "tokens/total": 728432, "tokens/train_per_sec_per_gpu": 31.52, "tokens/trainable": 10993}
+{"epoch": 0.09765625, "grad_norm": 1.6778419017791748, "learning_rate": 9.6e-05, "loss": 0.04129436984658241, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.04216, "step": 25, "tokens/total": 756656, "tokens/train_per_sec_per_gpu": 39.78, "tokens/trainable": 11417}
+{"epoch": 0.1015625, "grad_norm": 0.5760903358459473, "learning_rate": 0.0001, "loss": 0.008185511454939842, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00822, "step": 26, "tokens/total": 787120, "tokens/train_per_sec_per_gpu": 34.31, "tokens/trainable": 11874}
+{"epoch": 0.10546875, "grad_norm": 5.235631942749023, "learning_rate": 9.99990636831587e-05, "loss": 0.010828999802470207, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.01089, "step": 27, "tokens/total": 817504, "tokens/train_per_sec_per_gpu": 31.62, "tokens/trainable": 12316}
+{"epoch": 0.109375, "grad_norm": 1.404441475868225, "learning_rate": 9.999625477159879e-05, "loss": 0.051578108221292496, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.05293, "step": 28, "tokens/total": 848128, "tokens/train_per_sec_per_gpu": 37.76, "tokens/trainable": 12800}
+{"epoch": 0.11328125, "grad_norm": 3.0178139209747314, "learning_rate": 9.999157338221051e-05, "loss": 0.051813721656799316, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.05318, "step": 29, "tokens/total": 878448, "tokens/train_per_sec_per_gpu": 35.43, "tokens/trainable": 13270}
+{"epoch": 0.1171875, "grad_norm": 1.413711428642273, "learning_rate": 9.998501970980562e-05, "loss": 0.025016000494360924, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.02533, "step": 30, "tokens/total": 909008, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 13752}
+{"epoch": 0.12109375, "grad_norm": 1.1610053777694702, "learning_rate": 9.997659402710915e-05, "loss": 0.008201262913644314, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00823, "step": 31, "tokens/total": 939312, "tokens/train_per_sec_per_gpu": 34.0, "tokens/trainable": 14218}
+{"epoch": 0.125, "grad_norm": 0.2630869448184967, "learning_rate": 9.996629668474818e-05, "loss": 0.006646217778325081, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00667, "step": 32, "tokens/total": 969264, "tokens/train_per_sec_per_gpu": 34.38, "tokens/trainable": 14655}
+{"epoch": 0.12890625, "grad_norm": 0.7988011240959167, "learning_rate": 9.995412811123711e-05, "loss": 0.017185213044285774, "memory/device_reserved (GiB)": 37.82, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.01733, "step": 33, "tokens/total": 999792, "tokens/train_per_sec_per_gpu": 33.23, "tokens/trainable": 15087}
+{"epoch": 0.1328125, "grad_norm": 0.5069737434387207, "learning_rate": 9.994008881295999e-05, "loss": 0.02045844867825508, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.02067, "step": 34, "tokens/total": 1029840, "tokens/train_per_sec_per_gpu": 33.81, "tokens/trainable": 15548}
+{"epoch": 0.13671875, "grad_norm": 0.8837020993232727, "learning_rate": 9.992417937414932e-05, "loss": 0.024748671799898148, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.02506, "step": 35, "tokens/total": 1060416, "tokens/train_per_sec_per_gpu": 33.69, "tokens/trainable": 16037}
+{"epoch": 0.140625, "grad_norm": 0.7192158102989197, "learning_rate": 9.99064004568618e-05, "loss": 0.0077397022396326065, "memory/device_reserved (GiB)": 35.48, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00777, "step": 36, "tokens/total": 1090464, "tokens/train_per_sec_per_gpu": 31.0, "tokens/trainable": 16475}
+{"epoch": 0.14453125, "grad_norm": 0.8887882232666016, "learning_rate": 9.988675280095074e-05, "loss": 0.024969205260276794, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.02528, "step": 37, "tokens/total": 1120944, "tokens/train_per_sec_per_gpu": 34.32, "tokens/trainable": 16919}
+{"epoch": 0.1484375, "grad_norm": 1.0087132453918457, "learning_rate": 9.986523722403528e-05, "loss": 0.02261659875512123, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.02287, "step": 38, "tokens/total": 1151360, "tokens/train_per_sec_per_gpu": 28.9, "tokens/trainable": 17333}
+{"epoch": 0.15234375, "grad_norm": 0.5544025897979736, "learning_rate": 9.984185462146642e-05, "loss": 0.009409069083631039, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00945, "step": 39, "tokens/total": 1181728, "tokens/train_per_sec_per_gpu": 36.09, "tokens/trainable": 17806}
+{"epoch": 0.15625, "grad_norm": 0.6828466653823853, "learning_rate": 9.98166059662897e-05, "loss": 0.015685085207223892, "memory/device_reserved (GiB)": 35.86, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01581, "step": 40, "tokens/total": 1212064, "tokens/train_per_sec_per_gpu": 35.58, "tokens/trainable": 18290}
+{"epoch": 0.16015625, "grad_norm": 0.32031646370887756, "learning_rate": 9.978949230920472e-05, "loss": 0.006071081385016441, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00609, "step": 41, "tokens/total": 1242592, "tokens/train_per_sec_per_gpu": 31.93, "tokens/trainable": 18747}
+{"epoch": 0.1640625, "grad_norm": 1.35106360912323, "learning_rate": 9.976051477852141e-05, "loss": 0.038804396986961365, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.03957, "step": 42, "tokens/total": 1272800, "tokens/train_per_sec_per_gpu": 31.83, "tokens/trainable": 19172}
+{"epoch": 0.16796875, "grad_norm": 0.8709872961044312, "learning_rate": 9.972967458011312e-05, "loss": 0.019240852445364, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 34.04, "memory/max_allocated (GiB)": 34.04, "ppl": 1.01943, "step": 43, "tokens/total": 1303600, "tokens/train_per_sec_per_gpu": 31.21, "tokens/trainable": 19622}
+{"epoch": 0.171875, "grad_norm": 0.7335953116416931, "learning_rate": 9.96969729973664e-05, "loss": 0.012263014912605286, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.01234, "step": 44, "tokens/total": 1333936, "tokens/train_per_sec_per_gpu": 33.31, "tokens/trainable": 20073}
+{"epoch": 0.17578125, "grad_norm": 1.9549319744110107, "learning_rate": 9.966241139112754e-05, "loss": 0.012868007645010948, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01295, "step": 45, "tokens/total": 1364288, "tokens/train_per_sec_per_gpu": 31.46, "tokens/trainable": 20497}
+{"epoch": 0.1796875, "grad_norm": 0.5352866053581238, "learning_rate": 9.96259911996461e-05, "loss": 0.006207342725247145, "memory/device_reserved (GiB)": 36.03, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00623, "step": 46, "tokens/total": 1394368, "tokens/train_per_sec_per_gpu": 34.21, "tokens/trainable": 20940}
+{"epoch": 0.18359375, "grad_norm": 0.6218534708023071, "learning_rate": 9.958771393851491e-05, "loss": 0.006636852398514748, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00666, "step": 47, "tokens/total": 1425024, "tokens/train_per_sec_per_gpu": 34.94, "tokens/trainable": 21405}
+{"epoch": 0.1875, "grad_norm": 0.7587782144546509, "learning_rate": 9.954758120060702e-05, "loss": 0.008137413300573826, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00817, "step": 48, "tokens/total": 1455392, "tokens/train_per_sec_per_gpu": 35.03, "tokens/trainable": 21828}
+{"epoch": 0.19140625, "grad_norm": 0.6957025527954102, "learning_rate": 9.950559465600948e-05, "loss": 0.019302664324641228, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01949, "step": 49, "tokens/total": 1485616, "tokens/train_per_sec_per_gpu": 29.43, "tokens/trainable": 22241}
+{"epoch": 0.1953125, "grad_norm": 0.7795535922050476, "learning_rate": 9.946175605195379e-05, "loss": 0.0063660056330263615, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00639, "step": 50, "tokens/total": 1515984, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 22688}
+{"epoch": 0.19921875, "grad_norm": 0.20801189541816711, "learning_rate": 9.941606721274322e-05, "loss": 0.0021921340376138687, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00219, "step": 51, "tokens/total": 1546224, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 23136}
+{"epoch": 0.203125, "grad_norm": 0.7914130687713623, "learning_rate": 9.936853003967685e-05, "loss": 0.012165859341621399, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01224, "step": 52, "tokens/total": 1576528, "tokens/train_per_sec_per_gpu": 36.92, "tokens/trainable": 23631}
+{"epoch": 0.20703125, "grad_norm": 0.7185864448547363, "learning_rate": 9.93191465109705e-05, "loss": 0.008950375951826572, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00899, "step": 53, "tokens/total": 1606768, "tokens/train_per_sec_per_gpu": 34.05, "tokens/trainable": 24096}
+{"epoch": 0.2109375, "grad_norm": 0.8439444303512573, "learning_rate": 9.926791868167438e-05, "loss": 0.011410758830606937, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.01148, "step": 54, "tokens/total": 1636640, "tokens/train_per_sec_per_gpu": 37.77, "tokens/trainable": 24569}
+{"epoch": 0.21484375, "grad_norm": 0.47947707772254944, "learning_rate": 9.921484868358753e-05, "loss": 0.004429137334227562, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00444, "step": 55, "tokens/total": 1667136, "tokens/train_per_sec_per_gpu": 33.86, "tokens/trainable": 25067}
+{"epoch": 0.21875, "grad_norm": 0.4257976710796356, "learning_rate": 9.915993872516924e-05, "loss": 0.00340340007096529, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00341, "step": 56, "tokens/total": 1697472, "tokens/train_per_sec_per_gpu": 36.47, "tokens/trainable": 25541}
+{"epoch": 0.22265625, "grad_norm": 0.7323225736618042, "learning_rate": 9.9103191091447e-05, "loss": 0.010487878695130348, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01054, "step": 57, "tokens/total": 1727680, "tokens/train_per_sec_per_gpu": 33.08, "tokens/trainable": 25965}
+{"epoch": 0.2265625, "grad_norm": 2.3697221279144287, "learning_rate": 9.904460814392147e-05, "loss": 0.004758198279887438, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00477, "step": 58, "tokens/total": 1758192, "tokens/train_per_sec_per_gpu": 33.79, "tokens/trainable": 26409}
+{"epoch": 0.23046875, "grad_norm": 1.732944369316101, "learning_rate": 9.898419232046825e-05, "loss": 0.00296807405538857, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00297, "step": 59, "tokens/total": 1788816, "tokens/train_per_sec_per_gpu": 33.99, "tokens/trainable": 26883}
+{"epoch": 0.234375, "grad_norm": 4.904305934906006, "learning_rate": 9.892194613523633e-05, "loss": 0.005324847996234894, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00534, "step": 60, "tokens/total": 1819056, "tokens/train_per_sec_per_gpu": 33.67, "tokens/trainable": 27323}
+{"epoch": 0.23828125, "grad_norm": 0.08419182896614075, "learning_rate": 9.885787217854357e-05, "loss": 0.0007892341818660498, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00079, "step": 61, "tokens/total": 1849664, "tokens/train_per_sec_per_gpu": 39.09, "tokens/trainable": 27830}
+{"epoch": 0.2421875, "grad_norm": 0.7355648279190063, "learning_rate": 9.879197311676887e-05, "loss": 0.017353510484099388, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.0175, "step": 62, "tokens/total": 1880272, "tokens/train_per_sec_per_gpu": 32.21, "tokens/trainable": 28281}
+{"epoch": 0.24609375, "grad_norm": 0.925104558467865, "learning_rate": 9.872425169224113e-05, "loss": 0.01441475935280323, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01452, "step": 63, "tokens/total": 1910752, "tokens/train_per_sec_per_gpu": 32.16, "tokens/trainable": 28742}
+{"epoch": 0.25, "grad_norm": 0.2562558352947235, "learning_rate": 9.865471072312528e-05, "loss": 0.0025585039984434843, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00256, "step": 64, "tokens/total": 1940848, "tokens/train_per_sec_per_gpu": 33.54, "tokens/trainable": 29183}
+{"epoch": 0.25390625, "grad_norm": 0.5092302560806274, "learning_rate": 9.858335310330492e-05, "loss": 0.007829459384083748, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00786, "step": 65, "tokens/total": 1971344, "tokens/train_per_sec_per_gpu": 31.45, "tokens/trainable": 29648}
+{"epoch": 0.2578125, "grad_norm": 0.5972490310668945, "learning_rate": 9.851018180226185e-05, "loss": 0.0017629603389650583, "memory/device_reserved (GiB)": 35.06, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00176, "step": 66, "tokens/total": 2001712, "tokens/train_per_sec_per_gpu": 33.44, "tokens/trainable": 30075}
+{"epoch": 0.26171875, "grad_norm": 0.6682910919189453, "learning_rate": 9.843519986495259e-05, "loss": 0.005682674702256918, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.0057, "step": 67, "tokens/total": 2029936, "tokens/train_per_sec_per_gpu": 37.53, "tokens/trainable": 30546}
+{"epoch": 0.265625, "grad_norm": 2.3002567291259766, "learning_rate": 9.835841041168162e-05, "loss": 0.010535864159464836, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01059, "step": 68, "tokens/total": 2060384, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 31025}
+{"epoch": 0.26953125, "grad_norm": 2.2319464683532715, "learning_rate": 9.82798166379715e-05, "loss": 0.016697486862540245, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.01684, "step": 69, "tokens/total": 2090688, "tokens/train_per_sec_per_gpu": 33.87, "tokens/trainable": 31505}
+{"epoch": 0.2734375, "grad_norm": 0.4549625515937805, "learning_rate": 9.819942181443002e-05, "loss": 0.005516475066542625, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00553, "step": 70, "tokens/total": 2120848, "tokens/train_per_sec_per_gpu": 32.21, "tokens/trainable": 31956}
+{"epoch": 0.27734375, "grad_norm": 0.7483011484146118, "learning_rate": 9.811722928661392e-05, "loss": 0.008852789178490639, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.35, "memory/max_allocated (GiB)": 33.35, "ppl": 1.00889, "step": 71, "tokens/total": 2149120, "tokens/train_per_sec_per_gpu": 33.33, "tokens/trainable": 32374}
+{"epoch": 0.28125, "grad_norm": 0.4358912706375122, "learning_rate": 9.803324247488975e-05, "loss": 0.0014208657667040825, "memory/device_reserved (GiB)": 35.64, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00142, "step": 72, "tokens/total": 2179472, "tokens/train_per_sec_per_gpu": 34.46, "tokens/trainable": 32833}
+{"epoch": 0.28515625, "grad_norm": 0.32716670632362366, "learning_rate": 9.794746487429161e-05, "loss": 0.005963748786598444, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00598, "step": 73, "tokens/total": 2209984, "tokens/train_per_sec_per_gpu": 35.67, "tokens/trainable": 33304}
+{"epoch": 0.2890625, "grad_norm": 0.5767021775245667, "learning_rate": 9.785990005437554e-05, "loss": 0.00536112766712904, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00538, "step": 74, "tokens/total": 2240512, "tokens/train_per_sec_per_gpu": 32.33, "tokens/trainable": 33771}
+{"epoch": 0.29296875, "grad_norm": 0.9307782649993896, "learning_rate": 9.777055165907117e-05, "loss": 0.006907512433826923, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00693, "step": 75, "tokens/total": 2270608, "tokens/train_per_sec_per_gpu": 34.46, "tokens/trainable": 34186}
+{"epoch": 0.296875, "grad_norm": 0.517152726650238, "learning_rate": 9.767942340652993e-05, "loss": 0.0017031149473041296, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.0017, "step": 76, "tokens/total": 2300864, "tokens/train_per_sec_per_gpu": 34.83, "tokens/trainable": 34667}
+{"epoch": 0.30078125, "grad_norm": 4.431775093078613, "learning_rate": 9.758651908897035e-05, "loss": 0.01024378277361393, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.0103, "step": 77, "tokens/total": 2330960, "tokens/train_per_sec_per_gpu": 31.71, "tokens/trainable": 35078}
+{"epoch": 0.3046875, "grad_norm": 1.6990809440612793, "learning_rate": 9.749184257252033e-05, "loss": 0.016885127872228622, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.01703, "step": 78, "tokens/total": 2361584, "tokens/train_per_sec_per_gpu": 34.8, "tokens/trainable": 35550}
+{"epoch": 0.30859375, "grad_norm": 1.1758185625076294, "learning_rate": 9.739539779705614e-05, "loss": 0.005958402995020151, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00598, "step": 79, "tokens/total": 2391936, "tokens/train_per_sec_per_gpu": 29.46, "tokens/trainable": 35977}
+{"epoch": 0.3125, "grad_norm": 2.7611470222473145, "learning_rate": 9.729718877603861e-05, "loss": 0.034006115049123764, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.03459, "step": 80, "tokens/total": 2422432, "tokens/train_per_sec_per_gpu": 38.2, "tokens/trainable": 36462}
+{"epoch": 0.31640625, "grad_norm": 0.017304062843322754, "learning_rate": 9.719721959634592e-05, "loss": 0.00017258770822081715, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00017, "step": 81, "tokens/total": 2452752, "tokens/train_per_sec_per_gpu": 33.18, "tokens/trainable": 36919}
+{"epoch": 0.3203125, "grad_norm": 0.6304571032524109, "learning_rate": 9.709549441810375e-05, "loss": 0.00832848995923996, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.44, "memory/max_allocated (GiB)": 33.44, "ppl": 1.00836, "step": 82, "tokens/total": 2481216, "tokens/train_per_sec_per_gpu": 33.4, "tokens/trainable": 37379}
+{"epoch": 0.32421875, "grad_norm": 0.08372113853693008, "learning_rate": 9.699201747451195e-05, "loss": 0.0007960593793541193, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.0008, "step": 83, "tokens/total": 2511488, "tokens/train_per_sec_per_gpu": 33.05, "tokens/trainable": 37832}
+{"epoch": 0.328125, "grad_norm": 0.4056459367275238, "learning_rate": 9.688679307166854e-05, "loss": 0.0015380227705463767, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00154, "step": 84, "tokens/total": 2541984, "tokens/train_per_sec_per_gpu": 33.87, "tokens/trainable": 38303}
+{"epoch": 0.33203125, "grad_norm": 0.1656680703163147, "learning_rate": 9.677982558839042e-05, "loss": 0.0007727849297225475, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00077, "step": 85, "tokens/total": 2572224, "tokens/train_per_sec_per_gpu": 33.62, "tokens/trainable": 38758}
+{"epoch": 0.3359375, "grad_norm": 1.7784565687179565, "learning_rate": 9.66711194760312e-05, "loss": 0.00887264870107174, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00891, "step": 86, "tokens/total": 2602704, "tokens/train_per_sec_per_gpu": 30.62, "tokens/trainable": 39179}
+{"epoch": 0.33984375, "grad_norm": 0.6024295687675476, "learning_rate": 9.656067925829593e-05, "loss": 0.013714440166950226, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01381, "step": 87, "tokens/total": 2633072, "tokens/train_per_sec_per_gpu": 35.29, "tokens/trainable": 39679}
+{"epoch": 0.34375, "grad_norm": 3.774395704269409, "learning_rate": 9.644850953105288e-05, "loss": 0.021049978211522102, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.02127, "step": 88, "tokens/total": 2663232, "tokens/train_per_sec_per_gpu": 32.65, "tokens/trainable": 40118}
+{"epoch": 0.34765625, "grad_norm": 0.20895183086395264, "learning_rate": 9.633461496214225e-05, "loss": 0.0025408377405256033, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00254, "step": 89, "tokens/total": 2693696, "tokens/train_per_sec_per_gpu": 32.76, "tokens/trainable": 40571}
+{"epoch": 0.3515625, "grad_norm": 0.3318273425102234, "learning_rate": 9.621900029118195e-05, "loss": 0.0036354511976242065, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00364, "step": 90, "tokens/total": 2723984, "tokens/train_per_sec_per_gpu": 30.75, "tokens/trainable": 40997}
+{"epoch": 0.35546875, "grad_norm": 0.44360974431037903, "learning_rate": 9.610167032937036e-05, "loss": 0.007065630983561277, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00709, "step": 91, "tokens/total": 2754240, "tokens/train_per_sec_per_gpu": 36.46, "tokens/trainable": 41462}
+{"epoch": 0.359375, "grad_norm": 0.887543797492981, "learning_rate": 9.598262995928611e-05, "loss": 0.01106376200914383, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.01113, "step": 92, "tokens/total": 2784672, "tokens/train_per_sec_per_gpu": 36.79, "tokens/trainable": 41927}
+{"epoch": 0.36328125, "grad_norm": 1.7098511457443237, "learning_rate": 9.586188413468492e-05, "loss": 0.02100781723856926, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.02123, "step": 93, "tokens/total": 2815120, "tokens/train_per_sec_per_gpu": 34.84, "tokens/trainable": 42397}
+{"epoch": 0.3671875, "grad_norm": 0.05345788225531578, "learning_rate": 9.57394378802934e-05, "loss": 0.0007664336590096354, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00077, "step": 94, "tokens/total": 2845600, "tokens/train_per_sec_per_gpu": 36.55, "tokens/trainable": 42883}
+{"epoch": 0.37109375, "grad_norm": 0.3566146492958069, "learning_rate": 9.56152962916e-05, "loss": 0.001947208889760077, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00195, "step": 95, "tokens/total": 2875952, "tokens/train_per_sec_per_gpu": 38.67, "tokens/trainable": 43381}
+{"epoch": 0.375, "grad_norm": 0.6130653023719788, "learning_rate": 9.548946453464296e-05, "loss": 0.011625099927186966, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.01169, "step": 96, "tokens/total": 2906288, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 43824}
+{"epoch": 0.37890625, "grad_norm": 0.5786341428756714, "learning_rate": 9.53619478457953e-05, "loss": 0.0046112253330647945, "memory/device_reserved (GiB)": 35.98, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00462, "step": 97, "tokens/total": 2936752, "tokens/train_per_sec_per_gpu": 31.51, "tokens/trainable": 44255}
+{"epoch": 0.3828125, "grad_norm": 0.4321880340576172, "learning_rate": 9.523275153154695e-05, "loss": 0.003710552118718624, "memory/device_reserved (GiB)": 35.19, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00372, "step": 98, "tokens/total": 2967072, "tokens/train_per_sec_per_gpu": 37.74, "tokens/trainable": 44739}
+{"epoch": 0.38671875, "grad_norm": 0.16571089625358582, "learning_rate": 9.51018809682839e-05, "loss": 0.00260853860527277, "memory/device_reserved (GiB)": 35.76, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00261, "step": 99, "tokens/total": 2997664, "tokens/train_per_sec_per_gpu": 36.2, "tokens/trainable": 45202}
+{"epoch": 0.390625, "grad_norm": 0.4386274516582489, "learning_rate": 9.49693416020645e-05, "loss": 0.005652088671922684, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00567, "step": 100, "tokens/total": 3028032, "tokens/train_per_sec_per_gpu": 33.05, "tokens/trainable": 45646}
+{"epoch": 0.39453125, "grad_norm": 0.12225741147994995, "learning_rate": 9.483513894839276e-05, "loss": 0.0015262231463566422, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00153, "step": 101, "tokens/total": 3058272, "tokens/train_per_sec_per_gpu": 33.08, "tokens/trainable": 46089}
+{"epoch": 0.3984375, "grad_norm": 0.779109537601471, "learning_rate": 9.469927859198888e-05, "loss": 0.007584188133478165, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00761, "step": 102, "tokens/total": 3088848, "tokens/train_per_sec_per_gpu": 32.55, "tokens/trainable": 46547}
+{"epoch": 0.40234375, "grad_norm": 0.19508829712867737, "learning_rate": 9.456176618655689e-05, "loss": 0.0021933168172836304, "memory/device_reserved (GiB)": 35.99, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.0022, "step": 103, "tokens/total": 3119424, "tokens/train_per_sec_per_gpu": 31.96, "tokens/trainable": 47014}
+{"epoch": 0.40625, "grad_norm": 0.349365770816803, "learning_rate": 9.442260745454927e-05, "loss": 0.007655802182853222, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00769, "step": 104, "tokens/total": 3149696, "tokens/train_per_sec_per_gpu": 37.71, "tokens/trainable": 47503}
+{"epoch": 0.41015625, "grad_norm": 0.3205769956111908, "learning_rate": 9.428180818692884e-05, "loss": 0.0014675151323899627, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00147, "step": 105, "tokens/total": 3180064, "tokens/train_per_sec_per_gpu": 36.27, "tokens/trainable": 47996}
+{"epoch": 0.4140625, "grad_norm": 0.23203934729099274, "learning_rate": 9.413937424292791e-05, "loss": 0.001969895325601101, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.00197, "step": 106, "tokens/total": 3208320, "tokens/train_per_sec_per_gpu": 34.06, "tokens/trainable": 48447}
+{"epoch": 0.41796875, "grad_norm": 0.050742585211992264, "learning_rate": 9.399531154980424e-05, "loss": 0.00028503642533905804, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00029, "step": 107, "tokens/total": 3238416, "tokens/train_per_sec_per_gpu": 30.69, "tokens/trainable": 48889}
+{"epoch": 0.421875, "grad_norm": 0.68248450756073, "learning_rate": 9.384962610259455e-05, "loss": 0.0026741819456219673, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00268, "step": 108, "tokens/total": 3268784, "tokens/train_per_sec_per_gpu": 37.89, "tokens/trainable": 49360}
+{"epoch": 0.42578125, "grad_norm": 0.2664353549480438, "learning_rate": 9.370232396386494e-05, "loss": 0.0014279462629929185, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00143, "step": 109, "tokens/total": 3298912, "tokens/train_per_sec_per_gpu": 36.78, "tokens/trainable": 49827}
+{"epoch": 0.4296875, "grad_norm": 0.12221308797597885, "learning_rate": 9.355341126345868e-05, "loss": 0.0007402317132800817, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00074, "step": 110, "tokens/total": 3329232, "tokens/train_per_sec_per_gpu": 34.31, "tokens/trainable": 50267}
+{"epoch": 0.43359375, "grad_norm": 1.2223896980285645, "learning_rate": 9.340289419824107e-05, "loss": 0.020261401310563087, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.02047, "step": 111, "tokens/total": 3359440, "tokens/train_per_sec_per_gpu": 33.41, "tokens/trainable": 50711}
+{"epoch": 0.4375, "grad_norm": 1.2567111253738403, "learning_rate": 9.325077903184159e-05, "loss": 0.012774579226970673, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.01286, "step": 112, "tokens/total": 3389472, "tokens/train_per_sec_per_gpu": 31.59, "tokens/trainable": 51157}
+{"epoch": 0.44140625, "grad_norm": 0.8144268989562988, "learning_rate": 9.30970720943932e-05, "loss": 0.005420445930212736, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00544, "step": 113, "tokens/total": 3419920, "tokens/train_per_sec_per_gpu": 32.71, "tokens/trainable": 51613}
+{"epoch": 0.4453125, "grad_norm": 1.3511962890625, "learning_rate": 9.2941779782269e-05, "loss": 0.004596072714775801, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00461, "step": 114, "tokens/total": 3450176, "tokens/train_per_sec_per_gpu": 29.1, "tokens/trainable": 52017}
+{"epoch": 0.44921875, "grad_norm": 2.76552677154541, "learning_rate": 9.278490855781596e-05, "loss": 0.005162997171282768, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00518, "step": 115, "tokens/total": 3480544, "tokens/train_per_sec_per_gpu": 36.22, "tokens/trainable": 52506}
+{"epoch": 0.453125, "grad_norm": 1.7830004692077637, "learning_rate": 9.262646494908604e-05, "loss": 0.024823248386383057, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.02513, "step": 116, "tokens/total": 3510848, "tokens/train_per_sec_per_gpu": 38.63, "tokens/trainable": 53008}
+{"epoch": 0.45703125, "grad_norm": 0.8680487275123596, "learning_rate": 9.246645554956457e-05, "loss": 0.012252703309059143, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.01233, "step": 117, "tokens/total": 3541296, "tokens/train_per_sec_per_gpu": 35.36, "tokens/trainable": 53467}
+{"epoch": 0.4609375, "grad_norm": 0.8030971884727478, "learning_rate": 9.230488701789578e-05, "loss": 0.01154071744531393, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.01161, "step": 118, "tokens/total": 3571936, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 53923}
+{"epoch": 0.46484375, "grad_norm": 1.1320112943649292, "learning_rate": 9.214176607760577e-05, "loss": 0.00475259218364954, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00476, "step": 119, "tokens/total": 3602000, "tokens/train_per_sec_per_gpu": 35.58, "tokens/trainable": 54363}
+{"epoch": 0.46875, "grad_norm": 0.25778594613075256, "learning_rate": 9.197709951682268e-05, "loss": 0.00691817793995142, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00694, "step": 120, "tokens/total": 3632288, "tokens/train_per_sec_per_gpu": 35.18, "tokens/trainable": 54839}
+{"epoch": 0.47265625, "grad_norm": 0.3193860948085785, "learning_rate": 9.181089418799428e-05, "loss": 0.0037792143411934376, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00379, "step": 121, "tokens/total": 3662608, "tokens/train_per_sec_per_gpu": 32.59, "tokens/trainable": 55267}
+{"epoch": 0.4765625, "grad_norm": 0.34393274784088135, "learning_rate": 9.164315700760271e-05, "loss": 0.0042166681960225105, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00423, "step": 122, "tokens/total": 3692864, "tokens/train_per_sec_per_gpu": 30.52, "tokens/trainable": 55700}
+{"epoch": 0.48046875, "grad_norm": 0.4916401505470276, "learning_rate": 9.147389495587671e-05, "loss": 0.0038735629059374332, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00388, "step": 123, "tokens/total": 3722864, "tokens/train_per_sec_per_gpu": 33.39, "tokens/trainable": 56144}
+{"epoch": 0.484375, "grad_norm": 0.39921140670776367, "learning_rate": 9.130311507650116e-05, "loss": 0.004010441247373819, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00402, "step": 124, "tokens/total": 3753056, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 56589}
+{"epoch": 0.48828125, "grad_norm": 0.4556286633014679, "learning_rate": 9.113082447632394e-05, "loss": 0.005605827085673809, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00562, "step": 125, "tokens/total": 3783312, "tokens/train_per_sec_per_gpu": 36.68, "tokens/trainable": 57081}
+{"epoch": 0.4921875, "grad_norm": 0.041882507503032684, "learning_rate": 9.09570303250602e-05, "loss": 0.0005784470704384148, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00058, "step": 126, "tokens/total": 3813712, "tokens/train_per_sec_per_gpu": 32.87, "tokens/trainable": 57527}
+{"epoch": 0.49609375, "grad_norm": 0.21823598444461823, "learning_rate": 9.078173985499394e-05, "loss": 0.0024290401488542557, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00243, "step": 127, "tokens/total": 3844336, "tokens/train_per_sec_per_gpu": 34.84, "tokens/trainable": 58005}
+{"epoch": 0.5, "grad_norm": 0.4637400209903717, "learning_rate": 9.060496036067713e-05, "loss": 0.001603663433343172, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.0016, "step": 128, "tokens/total": 3872688, "tokens/train_per_sec_per_gpu": 35.21, "tokens/trainable": 58454}
+{"epoch": 0.50390625, "grad_norm": 0.03311942145228386, "learning_rate": 9.042669919862615e-05, "loss": 0.00027762592071667314, "memory/device_reserved (GiB)": 36.0, "memory/max_active (GiB)": 33.64, "memory/max_allocated (GiB)": 33.64, "ppl": 1.00028, "step": 129, "tokens/total": 3902640, "tokens/train_per_sec_per_gpu": 31.97, "tokens/trainable": 58896}
+{"epoch": 0.5078125, "grad_norm": 0.019396035000681877, "learning_rate": 9.024696378701557e-05, "loss": 8.183176396414638e-05, "memory/device_reserved (GiB)": 34.68, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00008, "step": 130, "tokens/total": 3933104, "tokens/train_per_sec_per_gpu": 35.8, "tokens/trainable": 59356}
+{"epoch": 0.51171875, "grad_norm": 0.3952345848083496, "learning_rate": 9.006576160536948e-05, "loss": 0.005079061258584261, "memory/device_reserved (GiB)": 34.92, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00509, "step": 131, "tokens/total": 3963184, "tokens/train_per_sec_per_gpu": 32.21, "tokens/trainable": 59792}
+{"epoch": 0.515625, "grad_norm": 0.03258282318711281, "learning_rate": 8.988310019425035e-05, "loss": 0.00015263023669831455, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00015, "step": 132, "tokens/total": 3993648, "tokens/train_per_sec_per_gpu": 35.68, "tokens/trainable": 60266}
+{"epoch": 0.51953125, "grad_norm": 0.10869178175926208, "learning_rate": 8.969898715494506e-05, "loss": 0.00033613567939028144, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00034, "step": 133, "tokens/total": 4024000, "tokens/train_per_sec_per_gpu": 35.44, "tokens/trainable": 60759}
+{"epoch": 0.5234375, "grad_norm": 0.016073748469352722, "learning_rate": 8.951343014914869e-05, "loss": 0.00014354957966133952, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.39, "memory/max_allocated (GiB)": 33.39, "ppl": 1.00014, "step": 134, "tokens/total": 4052480, "tokens/train_per_sec_per_gpu": 31.76, "tokens/trainable": 61204}
+{"epoch": 0.52734375, "grad_norm": 0.5635335445404053, "learning_rate": 8.932643689864568e-05, "loss": 0.0033080782741308212, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00331, "step": 135, "tokens/total": 4082864, "tokens/train_per_sec_per_gpu": 35.93, "tokens/trainable": 61684}
+{"epoch": 0.53125, "grad_norm": 0.12933659553527832, "learning_rate": 8.913801518498845e-05, "loss": 0.0003551081463228911, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.7, "memory/max_allocated (GiB)": 33.7, "ppl": 1.00036, "step": 136, "tokens/total": 4113040, "tokens/train_per_sec_per_gpu": 38.22, "tokens/trainable": 62159}
+{"epoch": 0.53515625, "grad_norm": 0.038116879761219025, "learning_rate": 8.894817284917364e-05, "loss": 0.0002933211508207023, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00029, "step": 137, "tokens/total": 4143344, "tokens/train_per_sec_per_gpu": 35.28, "tokens/trainable": 62649}
+{"epoch": 0.5390625, "grad_norm": 0.4708496928215027, "learning_rate": 8.875691779131569e-05, "loss": 0.009557873010635376, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.0096, "step": 138, "tokens/total": 4173648, "tokens/train_per_sec_per_gpu": 36.65, "tokens/trainable": 63100}
+{"epoch": 0.54296875, "grad_norm": 0.6401950120925903, "learning_rate": 8.856425797031829e-05, "loss": 0.006467439234256744, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00649, "step": 139, "tokens/total": 4204080, "tokens/train_per_sec_per_gpu": 37.56, "tokens/trainable": 63588}
+{"epoch": 0.546875, "grad_norm": 0.9977340698242188, "learning_rate": 8.837020140354295e-05, "loss": 0.01085682213306427, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.42, "memory/max_allocated (GiB)": 33.42, "ppl": 1.01092, "step": 140, "tokens/total": 4232400, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 64005}
+{"epoch": 0.55078125, "grad_norm": 0.9310302734375, "learning_rate": 8.817475616647554e-05, "loss": 0.002832747297361493, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00284, "step": 141, "tokens/total": 4262624, "tokens/train_per_sec_per_gpu": 35.12, "tokens/trainable": 64462}
+{"epoch": 0.5546875, "grad_norm": 0.8448959589004517, "learning_rate": 8.797793039239017e-05, "loss": 0.00225994479842484, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00226, "step": 142, "tokens/total": 4293168, "tokens/train_per_sec_per_gpu": 33.93, "tokens/trainable": 64900}
+{"epoch": 0.55859375, "grad_norm": 0.1500774621963501, "learning_rate": 8.777973227201069e-05, "loss": 0.001096544205211103, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0011, "step": 143, "tokens/total": 4323248, "tokens/train_per_sec_per_gpu": 37.52, "tokens/trainable": 65379}
+{"epoch": 0.5625, "grad_norm": 0.7080555558204651, "learning_rate": 8.758017005316988e-05, "loss": 0.02480880171060562, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.02512, "step": 144, "tokens/total": 4353344, "tokens/train_per_sec_per_gpu": 36.69, "tokens/trainable": 65845}
+{"epoch": 0.56640625, "grad_norm": 0.8302886486053467, "learning_rate": 8.737925204046629e-05, "loss": 0.011632976122200489, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.0117, "step": 145, "tokens/total": 4383584, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 66285}
+{"epoch": 0.5703125, "grad_norm": 0.48007094860076904, "learning_rate": 8.717698659491851e-05, "loss": 0.006236276589334011, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00626, "step": 146, "tokens/total": 4413856, "tokens/train_per_sec_per_gpu": 31.52, "tokens/trainable": 66736}
+{"epoch": 0.57421875, "grad_norm": 0.2671569585800171, "learning_rate": 8.697338213361735e-05, "loss": 0.004551246762275696, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00456, "step": 147, "tokens/total": 4444112, "tokens/train_per_sec_per_gpu": 35.26, "tokens/trainable": 67150}
+{"epoch": 0.578125, "grad_norm": 0.2333666980266571, "learning_rate": 8.676844712937552e-05, "loss": 0.005151683464646339, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00516, "step": 148, "tokens/total": 4474464, "tokens/train_per_sec_per_gpu": 35.08, "tokens/trainable": 67633}
+{"epoch": 0.58203125, "grad_norm": 1.3599361181259155, "learning_rate": 8.656219011037509e-05, "loss": 0.010383867658674717, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01044, "step": 149, "tokens/total": 4504880, "tokens/train_per_sec_per_gpu": 33.82, "tokens/trainable": 68077}
+{"epoch": 0.5859375, "grad_norm": 0.2275276631116867, "learning_rate": 8.63546196598125e-05, "loss": 0.0016249704640358686, "memory/device_reserved (GiB)": 35.79, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00163, "step": 150, "tokens/total": 4534976, "tokens/train_per_sec_per_gpu": 33.18, "tokens/trainable": 68545}
+{"epoch": 0.58984375, "grad_norm": 0.6304328441619873, "learning_rate": 8.614574441554145e-05, "loss": 0.02714025229215622, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.02751, "step": 151, "tokens/total": 4565088, "tokens/train_per_sec_per_gpu": 30.26, "tokens/trainable": 69011}
+{"epoch": 0.59375, "grad_norm": 0.3306228518486023, "learning_rate": 8.593557306971349e-05, "loss": 0.006482407916337252, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.0065, "step": 152, "tokens/total": 4595600, "tokens/train_per_sec_per_gpu": 35.37, "tokens/trainable": 69465}
+{"epoch": 0.59765625, "grad_norm": 2.509195566177368, "learning_rate": 8.572411436841618e-05, "loss": 0.014876915141940117, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01499, "step": 153, "tokens/total": 4626032, "tokens/train_per_sec_per_gpu": 33.89, "tokens/trainable": 69928}
+{"epoch": 0.6015625, "grad_norm": 0.08186601847410202, "learning_rate": 8.551137711130922e-05, "loss": 0.0014128303155303001, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00141, "step": 154, "tokens/total": 4656352, "tokens/train_per_sec_per_gpu": 36.67, "tokens/trainable": 70395}
+{"epoch": 0.60546875, "grad_norm": 0.20300441980361938, "learning_rate": 8.529737015125824e-05, "loss": 0.0016238696407526731, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00163, "step": 155, "tokens/total": 4684672, "tokens/train_per_sec_per_gpu": 35.58, "tokens/trainable": 70829}
+{"epoch": 0.609375, "grad_norm": 0.10475975275039673, "learning_rate": 8.508210239396639e-05, "loss": 0.0014675778802484274, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00147, "step": 156, "tokens/total": 4715216, "tokens/train_per_sec_per_gpu": 33.88, "tokens/trainable": 71287}
+{"epoch": 0.61328125, "grad_norm": 0.2740918695926666, "learning_rate": 8.486558279760375e-05, "loss": 0.004101475700736046, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00411, "step": 157, "tokens/total": 4745376, "tokens/train_per_sec_per_gpu": 34.95, "tokens/trainable": 71749}
+{"epoch": 0.6171875, "grad_norm": 0.30794957280158997, "learning_rate": 8.464782037243449e-05, "loss": 0.0044078221544623375, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00442, "step": 158, "tokens/total": 4775776, "tokens/train_per_sec_per_gpu": 32.18, "tokens/trainable": 72185}
+{"epoch": 0.62109375, "grad_norm": 0.3344379961490631, "learning_rate": 8.442882418044202e-05, "loss": 0.0035248196218162775, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00353, "step": 159, "tokens/total": 4805856, "tokens/train_per_sec_per_gpu": 30.54, "tokens/trainable": 72604}
+{"epoch": 0.625, "grad_norm": 0.6211779117584229, "learning_rate": 8.420860333495179e-05, "loss": 0.015572377480566502, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01569, "step": 160, "tokens/total": 4836304, "tokens/train_per_sec_per_gpu": 31.56, "tokens/trainable": 73061}
+{"epoch": 0.62890625, "grad_norm": 0.5694647431373596, "learning_rate": 8.398716700025208e-05, "loss": 0.005395190790295601, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00541, "step": 161, "tokens/total": 4866544, "tokens/train_per_sec_per_gpu": 29.98, "tokens/trainable": 73474}
+{"epoch": 0.6328125, "grad_norm": 0.3221203088760376, "learning_rate": 8.376452439121266e-05, "loss": 0.004907120484858751, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00492, "step": 162, "tokens/total": 4897120, "tokens/train_per_sec_per_gpu": 32.65, "tokens/trainable": 73923}
+{"epoch": 0.63671875, "grad_norm": 0.2562556862831116, "learning_rate": 8.354068477290124e-05, "loss": 0.011767375282943249, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01184, "step": 163, "tokens/total": 4927440, "tokens/train_per_sec_per_gpu": 35.35, "tokens/trainable": 74379}
+{"epoch": 0.640625, "grad_norm": 0.37376344203948975, "learning_rate": 8.331565746019807e-05, "loss": 0.014833642169833183, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.01494, "step": 164, "tokens/total": 4958032, "tokens/train_per_sec_per_gpu": 33.45, "tokens/trainable": 74830}
+{"epoch": 0.64453125, "grad_norm": 0.07679534703493118, "learning_rate": 8.308945181740812e-05, "loss": 0.0006800079718232155, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00068, "step": 165, "tokens/total": 4988368, "tokens/train_per_sec_per_gpu": 31.73, "tokens/trainable": 75262}
+{"epoch": 0.6484375, "grad_norm": 0.15940658748149872, "learning_rate": 8.286207725787153e-05, "loss": 0.0005167147610336542, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00052, "step": 166, "tokens/total": 5018672, "tokens/train_per_sec_per_gpu": 31.84, "tokens/trainable": 75703}
+{"epoch": 0.65234375, "grad_norm": 0.18865765631198883, "learning_rate": 8.263354324357182e-05, "loss": 0.003161062952131033, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00317, "step": 167, "tokens/total": 5049344, "tokens/train_per_sec_per_gpu": 31.3, "tokens/trainable": 76164}
+{"epoch": 0.65625, "grad_norm": 0.3047192096710205, "learning_rate": 8.240385928474219e-05, "loss": 0.010941199027001858, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.011, "step": 168, "tokens/total": 5079856, "tokens/train_per_sec_per_gpu": 34.45, "tokens/trainable": 76623}
+{"epoch": 0.66015625, "grad_norm": 0.39291632175445557, "learning_rate": 8.217303493946967e-05, "loss": 0.009001722559332848, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00904, "step": 169, "tokens/total": 5110192, "tokens/train_per_sec_per_gpu": 34.39, "tokens/trainable": 77047}
+{"epoch": 0.6640625, "grad_norm": 0.09876111894845963, "learning_rate": 8.194107981329746e-05, "loss": 0.0014627808704972267, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00146, "step": 170, "tokens/total": 5140592, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 77496}
+{"epoch": 0.66796875, "grad_norm": 0.02293679490685463, "learning_rate": 8.170800355882518e-05, "loss": 0.00041422739741392434, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00041, "step": 171, "tokens/total": 5170848, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 77956}
+{"epoch": 0.671875, "grad_norm": 0.2762586176395416, "learning_rate": 8.147381587530713e-05, "loss": 0.006397426128387451, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00642, "step": 172, "tokens/total": 5201280, "tokens/train_per_sec_per_gpu": 29.78, "tokens/trainable": 78410}
+{"epoch": 0.67578125, "grad_norm": 0.07085710763931274, "learning_rate": 8.123852650824877e-05, "loss": 0.0017267607618123293, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00173, "step": 173, "tokens/total": 5231712, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 78849}
+{"epoch": 0.6796875, "grad_norm": 0.4523530602455139, "learning_rate": 8.100214524900103e-05, "loss": 0.011936807073652744, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.62, "memory/max_allocated (GiB)": 33.62, "ppl": 1.01201, "step": 174, "tokens/total": 5261440, "tokens/train_per_sec_per_gpu": 30.07, "tokens/trainable": 79266}
+{"epoch": 0.68359375, "grad_norm": 0.1293293535709381, "learning_rate": 8.076468193435301e-05, "loss": 0.0021572443656623363, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00216, "step": 175, "tokens/total": 5291728, "tokens/train_per_sec_per_gpu": 32.53, "tokens/trainable": 79736}
+{"epoch": 0.6875, "grad_norm": 0.13209661841392517, "learning_rate": 8.052614644612253e-05, "loss": 0.0017078389646485448, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00171, "step": 176, "tokens/total": 5322240, "tokens/train_per_sec_per_gpu": 36.17, "tokens/trainable": 80214}
+{"epoch": 0.69140625, "grad_norm": 0.09734465181827545, "learning_rate": 8.028654871074489e-05, "loss": 0.0020819404162466526, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00208, "step": 177, "tokens/total": 5352576, "tokens/train_per_sec_per_gpu": 32.56, "tokens/trainable": 80661}
+{"epoch": 0.6953125, "grad_norm": 0.8079077005386353, "learning_rate": 8.004589869885986e-05, "loss": 0.006122267805039883, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00614, "step": 178, "tokens/total": 5382784, "tokens/train_per_sec_per_gpu": 34.82, "tokens/trainable": 81129}
+{"epoch": 0.69921875, "grad_norm": 0.806120753288269, "learning_rate": 7.980420642489674e-05, "loss": 0.003698066109791398, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.0037, "step": 179, "tokens/total": 5412784, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 81576}
+{"epoch": 0.703125, "grad_norm": 0.0535414032638073, "learning_rate": 7.95614819466576e-05, "loss": 0.0007224087603390217, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00072, "step": 180, "tokens/total": 5442960, "tokens/train_per_sec_per_gpu": 34.56, "tokens/trainable": 82049}
+{"epoch": 0.70703125, "grad_norm": 0.8787689208984375, "learning_rate": 7.931773536489872e-05, "loss": 0.014041105285286903, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.01414, "step": 181, "tokens/total": 5473216, "tokens/train_per_sec_per_gpu": 38.79, "tokens/trainable": 82566}
+{"epoch": 0.7109375, "grad_norm": 0.23288801312446594, "learning_rate": 7.907297682291035e-05, "loss": 0.0018348076846450567, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00184, "step": 182, "tokens/total": 5503568, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 83054}
+{"epoch": 0.71484375, "grad_norm": 0.08039269596338272, "learning_rate": 7.882721650609442e-05, "loss": 0.0011233665281906724, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00112, "step": 183, "tokens/total": 5533760, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 83501}
+{"epoch": 0.71875, "grad_norm": 0.028009524568915367, "learning_rate": 7.85804646415409e-05, "loss": 0.00045670082909055054, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00046, "step": 184, "tokens/total": 5564240, "tokens/train_per_sec_per_gpu": 35.48, "tokens/trainable": 84007}
+{"epoch": 0.72265625, "grad_norm": 0.04383641108870506, "learning_rate": 7.833273149760207e-05, "loss": 0.00067467603366822, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00067, "step": 185, "tokens/total": 5594448, "tokens/train_per_sec_per_gpu": 35.54, "tokens/trainable": 84491}
+{"epoch": 0.7265625, "grad_norm": 0.1170438677072525, "learning_rate": 7.808402738346527e-05, "loss": 0.0014282094780355692, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00143, "step": 186, "tokens/total": 5624816, "tokens/train_per_sec_per_gpu": 33.33, "tokens/trainable": 84952}
+{"epoch": 0.73046875, "grad_norm": 0.05614401772618294, "learning_rate": 7.783436264872382e-05, "loss": 0.0002470523177180439, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00025, "step": 187, "tokens/total": 5655104, "tokens/train_per_sec_per_gpu": 32.92, "tokens/trainable": 85400}
+{"epoch": 0.734375, "grad_norm": 0.30610957741737366, "learning_rate": 7.758374768294647e-05, "loss": 0.004403322469443083, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00441, "step": 188, "tokens/total": 5685360, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 85889}
+{"epoch": 0.73828125, "grad_norm": 0.433325856924057, "learning_rate": 7.733219291524489e-05, "loss": 0.005490327253937721, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00551, "step": 189, "tokens/total": 5715648, "tokens/train_per_sec_per_gpu": 36.22, "tokens/trainable": 86396}
+{"epoch": 0.7421875, "grad_norm": 0.4976643919944763, "learning_rate": 7.707970881383977e-05, "loss": 0.006626967806369066, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00665, "step": 190, "tokens/total": 5746240, "tokens/train_per_sec_per_gpu": 33.85, "tokens/trainable": 86867}
+{"epoch": 0.74609375, "grad_norm": 1.0000759363174438, "learning_rate": 7.682630588562518e-05, "loss": 0.009028773754835129, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00907, "step": 191, "tokens/total": 5776816, "tokens/train_per_sec_per_gpu": 36.07, "tokens/trainable": 87314}
+{"epoch": 0.75, "grad_norm": 0.006456763483583927, "learning_rate": 7.657199467573129e-05, "loss": 5.810559741803445e-05, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00006, "step": 192, "tokens/total": 5807248, "tokens/train_per_sec_per_gpu": 30.38, "tokens/trainable": 87736}
+{"epoch": 0.75390625, "grad_norm": 0.6312516331672668, "learning_rate": 7.631678576708561e-05, "loss": 0.0015132949920371175, "memory/device_reserved (GiB)": 36.02, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00151, "step": 193, "tokens/total": 5837680, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 88219}
+{"epoch": 0.7578125, "grad_norm": 0.003887161612510681, "learning_rate": 7.606068977997255e-05, "loss": 3.5049586585955694e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00004, "step": 194, "tokens/total": 5868000, "tokens/train_per_sec_per_gpu": 36.15, "tokens/trainable": 88699}
+{"epoch": 0.76171875, "grad_norm": 0.05365428328514099, "learning_rate": 7.580371737159148e-05, "loss": 0.00017142925935331732, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.38, "memory/max_allocated (GiB)": 33.38, "ppl": 1.00017, "step": 195, "tokens/total": 5896224, "tokens/train_per_sec_per_gpu": 30.39, "tokens/trainable": 89132}
+{"epoch": 0.765625, "grad_norm": 1.5210427045822144, "learning_rate": 7.554587923561324e-05, "loss": 0.007072226610034704, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.0071, "step": 196, "tokens/total": 5926624, "tokens/train_per_sec_per_gpu": 37.24, "tokens/trainable": 89594}
+{"epoch": 0.76953125, "grad_norm": 0.01594001241028309, "learning_rate": 7.528718610173511e-05, "loss": 0.00012487987987697124, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.34, "memory/max_allocated (GiB)": 33.34, "ppl": 1.00012, "step": 197, "tokens/total": 5954944, "tokens/train_per_sec_per_gpu": 38.08, "tokens/trainable": 90042}
+{"epoch": 0.7734375, "grad_norm": 0.48730939626693726, "learning_rate": 7.502764873523431e-05, "loss": 0.004653760232031345, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00466, "step": 198, "tokens/total": 5985200, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 90503}
+{"epoch": 0.77734375, "grad_norm": 0.06984082609415054, "learning_rate": 7.476727793652011e-05, "loss": 0.0007038296316750348, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 34.0, "memory/max_allocated (GiB)": 34.0, "ppl": 1.0007, "step": 199, "tokens/total": 6015952, "tokens/train_per_sec_per_gpu": 36.46, "tokens/trainable": 90982}
+{"epoch": 0.78125, "grad_norm": 0.09596561640501022, "learning_rate": 7.450608454068415e-05, "loss": 0.0010627463925629854, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00106, "step": 200, "tokens/total": 6046208, "tokens/train_per_sec_per_gpu": 39.3, "tokens/trainable": 91466}
+{"epoch": 0.78515625, "grad_norm": 0.03423153981566429, "learning_rate": 7.424407941704987e-05, "loss": 0.00040728453313931823, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00041, "step": 201, "tokens/total": 6076208, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 91912}
+{"epoch": 0.7890625, "grad_norm": 0.2397124022245407, "learning_rate": 7.398127346871986e-05, "loss": 0.0034314473159611225, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00344, "step": 202, "tokens/total": 6106560, "tokens/train_per_sec_per_gpu": 37.86, "tokens/trainable": 92385}
+{"epoch": 0.79296875, "grad_norm": 0.8799890875816345, "learning_rate": 7.371767763212238e-05, "loss": 0.021598415449261665, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.02183, "step": 203, "tokens/total": 6136640, "tokens/train_per_sec_per_gpu": 36.55, "tokens/trainable": 92842}
+{"epoch": 0.796875, "grad_norm": 0.12073227018117905, "learning_rate": 7.345330287655617e-05, "loss": 0.0006239307695068419, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00062, "step": 204, "tokens/total": 6167184, "tokens/train_per_sec_per_gpu": 33.97, "tokens/trainable": 93323}
+{"epoch": 0.80078125, "grad_norm": 0.3556082844734192, "learning_rate": 7.31881602037339e-05, "loss": 0.0026657599955797195, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00267, "step": 205, "tokens/total": 6197376, "tokens/train_per_sec_per_gpu": 34.79, "tokens/trainable": 93756}
+{"epoch": 0.8046875, "grad_norm": 0.24451178312301636, "learning_rate": 7.29222606473245e-05, "loss": 0.0015720583032816648, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00157, "step": 206, "tokens/total": 6227600, "tokens/train_per_sec_per_gpu": 37.54, "tokens/trainable": 94235}
+{"epoch": 0.80859375, "grad_norm": 0.23219481110572815, "learning_rate": 7.265561527249383e-05, "loss": 0.0031918964814394712, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0032, "step": 207, "tokens/total": 6257760, "tokens/train_per_sec_per_gpu": 36.52, "tokens/trainable": 94733}
+{"epoch": 0.8125, "grad_norm": 0.13691213726997375, "learning_rate": 7.238823517544436e-05, "loss": 0.0018865278689190745, "memory/device_reserved (GiB)": 35.93, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00189, "step": 208, "tokens/total": 6288048, "tokens/train_per_sec_per_gpu": 33.74, "tokens/trainable": 95218}
+{"epoch": 0.81640625, "grad_norm": 0.2183372676372528, "learning_rate": 7.212013148295333e-05, "loss": 0.0018021916039288044, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.0018, "step": 209, "tokens/total": 6318688, "tokens/train_per_sec_per_gpu": 29.23, "tokens/trainable": 95619}
+{"epoch": 0.8203125, "grad_norm": 0.8775327801704407, "learning_rate": 7.185131535190975e-05, "loss": 0.01249875407665968, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01258, "step": 210, "tokens/total": 6348944, "tokens/train_per_sec_per_gpu": 35.39, "tokens/trainable": 96076}
+{"epoch": 0.82421875, "grad_norm": 0.4041156768798828, "learning_rate": 7.158179796885005e-05, "loss": 0.0069442871026694775, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00697, "step": 211, "tokens/total": 6378960, "tokens/train_per_sec_per_gpu": 35.34, "tokens/trainable": 96505}
+{"epoch": 0.828125, "grad_norm": 0.5414943099021912, "learning_rate": 7.131159054949273e-05, "loss": 0.0017384262755513191, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00174, "step": 212, "tokens/total": 6409472, "tokens/train_per_sec_per_gpu": 32.68, "tokens/trainable": 96930}
+{"epoch": 0.83203125, "grad_norm": 0.010344896465539932, "learning_rate": 7.104070433827139e-05, "loss": 0.0002013841731240973, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0002, "step": 213, "tokens/total": 6439904, "tokens/train_per_sec_per_gpu": 32.72, "tokens/trainable": 97361}
+{"epoch": 0.8359375, "grad_norm": 0.5833696722984314, "learning_rate": 7.076915060786705e-05, "loss": 0.009723629802465439, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00977, "step": 214, "tokens/total": 6470176, "tokens/train_per_sec_per_gpu": 36.34, "tokens/trainable": 97858}
+{"epoch": 0.83984375, "grad_norm": 0.31467050313949585, "learning_rate": 7.049694065873882e-05, "loss": 0.004062815103679895, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00407, "step": 215, "tokens/total": 6500336, "tokens/train_per_sec_per_gpu": 37.47, "tokens/trainable": 98308}
+{"epoch": 0.84375, "grad_norm": 0.013619605451822281, "learning_rate": 7.022408581865382e-05, "loss": 0.0002783117815852165, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00028, "step": 216, "tokens/total": 6530688, "tokens/train_per_sec_per_gpu": 36.41, "tokens/trainable": 98781}
+{"epoch": 0.84765625, "grad_norm": 0.02845357358455658, "learning_rate": 6.99505974422157e-05, "loss": 0.00029517346411012113, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.0003, "step": 217, "tokens/total": 6561040, "tokens/train_per_sec_per_gpu": 36.76, "tokens/trainable": 99266}
+{"epoch": 0.8515625, "grad_norm": 0.17569833993911743, "learning_rate": 6.967648691039213e-05, "loss": 0.0029272970277816057, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00293, "step": 218, "tokens/total": 6589392, "tokens/train_per_sec_per_gpu": 39.65, "tokens/trainable": 99731}
+{"epoch": 0.85546875, "grad_norm": 0.25007033348083496, "learning_rate": 6.940176563004123e-05, "loss": 0.005346981342881918, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00536, "step": 219, "tokens/total": 6619824, "tokens/train_per_sec_per_gpu": 31.71, "tokens/trainable": 100164}
+{"epoch": 0.859375, "grad_norm": 0.36598774790763855, "learning_rate": 6.912644503343682e-05, "loss": 0.00675960024818778, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00678, "step": 220, "tokens/total": 6650224, "tokens/train_per_sec_per_gpu": 36.18, "tokens/trainable": 100620}
+{"epoch": 0.86328125, "grad_norm": 0.2870761454105377, "learning_rate": 6.885053657779273e-05, "loss": 0.004020639695227146, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00403, "step": 221, "tokens/total": 6680384, "tokens/train_per_sec_per_gpu": 32.94, "tokens/trainable": 101080}
+{"epoch": 0.8671875, "grad_norm": 0.05318336561322212, "learning_rate": 6.857405174478604e-05, "loss": 0.0006697649951092899, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00067, "step": 222, "tokens/total": 6710896, "tokens/train_per_sec_per_gpu": 37.8, "tokens/trainable": 101546}
+{"epoch": 0.87109375, "grad_norm": 0.016319693997502327, "learning_rate": 6.82970020400792e-05, "loss": 0.00020417713676579297, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.0002, "step": 223, "tokens/total": 6741280, "tokens/train_per_sec_per_gpu": 31.52, "tokens/trainable": 101987}
+{"epoch": 0.875, "grad_norm": 0.09935661405324936, "learning_rate": 6.801939899284132e-05, "loss": 0.0018507638014853, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00185, "step": 224, "tokens/total": 6771488, "tokens/train_per_sec_per_gpu": 35.37, "tokens/trainable": 102435}
+{"epoch": 0.87890625, "grad_norm": 0.045450225472450256, "learning_rate": 6.774125415526827e-05, "loss": 0.0007717214175499976, "memory/device_reserved (GiB)": 35.94, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00077, "step": 225, "tokens/total": 6801744, "tokens/train_per_sec_per_gpu": 28.86, "tokens/trainable": 102843}
+{"epoch": 0.8828125, "grad_norm": 0.1693895012140274, "learning_rate": 6.746257910210214e-05, "loss": 0.0015482499729841948, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00155, "step": 226, "tokens/total": 6832432, "tokens/train_per_sec_per_gpu": 37.39, "tokens/trainable": 103342}
+{"epoch": 0.88671875, "grad_norm": 0.011035765521228313, "learning_rate": 6.718338543014937e-05, "loss": 0.0002161034499295056, "memory/device_reserved (GiB)": 35.71, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00022, "step": 227, "tokens/total": 6862448, "tokens/train_per_sec_per_gpu": 40.81, "tokens/trainable": 103844}
+{"epoch": 0.890625, "grad_norm": 0.339762806892395, "learning_rate": 6.69036847577983e-05, "loss": 0.00712943309918046, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00715, "step": 228, "tokens/total": 6892832, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 104307}
+{"epoch": 0.89453125, "grad_norm": 0.022490613162517548, "learning_rate": 6.662348872453553e-05, "loss": 0.000386083556804806, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.69, "memory/max_allocated (GiB)": 33.69, "ppl": 1.00039, "step": 229, "tokens/total": 6922832, "tokens/train_per_sec_per_gpu": 33.49, "tokens/trainable": 104772}
+{"epoch": 0.8984375, "grad_norm": 0.030813584104180336, "learning_rate": 6.63428089904618e-05, "loss": 0.00022011388500686735, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00022, "step": 230, "tokens/total": 6953104, "tokens/train_per_sec_per_gpu": 36.82, "tokens/trainable": 105240}
+{"epoch": 0.90234375, "grad_norm": 0.13754640519618988, "learning_rate": 6.60616572358065e-05, "loss": 0.0012558111920952797, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00126, "step": 231, "tokens/total": 6983440, "tokens/train_per_sec_per_gpu": 33.76, "tokens/trainable": 105703}
+{"epoch": 0.90625, "grad_norm": 0.003156182123348117, "learning_rate": 6.578004516044172e-05, "loss": 6.873848178656772e-05, "memory/device_reserved (GiB)": 35.77, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00007, "step": 232, "tokens/total": 7013840, "tokens/train_per_sec_per_gpu": 33.02, "tokens/trainable": 106179}
+{"epoch": 0.91015625, "grad_norm": 0.021215464919805527, "learning_rate": 6.549798448339548e-05, "loss": 0.00016204064013436437, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00016, "step": 233, "tokens/total": 7044240, "tokens/train_per_sec_per_gpu": 38.0, "tokens/trainable": 106683}
+{"epoch": 0.9140625, "grad_norm": 0.03298190236091614, "learning_rate": 6.521548694236384e-05, "loss": 0.0003434696700423956, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00034, "step": 234, "tokens/total": 7074224, "tokens/train_per_sec_per_gpu": 33.86, "tokens/trainable": 107121}
+{"epoch": 0.91796875, "grad_norm": 0.08474111557006836, "learning_rate": 6.493256429322259e-05, "loss": 0.0006107806693762541, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00061, "step": 235, "tokens/total": 7104672, "tokens/train_per_sec_per_gpu": 32.7, "tokens/trainable": 107561}
+{"epoch": 0.921875, "grad_norm": 0.012000355869531631, "learning_rate": 6.464922830953799e-05, "loss": 0.00010977509373333305, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00011, "step": 236, "tokens/total": 7134784, "tokens/train_per_sec_per_gpu": 34.28, "tokens/trainable": 108018}
+{"epoch": 0.92578125, "grad_norm": 0.011812024749815464, "learning_rate": 6.436549078207688e-05, "loss": 6.36270415270701e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00006, "step": 237, "tokens/total": 7164960, "tokens/train_per_sec_per_gpu": 33.32, "tokens/trainable": 108446}
+{"epoch": 0.9296875, "grad_norm": 0.055643483996391296, "learning_rate": 6.408136351831592e-05, "loss": 0.0002590732474345714, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00026, "step": 238, "tokens/total": 7195344, "tokens/train_per_sec_per_gpu": 29.15, "tokens/trainable": 108866}
+{"epoch": 0.93359375, "grad_norm": 0.7225600481033325, "learning_rate": 6.379685834195036e-05, "loss": 0.004054174292832613, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00406, "step": 239, "tokens/total": 7225680, "tokens/train_per_sec_per_gpu": 33.86, "tokens/trainable": 109316}
+{"epoch": 0.9375, "grad_norm": 0.06819978356361389, "learning_rate": 6.351198709240186e-05, "loss": 0.00046806156751699746, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00047, "step": 240, "tokens/total": 7256112, "tokens/train_per_sec_per_gpu": 33.27, "tokens/trainable": 109787}
+{"epoch": 0.94140625, "grad_norm": 0.17400965094566345, "learning_rate": 6.32267616243259e-05, "loss": 0.0013036895543336868, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.0013, "step": 241, "tokens/total": 7286384, "tokens/train_per_sec_per_gpu": 38.35, "tokens/trainable": 110246}
+{"epoch": 0.9453125, "grad_norm": 0.0573316365480423, "learning_rate": 6.294119380711849e-05, "loss": 0.00048713054275140166, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00049, "step": 242, "tokens/total": 7316688, "tokens/train_per_sec_per_gpu": 29.22, "tokens/trainable": 110667}
+{"epoch": 0.94921875, "grad_norm": 0.8039842247962952, "learning_rate": 6.265529552442209e-05, "loss": 0.013966547325253487, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01406, "step": 243, "tokens/total": 7346992, "tokens/train_per_sec_per_gpu": 38.13, "tokens/trainable": 111154}
+{"epoch": 0.953125, "grad_norm": 0.03269768878817558, "learning_rate": 6.236907867363127e-05, "loss": 0.00011441730748629197, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00011, "step": 244, "tokens/total": 7377632, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 111611}
+{"epoch": 0.95703125, "grad_norm": 0.006486211437731981, "learning_rate": 6.208255516539749e-05, "loss": 4.603556953952648e-05, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00005, "step": 245, "tokens/total": 7407872, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 112102}
+{"epoch": 0.9609375, "grad_norm": 0.04383692517876625, "learning_rate": 6.179573692313344e-05, "loss": 0.0002824773546308279, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00028, "step": 246, "tokens/total": 7438448, "tokens/train_per_sec_per_gpu": 33.24, "tokens/trainable": 112569}
+{"epoch": 0.96484375, "grad_norm": 0.16995811462402344, "learning_rate": 6.150863588251694e-05, "loss": 0.0024534151889383793, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00246, "step": 247, "tokens/total": 7466768, "tokens/train_per_sec_per_gpu": 37.71, "tokens/trainable": 113023}
+{"epoch": 0.96875, "grad_norm": 0.016517437994480133, "learning_rate": 6.122126399099419e-05, "loss": 0.00018760509556159377, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00019, "step": 248, "tokens/total": 7497360, "tokens/train_per_sec_per_gpu": 35.12, "tokens/trainable": 113464}
+{"epoch": 0.97265625, "grad_norm": 0.27915316820144653, "learning_rate": 6.0933633207282615e-05, "loss": 0.0029680300503969193, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00297, "step": 249, "tokens/total": 7527504, "tokens/train_per_sec_per_gpu": 32.71, "tokens/trainable": 113908}
+{"epoch": 0.9765625, "grad_norm": 0.45557740330696106, "learning_rate": 6.064575550087316e-05, "loss": 0.005397973116487265, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00541, "step": 250, "tokens/total": 7555808, "tokens/train_per_sec_per_gpu": 35.3, "tokens/trainable": 114348}
+{"epoch": 0.98046875, "grad_norm": 0.03177047520875931, "learning_rate": 6.0357642851532245e-05, "loss": 0.00033563212491571903, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00034, "step": 251, "tokens/total": 7586320, "tokens/train_per_sec_per_gpu": 35.23, "tokens/trainable": 114830}
+{"epoch": 0.984375, "grad_norm": 0.2969627380371094, "learning_rate": 6.0069307248803294e-05, "loss": 0.0029213305097073317, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00293, "step": 252, "tokens/total": 7616576, "tokens/train_per_sec_per_gpu": 35.18, "tokens/trainable": 115278}
+{"epoch": 0.98828125, "grad_norm": 0.09316814690828323, "learning_rate": 5.9780760691507635e-05, "loss": 0.00040648062713444233, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00041, "step": 253, "tokens/total": 7647152, "tokens/train_per_sec_per_gpu": 32.29, "tokens/trainable": 115699}
+{"epoch": 0.9921875, "grad_norm": 0.2160760760307312, "learning_rate": 5.9492015187245334e-05, "loss": 0.0019942133221775293, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.002, "step": 254, "tokens/total": 7677584, "tokens/train_per_sec_per_gpu": 30.81, "tokens/trainable": 116133}
+{"epoch": 0.99609375, "grad_norm": 0.08144493401050568, "learning_rate": 5.920308275189541e-05, "loss": 0.0008475447539240122, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00085, "step": 255, "tokens/total": 7708016, "tokens/train_per_sec_per_gpu": 36.82, "tokens/trainable": 116591}
+{"epoch": 1.0, "grad_norm": 0.06008912995457649, "learning_rate": 5.8913975409115874e-05, "loss": 0.0007099541835486889, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00071, "step": 256, "tokens/total": 7738608, "tokens/train_per_sec_per_gpu": 32.11, "tokens/trainable": 117039}
+{"epoch": 1.00390625, "grad_norm": 0.01939917728304863, "learning_rate": 5.8624705189843395e-05, "loss": 0.00010193810157943517, "memory/device_reserved (GiB)": 36.14, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.0001, "step": 257, "tokens/total": 7768944, "tokens/train_per_sec_per_gpu": 32.49, "tokens/trainable": 117495}
+{"epoch": 1.0078125, "grad_norm": 0.020724618807435036, "learning_rate": 5.833528413179249e-05, "loss": 0.00015316747885663062, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00015, "step": 258, "tokens/total": 7799232, "tokens/train_per_sec_per_gpu": 30.54, "tokens/trainable": 117925}
+{"epoch": 1.01171875, "grad_norm": 0.002170708030462265, "learning_rate": 5.80457242789548e-05, "loss": 4.118717333767563e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00004, "step": 259, "tokens/total": 7829568, "tokens/train_per_sec_per_gpu": 31.66, "tokens/trainable": 118373}
+{"epoch": 1.015625, "grad_norm": 0.0011113203363493085, "learning_rate": 5.77560376810977e-05, "loss": 2.2629777959082276e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 260, "tokens/total": 7860000, "tokens/train_per_sec_per_gpu": 31.48, "tokens/trainable": 118816}
+{"epoch": 1.01953125, "grad_norm": 0.01329761277884245, "learning_rate": 5.7466236393263005e-05, "loss": 7.257944525917992e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00007, "step": 261, "tokens/total": 7890464, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 119263}
+{"epoch": 1.0234375, "grad_norm": 0.060381509363651276, "learning_rate": 5.717633247526522e-05, "loss": 5.40789587830659e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00005, "step": 262, "tokens/total": 7920944, "tokens/train_per_sec_per_gpu": 34.79, "tokens/trainable": 119772}
+{"epoch": 1.02734375, "grad_norm": 0.003907898440957069, "learning_rate": 5.688633799118971e-05, "loss": 5.063481512479484e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00005, "step": 263, "tokens/total": 7951296, "tokens/train_per_sec_per_gpu": 33.87, "tokens/trainable": 120269}
+{"epoch": 1.03125, "grad_norm": 0.008468744345009327, "learning_rate": 5.659626500889066e-05, "loss": 3.1418872822541744e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00003, "step": 264, "tokens/total": 7981856, "tokens/train_per_sec_per_gpu": 37.05, "tokens/trainable": 120746}
+{"epoch": 1.03515625, "grad_norm": 0.276324599981308, "learning_rate": 5.6306125599488905e-05, "loss": 0.0011993877124041319, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.0012, "step": 265, "tokens/total": 8012192, "tokens/train_per_sec_per_gpu": 38.3, "tokens/trainable": 121204}
+{"epoch": 1.0390625, "grad_norm": 0.3999474346637726, "learning_rate": 5.601593183686955e-05, "loss": 0.01104999240487814, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.01111, "step": 266, "tokens/total": 8042736, "tokens/train_per_sec_per_gpu": 31.23, "tokens/trainable": 121670}
+{"epoch": 1.04296875, "grad_norm": 0.16733741760253906, "learning_rate": 5.572569579717961e-05, "loss": 0.0020670127123594284, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00207, "step": 267, "tokens/total": 8073024, "tokens/train_per_sec_per_gpu": 40.11, "tokens/trainable": 122183}
+{"epoch": 1.046875, "grad_norm": 0.03506140410900116, "learning_rate": 5.543542955832538e-05, "loss": 0.0002631399256642908, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00026, "step": 268, "tokens/total": 8103280, "tokens/train_per_sec_per_gpu": 35.41, "tokens/trainable": 122645}
+{"epoch": 1.05078125, "grad_norm": 0.006357306614518166, "learning_rate": 5.514514519946986e-05, "loss": 7.129507866920903e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00007, "step": 269, "tokens/total": 8133440, "tokens/train_per_sec_per_gpu": 33.09, "tokens/trainable": 123105}
+{"epoch": 1.0546875, "grad_norm": 0.06588926911354065, "learning_rate": 5.485485480053015e-05, "loss": 0.0005020307726226747, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.0005, "step": 270, "tokens/total": 8163920, "tokens/train_per_sec_per_gpu": 36.24, "tokens/trainable": 123595}
+{"epoch": 1.05859375, "grad_norm": 0.056961171329021454, "learning_rate": 5.4564570441674645e-05, "loss": 0.0007095756009221077, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00071, "step": 271, "tokens/total": 8194032, "tokens/train_per_sec_per_gpu": 37.65, "tokens/trainable": 124077}
+{"epoch": 1.0625, "grad_norm": 0.09018661081790924, "learning_rate": 5.42743042028204e-05, "loss": 0.0002944314619526267, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00029, "step": 272, "tokens/total": 8224320, "tokens/train_per_sec_per_gpu": 34.77, "tokens/trainable": 124547}
+{"epoch": 1.06640625, "grad_norm": 0.13114672899246216, "learning_rate": 5.3984068163130464e-05, "loss": 0.0013651850167661905, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00137, "step": 273, "tokens/total": 8254368, "tokens/train_per_sec_per_gpu": 34.0, "tokens/trainable": 124977}
+{"epoch": 1.0703125, "grad_norm": 0.35149261355400085, "learning_rate": 5.369387440051111e-05, "loss": 0.003347770543769002, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00335, "step": 274, "tokens/total": 8284944, "tokens/train_per_sec_per_gpu": 34.56, "tokens/trainable": 125438}
+{"epoch": 1.07421875, "grad_norm": 0.33942294120788574, "learning_rate": 5.340373499110935e-05, "loss": 0.0012429209891706705, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00124, "step": 275, "tokens/total": 8315584, "tokens/train_per_sec_per_gpu": 31.49, "tokens/trainable": 125889}
+{"epoch": 1.078125, "grad_norm": 0.39785870909690857, "learning_rate": 5.3113662008810304e-05, "loss": 0.004628765396773815, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00464, "step": 276, "tokens/total": 8345952, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 126363}
+{"epoch": 1.08203125, "grad_norm": 0.2942987382411957, "learning_rate": 5.282366752473479e-05, "loss": 0.004942136351019144, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00495, "step": 277, "tokens/total": 8376416, "tokens/train_per_sec_per_gpu": 34.04, "tokens/trainable": 126828}
+{"epoch": 1.0859375, "grad_norm": 0.04225993901491165, "learning_rate": 5.2533763606737005e-05, "loss": 0.00015706325939390808, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00016, "step": 278, "tokens/total": 8406640, "tokens/train_per_sec_per_gpu": 38.21, "tokens/trainable": 127331}
+{"epoch": 1.08984375, "grad_norm": 0.008294228464365005, "learning_rate": 5.224396231890232e-05, "loss": 9.9621967819985e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.0001, "step": 279, "tokens/total": 8436784, "tokens/train_per_sec_per_gpu": 34.39, "tokens/trainable": 127798}
+{"epoch": 1.09375, "grad_norm": 0.017988065257668495, "learning_rate": 5.195427572104522e-05, "loss": 0.00021193400607444346, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00021, "step": 280, "tokens/total": 8467152, "tokens/train_per_sec_per_gpu": 32.84, "tokens/trainable": 128247}
+{"epoch": 1.09765625, "grad_norm": 0.055866386741399765, "learning_rate": 5.166471586820751e-05, "loss": 0.0003000127908308059, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.0003, "step": 281, "tokens/total": 8497424, "tokens/train_per_sec_per_gpu": 36.22, "tokens/trainable": 128729}
+{"epoch": 1.1015625, "grad_norm": 0.03349864110350609, "learning_rate": 5.1375294810156615e-05, "loss": 0.0003475213306955993, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00035, "step": 282, "tokens/total": 8527616, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 129203}
+{"epoch": 1.10546875, "grad_norm": 0.0013503885129466653, "learning_rate": 5.1086024590884144e-05, "loss": 2.522556678741239e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00003, "step": 283, "tokens/total": 8557984, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 129653}
+{"epoch": 1.109375, "grad_norm": 0.03435222804546356, "learning_rate": 5.079691724810461e-05, "loss": 0.00024076660338323563, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00024, "step": 284, "tokens/total": 8588384, "tokens/train_per_sec_per_gpu": 32.63, "tokens/trainable": 130100}
+{"epoch": 1.11328125, "grad_norm": 0.06312273442745209, "learning_rate": 5.0507984812754684e-05, "loss": 0.0001932874001795426, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00019, "step": 285, "tokens/total": 8618768, "tokens/train_per_sec_per_gpu": 30.49, "tokens/trainable": 130543}
+{"epoch": 1.1171875, "grad_norm": 0.01356665138155222, "learning_rate": 5.021923930849237e-05, "loss": 0.00014943527639843524, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00015, "step": 286, "tokens/total": 8648976, "tokens/train_per_sec_per_gpu": 34.29, "tokens/trainable": 131000}
+{"epoch": 1.12109375, "grad_norm": 0.008714325726032257, "learning_rate": 4.99306927511967e-05, "loss": 9.58888849709183e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.0001, "step": 287, "tokens/total": 8679152, "tokens/train_per_sec_per_gpu": 34.04, "tokens/trainable": 131449}
+{"epoch": 1.125, "grad_norm": 0.004380185157060623, "learning_rate": 4.964235714846775e-05, "loss": 3.911522071575746e-05, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00004, "step": 288, "tokens/total": 8709504, "tokens/train_per_sec_per_gpu": 35.97, "tokens/trainable": 131923}
+{"epoch": 1.12890625, "grad_norm": 0.3479631245136261, "learning_rate": 4.9354244499126866e-05, "loss": 0.003404036397114396, "memory/device_reserved (GiB)": 35.91, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00341, "step": 289, "tokens/total": 8739680, "tokens/train_per_sec_per_gpu": 33.15, "tokens/trainable": 132397}
+{"epoch": 1.1328125, "grad_norm": 0.0024364532437175512, "learning_rate": 4.90663667927174e-05, "loss": 2.929499896708876e-05, "memory/device_reserved (GiB)": 35.57, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00003, "step": 290, "tokens/total": 8770112, "tokens/train_per_sec_per_gpu": 36.4, "tokens/trainable": 132846}
+{"epoch": 1.13671875, "grad_norm": 0.5378495454788208, "learning_rate": 4.877873600900581e-05, "loss": 0.0011512563796713948, "memory/device_reserved (GiB)": 35.57, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00115, "step": 291, "tokens/total": 8800368, "tokens/train_per_sec_per_gpu": 35.74, "tokens/trainable": 133347}
+{"epoch": 1.140625, "grad_norm": 0.005875240080058575, "learning_rate": 4.849136411748306e-05, "loss": 4.1250437789130956e-05, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00004, "step": 292, "tokens/total": 8830688, "tokens/train_per_sec_per_gpu": 34.29, "tokens/trainable": 133779}
+{"epoch": 1.14453125, "grad_norm": 0.04661871865391731, "learning_rate": 4.8204263076866574e-05, "loss": 0.00011396713671274483, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00011, "step": 293, "tokens/total": 8861008, "tokens/train_per_sec_per_gpu": 36.33, "tokens/trainable": 134251}
+{"epoch": 1.1484375, "grad_norm": 0.0046598357148468494, "learning_rate": 4.791744483460251e-05, "loss": 3.974197170464322e-05, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00004, "step": 294, "tokens/total": 8891760, "tokens/train_per_sec_per_gpu": 33.55, "tokens/trainable": 134683}
+{"epoch": 1.15234375, "grad_norm": 0.0034782576840370893, "learning_rate": 4.7630921326368736e-05, "loss": 1.7245467461179942e-05, "memory/device_reserved (GiB)": 35.59, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00002, "step": 295, "tokens/total": 8922192, "tokens/train_per_sec_per_gpu": 35.57, "tokens/trainable": 135165}
+{"epoch": 1.15625, "grad_norm": 0.007050055079162121, "learning_rate": 4.7344704475577916e-05, "loss": 6.337351078400388e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00006, "step": 296, "tokens/total": 8952496, "tokens/train_per_sec_per_gpu": 33.24, "tokens/trainable": 135599}
+{"epoch": 1.16015625, "grad_norm": 0.0338653139770031, "learning_rate": 4.705880619288153e-05, "loss": 0.00016246488667093217, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00016, "step": 297, "tokens/total": 8982768, "tokens/train_per_sec_per_gpu": 32.67, "tokens/trainable": 136038}
+{"epoch": 1.1640625, "grad_norm": 0.014626193791627884, "learning_rate": 4.677323837567412e-05, "loss": 0.00016142457025125623, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00016, "step": 298, "tokens/total": 9013248, "tokens/train_per_sec_per_gpu": 32.9, "tokens/trainable": 136495}
+{"epoch": 1.16796875, "grad_norm": 0.0024293530732393265, "learning_rate": 4.6488012907598146e-05, "loss": 2.855894126696512e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00003, "step": 299, "tokens/total": 9043632, "tokens/train_per_sec_per_gpu": 35.34, "tokens/trainable": 136953}
+{"epoch": 1.171875, "grad_norm": 0.027738185599446297, "learning_rate": 4.620314165804964e-05, "loss": 0.00019799786969088018, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.0002, "step": 300, "tokens/total": 9073936, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 137412}
+{"epoch": 1.17578125, "grad_norm": 0.22624947130680084, "learning_rate": 4.591863648168407e-05, "loss": 0.0013185559073463082, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00132, "step": 301, "tokens/total": 9104416, "tokens/train_per_sec_per_gpu": 35.68, "tokens/trainable": 137907}
+{"epoch": 1.1796875, "grad_norm": 0.034625228494405746, "learning_rate": 4.5634509217923135e-05, "loss": 0.00012454690295271575, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00012, "step": 302, "tokens/total": 9134688, "tokens/train_per_sec_per_gpu": 34.33, "tokens/trainable": 138383}
+{"epoch": 1.18359375, "grad_norm": 0.005657304544001818, "learning_rate": 4.535077169046201e-05, "loss": 5.558782140724361e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00006, "step": 303, "tokens/total": 9165088, "tokens/train_per_sec_per_gpu": 36.4, "tokens/trainable": 138842}
+{"epoch": 1.1875, "grad_norm": 0.0075146509334445, "learning_rate": 4.506743570677743e-05, "loss": 1.902109943330288e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 304, "tokens/total": 9195584, "tokens/train_per_sec_per_gpu": 30.67, "tokens/trainable": 139269}
+{"epoch": 1.19140625, "grad_norm": 0.0024459187407046556, "learning_rate": 4.478451305763618e-05, "loss": 2.344881431781687e-05, "memory/device_reserved (GiB)": 35.67, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00002, "step": 305, "tokens/total": 9225760, "tokens/train_per_sec_per_gpu": 34.45, "tokens/trainable": 139736}
+{"epoch": 1.1953125, "grad_norm": 0.18014587461948395, "learning_rate": 4.450201551660454e-05, "loss": 0.0005937310634180903, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00059, "step": 306, "tokens/total": 9256352, "tokens/train_per_sec_per_gpu": 37.01, "tokens/trainable": 140226}
+{"epoch": 1.19921875, "grad_norm": 0.0010275261010974646, "learning_rate": 4.4219954839558276e-05, "loss": 2.1100560843478888e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.19, "memory/max_allocated (GiB)": 33.19, "ppl": 1.00002, "step": 307, "tokens/total": 9284256, "tokens/train_per_sec_per_gpu": 33.98, "tokens/trainable": 140664}
+{"epoch": 1.203125, "grad_norm": 0.0054944949224591255, "learning_rate": 4.393834276419352e-05, "loss": 5.6230866903206334e-05, "memory/device_reserved (GiB)": 35.9, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00006, "step": 308, "tokens/total": 9314768, "tokens/train_per_sec_per_gpu": 29.84, "tokens/trainable": 141115}
+{"epoch": 1.20703125, "grad_norm": 0.0006827972829341888, "learning_rate": 4.36571910095382e-05, "loss": 1.4492828995571472e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 309, "tokens/total": 9344976, "tokens/train_per_sec_per_gpu": 28.87, "tokens/trainable": 141530}
+{"epoch": 1.2109375, "grad_norm": 0.7971475720405579, "learning_rate": 4.337651127546448e-05, "loss": 0.0021856159437447786, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00219, "step": 310, "tokens/total": 9375280, "tokens/train_per_sec_per_gpu": 34.2, "tokens/trainable": 141964}
+{"epoch": 1.21484375, "grad_norm": 0.00386450975202024, "learning_rate": 4.3096315242201736e-05, "loss": 1.9838389562210068e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00002, "step": 311, "tokens/total": 9405536, "tokens/train_per_sec_per_gpu": 33.98, "tokens/trainable": 142428}
+{"epoch": 1.21875, "grad_norm": 0.058126050978899, "learning_rate": 4.2816614569850635e-05, "loss": 0.0002090351772494614, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00021, "step": 312, "tokens/total": 9435696, "tokens/train_per_sec_per_gpu": 38.16, "tokens/trainable": 142894}
+{"epoch": 1.22265625, "grad_norm": 0.019857943058013916, "learning_rate": 4.2537420897897864e-05, "loss": 0.00014800178178120404, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00015, "step": 313, "tokens/total": 9466080, "tokens/train_per_sec_per_gpu": 36.37, "tokens/trainable": 143368}
+{"epoch": 1.2265625, "grad_norm": 0.05821572616696358, "learning_rate": 4.225874584473174e-05, "loss": 0.00014745524094905704, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00015, "step": 314, "tokens/total": 9496400, "tokens/train_per_sec_per_gpu": 32.87, "tokens/trainable": 143824}
+{"epoch": 1.23046875, "grad_norm": 0.00997456256300211, "learning_rate": 4.19806010071587e-05, "loss": 0.00011511147022247314, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00012, "step": 315, "tokens/total": 9526880, "tokens/train_per_sec_per_gpu": 34.58, "tokens/trainable": 144306}
+{"epoch": 1.234375, "grad_norm": 0.009190126322209835, "learning_rate": 4.170299795992081e-05, "loss": 6.155027949716896e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00006, "step": 316, "tokens/total": 9557280, "tokens/train_per_sec_per_gpu": 35.45, "tokens/trainable": 144763}
+{"epoch": 1.23828125, "grad_norm": 0.0016322726150974631, "learning_rate": 4.142594825521398e-05, "loss": 2.0669946025009267e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 317, "tokens/total": 9587712, "tokens/train_per_sec_per_gpu": 37.45, "tokens/trainable": 145255}
+{"epoch": 1.2421875, "grad_norm": 0.00745917996391654, "learning_rate": 4.114946342220728e-05, "loss": 5.585006874753162e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00006, "step": 318, "tokens/total": 9618064, "tokens/train_per_sec_per_gpu": 36.7, "tokens/trainable": 145714}
+{"epoch": 1.24609375, "grad_norm": 0.07343094050884247, "learning_rate": 4.087355496656321e-05, "loss": 0.00043384716263972223, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00043, "step": 319, "tokens/total": 9648336, "tokens/train_per_sec_per_gpu": 36.04, "tokens/trainable": 146200}
+{"epoch": 1.25, "grad_norm": 0.007983052171766758, "learning_rate": 4.05982343699588e-05, "loss": 4.524823816609569e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00005, "step": 320, "tokens/total": 9678688, "tokens/train_per_sec_per_gpu": 34.28, "tokens/trainable": 146668}
+{"epoch": 1.25390625, "grad_norm": 0.01599739119410515, "learning_rate": 4.0323513089607876e-05, "loss": 6.388167093973607e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00006, "step": 321, "tokens/total": 9708848, "tokens/train_per_sec_per_gpu": 35.35, "tokens/trainable": 147114}
+{"epoch": 1.2578125, "grad_norm": 0.0042083412408828735, "learning_rate": 4.004940255778431e-05, "loss": 2.1792850020574406e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00002, "step": 322, "tokens/total": 9739360, "tokens/train_per_sec_per_gpu": 35.18, "tokens/trainable": 147564}
+{"epoch": 1.26171875, "grad_norm": 0.009429940953850746, "learning_rate": 3.977591418134619e-05, "loss": 3.834946619463153e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00004, "step": 323, "tokens/total": 9769776, "tokens/train_per_sec_per_gpu": 34.24, "tokens/trainable": 148013}
+{"epoch": 1.265625, "grad_norm": 0.05088210478425026, "learning_rate": 3.95030593412612e-05, "loss": 0.00018766832363326102, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.00019, "step": 324, "tokens/total": 9800096, "tokens/train_per_sec_per_gpu": 38.06, "tokens/trainable": 148470}
+{"epoch": 1.26953125, "grad_norm": 0.9416317939758301, "learning_rate": 3.923084939213296e-05, "loss": 0.01083211787045002, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.01089, "step": 325, "tokens/total": 9830304, "tokens/train_per_sec_per_gpu": 36.53, "tokens/trainable": 148926}
+{"epoch": 1.2734375, "grad_norm": 0.00033186975633725524, "learning_rate": 3.895929566172861e-05, "loss": 8.612486453785095e-06, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 326, "tokens/total": 9860608, "tokens/train_per_sec_per_gpu": 37.82, "tokens/trainable": 149420}
+{"epoch": 1.27734375, "grad_norm": 0.0005477021913975477, "learning_rate": 3.868840945050728e-05, "loss": 1.5825649825274013e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00002, "step": 327, "tokens/total": 9890560, "tokens/train_per_sec_per_gpu": 39.41, "tokens/trainable": 149873}
+{"epoch": 1.28125, "grad_norm": 0.0008259841124527156, "learning_rate": 3.841820203114995e-05, "loss": 1.642670395085588e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 328, "tokens/total": 9920880, "tokens/train_per_sec_per_gpu": 33.86, "tokens/trainable": 150347}
+{"epoch": 1.28515625, "grad_norm": 0.0011461537797003984, "learning_rate": 3.814868464809027e-05, "loss": 1.7516158550279215e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00002, "step": 329, "tokens/total": 9951280, "tokens/train_per_sec_per_gpu": 31.83, "tokens/trainable": 150822}
+{"epoch": 1.2890625, "grad_norm": 0.0022519829217344522, "learning_rate": 3.787986851704667e-05, "loss": 2.000835411308799e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 330, "tokens/total": 9981600, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 151266}
+{"epoch": 1.29296875, "grad_norm": 0.08718931674957275, "learning_rate": 3.7611764824555654e-05, "loss": 0.0003773289208766073, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00038, "step": 331, "tokens/total": 10012016, "tokens/train_per_sec_per_gpu": 37.15, "tokens/trainable": 151734}
+{"epoch": 1.296875, "grad_norm": 0.000883373199030757, "learning_rate": 3.734438472750619e-05, "loss": 1.1746728887374047e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00001, "step": 332, "tokens/total": 10042448, "tokens/train_per_sec_per_gpu": 34.01, "tokens/trainable": 152184}
+{"epoch": 1.30078125, "grad_norm": 0.011426394805312157, "learning_rate": 3.707773935267552e-05, "loss": 5.219353988650255e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00005, "step": 333, "tokens/total": 10073024, "tokens/train_per_sec_per_gpu": 31.85, "tokens/trainable": 152623}
+{"epoch": 1.3046875, "grad_norm": 0.5781233310699463, "learning_rate": 3.68118397962661e-05, "loss": 0.0034865224733948708, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00349, "step": 334, "tokens/total": 10103504, "tokens/train_per_sec_per_gpu": 33.44, "tokens/trainable": 153040}
+{"epoch": 1.30859375, "grad_norm": 0.7213758230209351, "learning_rate": 3.654669712344384e-05, "loss": 0.004715530201792717, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.31, "memory/max_allocated (GiB)": 33.31, "ppl": 1.00473, "step": 335, "tokens/total": 10131632, "tokens/train_per_sec_per_gpu": 31.78, "tokens/trainable": 153463}
+{"epoch": 1.3125, "grad_norm": 0.03131686523556709, "learning_rate": 3.628232236787763e-05, "loss": 0.0002346257824683562, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00023, "step": 336, "tokens/total": 10161824, "tokens/train_per_sec_per_gpu": 33.31, "tokens/trainable": 153908}
+{"epoch": 1.31640625, "grad_norm": 0.03332929685711861, "learning_rate": 3.6018726531280144e-05, "loss": 0.00011981122952420264, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00012, "step": 337, "tokens/total": 10192448, "tokens/train_per_sec_per_gpu": 33.72, "tokens/trainable": 154387}
+{"epoch": 1.3203125, "grad_norm": 0.0045038131065666676, "learning_rate": 3.575592058295017e-05, "loss": 3.5494955227477476e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 338, "tokens/total": 10222704, "tokens/train_per_sec_per_gpu": 33.94, "tokens/trainable": 154836}
+{"epoch": 1.32421875, "grad_norm": 0.0012434936361387372, "learning_rate": 3.549391545931585e-05, "loss": 1.7894679331220686e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00002, "step": 339, "tokens/total": 10253168, "tokens/train_per_sec_per_gpu": 29.81, "tokens/trainable": 155261}
+{"epoch": 1.328125, "grad_norm": 0.3250766098499298, "learning_rate": 3.5232722063479914e-05, "loss": 0.00279021542519331, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00279, "step": 340, "tokens/total": 10283552, "tokens/train_per_sec_per_gpu": 31.24, "tokens/trainable": 155674}
+{"epoch": 1.33203125, "grad_norm": 0.036490436643362045, "learning_rate": 3.49723512647657e-05, "loss": 0.00013251493510324508, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00013, "step": 341, "tokens/total": 10313872, "tokens/train_per_sec_per_gpu": 35.78, "tokens/trainable": 156164}
+{"epoch": 1.3359375, "grad_norm": 0.2706057131290436, "learning_rate": 3.471281389826491e-05, "loss": 0.0021429890766739845, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00215, "step": 342, "tokens/total": 10344256, "tokens/train_per_sec_per_gpu": 33.65, "tokens/trainable": 156611}
+{"epoch": 1.33984375, "grad_norm": 0.01915346086025238, "learning_rate": 3.4454120764386764e-05, "loss": 0.00013777356070932, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00014, "step": 343, "tokens/total": 10374544, "tokens/train_per_sec_per_gpu": 33.68, "tokens/trainable": 157073}
+{"epoch": 1.34375, "grad_norm": 0.02540464885532856, "learning_rate": 3.4196282628408526e-05, "loss": 5.135099490871653e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00005, "step": 344, "tokens/total": 10405040, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 157523}
+{"epoch": 1.34765625, "grad_norm": 0.39257746934890747, "learning_rate": 3.3939310220027456e-05, "loss": 0.009803109802305698, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00985, "step": 345, "tokens/total": 10435424, "tokens/train_per_sec_per_gpu": 35.16, "tokens/trainable": 157964}
+{"epoch": 1.3515625, "grad_norm": 0.0029070202726870775, "learning_rate": 3.3683214232914404e-05, "loss": 2.9972559786983766e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00003, "step": 346, "tokens/total": 10465760, "tokens/train_per_sec_per_gpu": 30.44, "tokens/trainable": 158390}
+{"epoch": 1.35546875, "grad_norm": 0.020276719704270363, "learning_rate": 3.342800532426873e-05, "loss": 7.613154593855143e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00008, "step": 347, "tokens/total": 10495904, "tokens/train_per_sec_per_gpu": 35.57, "tokens/trainable": 158848}
+{"epoch": 1.359375, "grad_norm": 0.14866818487644196, "learning_rate": 3.317369411437484e-05, "loss": 0.0011244683992117643, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 34.0, "memory/max_allocated (GiB)": 34.0, "ppl": 1.00113, "step": 348, "tokens/total": 10526624, "tokens/train_per_sec_per_gpu": 28.71, "tokens/trainable": 159270}
+{"epoch": 1.36328125, "grad_norm": 0.015379665419459343, "learning_rate": 3.292029118616024e-05, "loss": 7.920589268906042e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00008, "step": 349, "tokens/total": 10556752, "tokens/train_per_sec_per_gpu": 32.51, "tokens/trainable": 159720}
+{"epoch": 1.3671875, "grad_norm": 0.0023695260751992464, "learning_rate": 3.266780708475511e-05, "loss": 1.4566459867637604e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.99, "memory/max_allocated (GiB)": 33.99, "ppl": 1.00001, "step": 350, "tokens/total": 10587264, "tokens/train_per_sec_per_gpu": 35.14, "tokens/trainable": 160187}
+{"epoch": 1.37109375, "grad_norm": 0.00954374298453331, "learning_rate": 3.241625231705354e-05, "loss": 6.530048267450184e-05, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00007, "step": 351, "tokens/total": 10617504, "tokens/train_per_sec_per_gpu": 32.63, "tokens/trainable": 160645}
+{"epoch": 1.375, "grad_norm": 0.6504904627799988, "learning_rate": 3.216563735127618e-05, "loss": 0.01059151440858841, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.01065, "step": 352, "tokens/total": 10647760, "tokens/train_per_sec_per_gpu": 34.49, "tokens/trainable": 161102}
+{"epoch": 1.37890625, "grad_norm": 0.024732626974582672, "learning_rate": 3.191597261653475e-05, "loss": 0.0001429672265658155, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00014, "step": 353, "tokens/total": 10678080, "tokens/train_per_sec_per_gpu": 28.93, "tokens/trainable": 161555}
+{"epoch": 1.3828125, "grad_norm": 0.037309419363737106, "learning_rate": 3.166726850239794e-05, "loss": 7.214388460852206e-05, "memory/device_reserved (GiB)": 35.41, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00007, "step": 354, "tokens/total": 10708544, "tokens/train_per_sec_per_gpu": 36.75, "tokens/trainable": 162020}
+{"epoch": 1.38671875, "grad_norm": 0.00885045062750578, "learning_rate": 3.141953535845912e-05, "loss": 0.00011830432777060196, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00012, "step": 355, "tokens/total": 10739024, "tokens/train_per_sec_per_gpu": 32.1, "tokens/trainable": 162448}
+{"epoch": 1.390625, "grad_norm": 0.06791900098323822, "learning_rate": 3.11727834939056e-05, "loss": 0.0003609730047173798, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.61, "memory/max_allocated (GiB)": 33.61, "ppl": 1.00036, "step": 356, "tokens/total": 10769024, "tokens/train_per_sec_per_gpu": 28.26, "tokens/trainable": 162863}
+{"epoch": 1.39453125, "grad_norm": 0.0019508687546476722, "learning_rate": 3.092702317708967e-05, "loss": 3.1739040423417464e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00003, "step": 357, "tokens/total": 10799328, "tokens/train_per_sec_per_gpu": 34.89, "tokens/trainable": 163318}
+{"epoch": 1.3984375, "grad_norm": 0.7139555811882019, "learning_rate": 3.0682264635101276e-05, "loss": 0.010268578305840492, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.01032, "step": 358, "tokens/total": 10829488, "tokens/train_per_sec_per_gpu": 33.08, "tokens/trainable": 163767}
+{"epoch": 1.40234375, "grad_norm": 0.006852464284747839, "learning_rate": 3.0438518053342407e-05, "loss": 4.873241778113879e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00005, "step": 359, "tokens/total": 10859936, "tokens/train_per_sec_per_gpu": 31.94, "tokens/trainable": 164208}
+{"epoch": 1.40625, "grad_norm": 0.0018247900297865272, "learning_rate": 3.0195793575103266e-05, "loss": 3.88835760531947e-05, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00004, "step": 360, "tokens/total": 10890400, "tokens/train_per_sec_per_gpu": 34.18, "tokens/trainable": 164655}
+{"epoch": 1.41015625, "grad_norm": 0.0436873696744442, "learning_rate": 2.9954101301140146e-05, "loss": 0.0003687943681143224, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00037, "step": 361, "tokens/total": 10920624, "tokens/train_per_sec_per_gpu": 34.15, "tokens/trainable": 165100}
+{"epoch": 1.4140625, "grad_norm": 0.26735150814056396, "learning_rate": 2.9713451289255123e-05, "loss": 0.00039605735219083726, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.0004, "step": 362, "tokens/total": 10951136, "tokens/train_per_sec_per_gpu": 34.91, "tokens/trainable": 165587}
+{"epoch": 1.41796875, "grad_norm": 0.0964483991265297, "learning_rate": 2.9473853553877484e-05, "loss": 0.0007330900407396257, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00073, "step": 363, "tokens/total": 10981344, "tokens/train_per_sec_per_gpu": 34.14, "tokens/trainable": 166034}
+{"epoch": 1.421875, "grad_norm": 0.05367618799209595, "learning_rate": 2.9235318065647e-05, "loss": 0.0005384897813200951, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00054, "step": 364, "tokens/total": 11011552, "tokens/train_per_sec_per_gpu": 34.1, "tokens/trainable": 166485}
+{"epoch": 1.42578125, "grad_norm": 0.10067912936210632, "learning_rate": 2.8997854750998964e-05, "loss": 0.0006868956843391061, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00069, "step": 365, "tokens/total": 11041872, "tokens/train_per_sec_per_gpu": 36.36, "tokens/trainable": 166973}
+{"epoch": 1.4296875, "grad_norm": 0.02697882056236267, "learning_rate": 2.8761473491751258e-05, "loss": 0.0002464794088155031, "memory/device_reserved (GiB)": 35.84, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00025, "step": 366, "tokens/total": 11072192, "tokens/train_per_sec_per_gpu": 29.12, "tokens/trainable": 167414}
+{"epoch": 1.43359375, "grad_norm": 0.0026214183308184147, "learning_rate": 2.8526184124692883e-05, "loss": 5.2629769925260916e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00005, "step": 367, "tokens/total": 11102736, "tokens/train_per_sec_per_gpu": 30.23, "tokens/trainable": 167851}
+{"epoch": 1.4375, "grad_norm": 0.0018140808679163456, "learning_rate": 2.829199644117484e-05, "loss": 2.9512597393477336e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00003, "step": 368, "tokens/total": 11133056, "tokens/train_per_sec_per_gpu": 35.56, "tokens/trainable": 168294}
+{"epoch": 1.44140625, "grad_norm": 0.012045321986079216, "learning_rate": 2.8058920186702553e-05, "loss": 7.606908184243366e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00008, "step": 369, "tokens/total": 11163568, "tokens/train_per_sec_per_gpu": 32.23, "tokens/trainable": 168769}
+{"epoch": 1.4453125, "grad_norm": 0.013453424908220768, "learning_rate": 2.782696506053033e-05, "loss": 0.00015325279673561454, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00015, "step": 370, "tokens/total": 11193936, "tokens/train_per_sec_per_gpu": 36.83, "tokens/trainable": 169270}
+{"epoch": 1.44921875, "grad_norm": 0.003458227962255478, "learning_rate": 2.7596140715257824e-05, "loss": 5.63332432648167e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00006, "step": 371, "tokens/total": 11224480, "tokens/train_per_sec_per_gpu": 35.08, "tokens/trainable": 169711}
+{"epoch": 1.453125, "grad_norm": 0.0063622924499213696, "learning_rate": 2.7366456756428184e-05, "loss": 9.534892160445452e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.0001, "step": 372, "tokens/total": 11254912, "tokens/train_per_sec_per_gpu": 32.88, "tokens/trainable": 170185}
+{"epoch": 1.45703125, "grad_norm": 0.04558980092406273, "learning_rate": 2.7137922742128486e-05, "loss": 0.000613630109000951, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00061, "step": 373, "tokens/total": 11285104, "tokens/train_per_sec_per_gpu": 33.12, "tokens/trainable": 170650}
+{"epoch": 1.4609375, "grad_norm": 0.014873786829411983, "learning_rate": 2.691054818259188e-05, "loss": 0.0002038514503510669, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.0002, "step": 374, "tokens/total": 11315600, "tokens/train_per_sec_per_gpu": 31.99, "tokens/trainable": 171126}
+{"epoch": 1.46484375, "grad_norm": 0.058739252388477325, "learning_rate": 2.6684342539801933e-05, "loss": 0.0010346119524911046, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00104, "step": 375, "tokens/total": 11345776, "tokens/train_per_sec_per_gpu": 35.6, "tokens/trainable": 171600}
+{"epoch": 1.46875, "grad_norm": 0.24105991423130035, "learning_rate": 2.645931522709877e-05, "loss": 0.0035716122947633266, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00358, "step": 376, "tokens/total": 11376208, "tokens/train_per_sec_per_gpu": 28.14, "tokens/trainable": 172017}
+{"epoch": 1.47265625, "grad_norm": 0.08793950080871582, "learning_rate": 2.6235475608787365e-05, "loss": 0.0007833336712792516, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00078, "step": 377, "tokens/total": 11406512, "tokens/train_per_sec_per_gpu": 36.07, "tokens/trainable": 172493}
+{"epoch": 1.4765625, "grad_norm": 0.009154544211924076, "learning_rate": 2.6012832999747916e-05, "loss": 0.00011752049613278359, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00012, "step": 378, "tokens/total": 11436544, "tokens/train_per_sec_per_gpu": 34.4, "tokens/trainable": 172951}
+{"epoch": 1.48046875, "grad_norm": 0.012918495573103428, "learning_rate": 2.579139666504821e-05, "loss": 0.00011567945330170915, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00012, "step": 379, "tokens/total": 11467008, "tokens/train_per_sec_per_gpu": 35.72, "tokens/trainable": 173435}
+{"epoch": 1.484375, "grad_norm": 0.008966504596173763, "learning_rate": 2.557117581955798e-05, "loss": 8.938623068388551e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00009, "step": 380, "tokens/total": 11497184, "tokens/train_per_sec_per_gpu": 30.47, "tokens/trainable": 173900}
+{"epoch": 1.48828125, "grad_norm": 0.03181751072406769, "learning_rate": 2.5352179627565532e-05, "loss": 0.00013083787052892148, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00013, "step": 381, "tokens/total": 11527600, "tokens/train_per_sec_per_gpu": 34.13, "tokens/trainable": 174371}
+{"epoch": 1.4921875, "grad_norm": 0.0009072935208678246, "learning_rate": 2.5134417202396277e-05, "loss": 2.5600436856620945e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00003, "step": 382, "tokens/total": 11557808, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 174826}
+{"epoch": 1.49609375, "grad_norm": 0.0050596860237419605, "learning_rate": 2.491789760603361e-05, "loss": 5.778766353614628e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00006, "step": 383, "tokens/total": 11588352, "tokens/train_per_sec_per_gpu": 32.88, "tokens/trainable": 175276}
+{"epoch": 1.5, "grad_norm": 0.008304890245199203, "learning_rate": 2.4702629848741764e-05, "loss": 0.00012505470658652484, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00013, "step": 384, "tokens/total": 11618640, "tokens/train_per_sec_per_gpu": 33.71, "tokens/trainable": 175720}
+{"epoch": 1.50390625, "grad_norm": 0.0035276354756206274, "learning_rate": 2.4488622888690785e-05, "loss": 4.851898120250553e-05, "memory/device_reserved (GiB)": 35.89, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00005, "step": 385, "tokens/total": 11649072, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 176203}
+{"epoch": 1.5078125, "grad_norm": 0.001593019813299179, "learning_rate": 2.427588563158384e-05, "loss": 3.330651088617742e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00003, "step": 386, "tokens/total": 11679552, "tokens/train_per_sec_per_gpu": 34.19, "tokens/trainable": 176651}
+{"epoch": 1.51171875, "grad_norm": 0.002167076338082552, "learning_rate": 2.406442693028651e-05, "loss": 3.344817378092557e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00003, "step": 387, "tokens/total": 11709760, "tokens/train_per_sec_per_gpu": 32.74, "tokens/trainable": 177115}
+{"epoch": 1.515625, "grad_norm": 0.003447546623647213, "learning_rate": 2.3854255584458547e-05, "loss": 5.5277254432439804e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00006, "step": 388, "tokens/total": 11740288, "tokens/train_per_sec_per_gpu": 36.13, "tokens/trainable": 177601}
+{"epoch": 1.51953125, "grad_norm": 0.0052979388274252415, "learning_rate": 2.3645380340187508e-05, "loss": 4.56162124464754e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00005, "step": 389, "tokens/total": 11770592, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 178050}
+{"epoch": 1.5234375, "grad_norm": 0.003032457083463669, "learning_rate": 2.3437809889624914e-05, "loss": 3.829343768302351e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00004, "step": 390, "tokens/total": 11800752, "tokens/train_per_sec_per_gpu": 36.42, "tokens/trainable": 178531}
+{"epoch": 1.52734375, "grad_norm": 0.03542603179812431, "learning_rate": 2.3231552870624487e-05, "loss": 0.0003605492820497602, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00036, "step": 391, "tokens/total": 11831360, "tokens/train_per_sec_per_gpu": 33.98, "tokens/trainable": 179000}
+{"epoch": 1.53125, "grad_norm": 0.004965933505445719, "learning_rate": 2.3026617866382657e-05, "loss": 7.091004226822406e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00007, "step": 392, "tokens/total": 11861552, "tokens/train_per_sec_per_gpu": 33.1, "tokens/trainable": 179457}
+{"epoch": 1.53515625, "grad_norm": 0.006339129991829395, "learning_rate": 2.2823013405081507e-05, "loss": 9.15761775104329e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00009, "step": 393, "tokens/total": 11891904, "tokens/train_per_sec_per_gpu": 32.81, "tokens/trainable": 179890}
+{"epoch": 1.5390625, "grad_norm": 0.03921438008546829, "learning_rate": 2.2620747959533722e-05, "loss": 0.00038214243249967694, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00038, "step": 394, "tokens/total": 11922208, "tokens/train_per_sec_per_gpu": 33.95, "tokens/trainable": 180341}
+{"epoch": 1.54296875, "grad_norm": 0.08640608936548233, "learning_rate": 2.2419829946830123e-05, "loss": 0.0006777399685233831, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00068, "step": 395, "tokens/total": 11952672, "tokens/train_per_sec_per_gpu": 32.99, "tokens/trainable": 180762}
+{"epoch": 1.546875, "grad_norm": 0.08661718666553497, "learning_rate": 2.2220267727989325e-05, "loss": 0.0007016840972937644, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.0007, "step": 396, "tokens/total": 11983088, "tokens/train_per_sec_per_gpu": 35.18, "tokens/trainable": 181226}
+{"epoch": 1.55078125, "grad_norm": 0.010069145821034908, "learning_rate": 2.202206960760984e-05, "loss": 0.00011120665294583887, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00011, "step": 397, "tokens/total": 12013488, "tokens/train_per_sec_per_gpu": 33.18, "tokens/trainable": 181714}
+{"epoch": 1.5546875, "grad_norm": 0.03158818930387497, "learning_rate": 2.182524383352446e-05, "loss": 0.00024857826065272093, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00025, "step": 398, "tokens/total": 12043968, "tokens/train_per_sec_per_gpu": 35.57, "tokens/trainable": 182142}
+{"epoch": 1.55859375, "grad_norm": 0.000873154669534415, "learning_rate": 2.1629798596457056e-05, "loss": 2.1218824258539826e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00002, "step": 399, "tokens/total": 12074160, "tokens/train_per_sec_per_gpu": 34.86, "tokens/trainable": 182604}
+{"epoch": 1.5625, "grad_norm": 0.009255572222173214, "learning_rate": 2.1435742029681725e-05, "loss": 0.0001349164522252977, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.67, "memory/max_allocated (GiB)": 33.67, "ppl": 1.00013, "step": 400, "tokens/total": 12104352, "tokens/train_per_sec_per_gpu": 25.31, "tokens/trainable": 183011}
+{"epoch": 1.56640625, "grad_norm": 0.010521743446588516, "learning_rate": 2.124308220868431e-05, "loss": 7.793466647854075e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.71, "memory/max_allocated (GiB)": 33.71, "ppl": 1.00008, "step": 401, "tokens/total": 12134240, "tokens/train_per_sec_per_gpu": 31.81, "tokens/trainable": 183459}
+{"epoch": 1.5703125, "grad_norm": 0.0013409090461209416, "learning_rate": 2.105182715082638e-05, "loss": 2.7722922823159024e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 402, "tokens/total": 12164480, "tokens/train_per_sec_per_gpu": 38.7, "tokens/trainable": 183959}
+{"epoch": 1.57421875, "grad_norm": 0.10361713916063309, "learning_rate": 2.0861984815011552e-05, "loss": 0.0014142843428999186, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00142, "step": 403, "tokens/total": 12194640, "tokens/train_per_sec_per_gpu": 32.5, "tokens/trainable": 184431}
+{"epoch": 1.578125, "grad_norm": 0.0026048943400382996, "learning_rate": 2.0673563101354323e-05, "loss": 4.14503738284111e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00004, "step": 404, "tokens/total": 12224960, "tokens/train_per_sec_per_gpu": 31.37, "tokens/trainable": 184867}
+{"epoch": 1.58203125, "grad_norm": 0.009411919862031937, "learning_rate": 2.0486569850851317e-05, "loss": 7.322158489841968e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.72, "memory/max_allocated (GiB)": 33.72, "ppl": 1.00007, "step": 405, "tokens/total": 12255008, "tokens/train_per_sec_per_gpu": 29.37, "tokens/trainable": 185300}
+{"epoch": 1.5859375, "grad_norm": 0.006792420521378517, "learning_rate": 2.0301012845054956e-05, "loss": 7.982828537933528e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00008, "step": 406, "tokens/total": 12285248, "tokens/train_per_sec_per_gpu": 31.48, "tokens/trainable": 185754}
+{"epoch": 1.58984375, "grad_norm": 0.013020367361605167, "learning_rate": 2.011689980574966e-05, "loss": 3.743396155186929e-05, "memory/device_reserved (GiB)": 35.88, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00004, "step": 407, "tokens/total": 12315552, "tokens/train_per_sec_per_gpu": 36.82, "tokens/trainable": 186213}
+{"epoch": 1.59375, "grad_norm": 0.0013089004205539823, "learning_rate": 1.993423839463052e-05, "loss": 2.306591341039166e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00002, "step": 408, "tokens/total": 12345904, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 186659}
+{"epoch": 1.59765625, "grad_norm": 0.0012270932784304023, "learning_rate": 1.975303621298445e-05, "loss": 2.1350417227949947e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 409, "tokens/total": 12376336, "tokens/train_per_sec_per_gpu": 32.97, "tokens/trainable": 187083}
+{"epoch": 1.6015625, "grad_norm": 0.011230082251131535, "learning_rate": 1.957330080137385e-05, "loss": 3.752015618374571e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00004, "step": 410, "tokens/total": 12406880, "tokens/train_per_sec_per_gpu": 35.01, "tokens/trainable": 187530}
+{"epoch": 1.60546875, "grad_norm": 0.004872993566095829, "learning_rate": 1.9395039639322864e-05, "loss": 5.5653974413871765e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00006, "step": 411, "tokens/total": 12437088, "tokens/train_per_sec_per_gpu": 34.32, "tokens/trainable": 187972}
+{"epoch": 1.609375, "grad_norm": 0.19391997158527374, "learning_rate": 1.9218260145006073e-05, "loss": 0.000892186420969665, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00089, "step": 412, "tokens/total": 12465440, "tokens/train_per_sec_per_gpu": 39.95, "tokens/trainable": 188417}
+{"epoch": 1.61328125, "grad_norm": 0.09062197804450989, "learning_rate": 1.904296967493982e-05, "loss": 0.0006304415874183178, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00063, "step": 413, "tokens/total": 12495968, "tokens/train_per_sec_per_gpu": 31.57, "tokens/trainable": 188868}
+{"epoch": 1.6171875, "grad_norm": 0.009724686853587627, "learning_rate": 1.8869175523676064e-05, "loss": 6.72380265314132e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00007, "step": 414, "tokens/total": 12526128, "tokens/train_per_sec_per_gpu": 36.95, "tokens/trainable": 189337}
+{"epoch": 1.62109375, "grad_norm": 0.19542834162712097, "learning_rate": 1.869688492349885e-05, "loss": 0.001907092402689159, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00191, "step": 415, "tokens/total": 12556256, "tokens/train_per_sec_per_gpu": 31.94, "tokens/trainable": 189750}
+{"epoch": 1.625, "grad_norm": 0.0465276800096035, "learning_rate": 1.85261050441233e-05, "loss": 0.0005661146715283394, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00057, "step": 416, "tokens/total": 12586736, "tokens/train_per_sec_per_gpu": 34.06, "tokens/trainable": 190216}
+{"epoch": 1.62890625, "grad_norm": 0.0012954205740243196, "learning_rate": 1.8356842992397304e-05, "loss": 1.8871982319978997e-05, "memory/device_reserved (GiB)": 35.96, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00002, "step": 417, "tokens/total": 12617168, "tokens/train_per_sec_per_gpu": 35.01, "tokens/trainable": 190708}
+{"epoch": 1.6328125, "grad_norm": 0.0018749319715425372, "learning_rate": 1.8189105812005714e-05, "loss": 3.220669532311149e-05, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00003, "step": 418, "tokens/total": 12647680, "tokens/train_per_sec_per_gpu": 30.9, "tokens/trainable": 191126}
+{"epoch": 1.63671875, "grad_norm": 0.0022480092011392117, "learning_rate": 1.802290048317732e-05, "loss": 2.6564141080598347e-05, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00003, "step": 419, "tokens/total": 12677952, "tokens/train_per_sec_per_gpu": 30.34, "tokens/trainable": 191540}
+{"epoch": 1.640625, "grad_norm": 0.1634059101343155, "learning_rate": 1.785823392239424e-05, "loss": 0.0009822181891649961, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00098, "step": 420, "tokens/total": 12708416, "tokens/train_per_sec_per_gpu": 36.69, "tokens/trainable": 192014}
+{"epoch": 1.64453125, "grad_norm": 0.0019404747290536761, "learning_rate": 1.7695112982104225e-05, "loss": 2.167341335734818e-05, "memory/device_reserved (GiB)": 35.8, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00002, "step": 421, "tokens/total": 12738640, "tokens/train_per_sec_per_gpu": 31.31, "tokens/trainable": 192463}
+{"epoch": 1.6484375, "grad_norm": 0.001059795613400638, "learning_rate": 1.7533544450435433e-05, "loss": 1.6543437595828436e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00002, "step": 422, "tokens/total": 12769232, "tokens/train_per_sec_per_gpu": 32.74, "tokens/trainable": 192918}
+{"epoch": 1.65234375, "grad_norm": 0.01760418340563774, "learning_rate": 1.7373535050913946e-05, "loss": 0.0001707405026536435, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00017, "step": 423, "tokens/total": 12799648, "tokens/train_per_sec_per_gpu": 37.46, "tokens/trainable": 193425}
+{"epoch": 1.65625, "grad_norm": 0.0023663989268243313, "learning_rate": 1.721509144218405e-05, "loss": 2.332203439436853e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 424, "tokens/total": 12829920, "tokens/train_per_sec_per_gpu": 35.25, "tokens/trainable": 193898}
+{"epoch": 1.66015625, "grad_norm": 0.0245437640696764, "learning_rate": 1.705822021773101e-05, "loss": 0.0001577093207743019, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00016, "step": 425, "tokens/total": 12860112, "tokens/train_per_sec_per_gpu": 29.29, "tokens/trainable": 194315}
+{"epoch": 1.6640625, "grad_norm": 0.0009002169244922698, "learning_rate": 1.69029279056068e-05, "loss": 1.4283305972639937e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00001, "step": 426, "tokens/total": 12890320, "tokens/train_per_sec_per_gpu": 34.92, "tokens/trainable": 194796}
+{"epoch": 1.66796875, "grad_norm": 0.00033997284481301904, "learning_rate": 1.6749220968158415e-05, "loss": 7.185776667029131e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00001, "step": 427, "tokens/total": 12920656, "tokens/train_per_sec_per_gpu": 33.96, "tokens/trainable": 195262}
+{"epoch": 1.671875, "grad_norm": 0.0005359657225199044, "learning_rate": 1.659710580175893e-05, "loss": 9.948088518285658e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 428, "tokens/total": 12951040, "tokens/train_per_sec_per_gpu": 31.91, "tokens/trainable": 195680}
+{"epoch": 1.67578125, "grad_norm": 0.0014890795573592186, "learning_rate": 1.644658873654133e-05, "loss": 2.028812377830036e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.75, "memory/max_allocated (GiB)": 33.75, "ppl": 1.00002, "step": 429, "tokens/total": 12981232, "tokens/train_per_sec_per_gpu": 32.89, "tokens/trainable": 196145}
+{"epoch": 1.6796875, "grad_norm": 0.05130980163812637, "learning_rate": 1.629767603613508e-05, "loss": 0.0003292672336101532, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00033, "step": 430, "tokens/total": 13011616, "tokens/train_per_sec_per_gpu": 32.42, "tokens/trainable": 196626}
+{"epoch": 1.68359375, "grad_norm": 0.0014070137403905392, "learning_rate": 1.615037389740547e-05, "loss": 1.0152909453609027e-05, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.00001, "step": 431, "tokens/total": 13042208, "tokens/train_per_sec_per_gpu": 31.17, "tokens/trainable": 197067}
+{"epoch": 1.6875, "grad_norm": 0.004116157069802284, "learning_rate": 1.600468845019576e-05, "loss": 3.514952550176531e-05, "memory/device_reserved (GiB)": 36.35, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00004, "step": 432, "tokens/total": 13072800, "tokens/train_per_sec_per_gpu": 38.68, "tokens/trainable": 197565}
+{"epoch": 1.69140625, "grad_norm": 0.01147819496691227, "learning_rate": 1.5860625757072092e-05, "loss": 7.313965761568397e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.95, "memory/max_allocated (GiB)": 33.95, "ppl": 1.00007, "step": 433, "tokens/total": 13103328, "tokens/train_per_sec_per_gpu": 32.96, "tokens/trainable": 198015}
+{"epoch": 1.6953125, "grad_norm": 0.030784178525209427, "learning_rate": 1.571819181307116e-05, "loss": 0.0002448821614962071, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00024, "step": 434, "tokens/total": 13133472, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 198442}
+{"epoch": 1.69921875, "grad_norm": 0.0022457120940089226, "learning_rate": 1.557739254545075e-05, "loss": 1.4446297427639365e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 435, "tokens/total": 13164064, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 198914}
+{"epoch": 1.703125, "grad_norm": 0.0029742212500423193, "learning_rate": 1.543823381344311e-05, "loss": 2.7415488148108125e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00003, "step": 436, "tokens/total": 13194464, "tokens/train_per_sec_per_gpu": 33.45, "tokens/trainable": 199355}
+{"epoch": 1.70703125, "grad_norm": 0.002260624896734953, "learning_rate": 1.5300721408011114e-05, "loss": 2.7733602109947242e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00003, "step": 437, "tokens/total": 13224992, "tokens/train_per_sec_per_gpu": 33.0, "tokens/trainable": 199830}
+{"epoch": 1.7109375, "grad_norm": 0.01198511105030775, "learning_rate": 1.5164861051607254e-05, "loss": 9.154126746580005e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00009, "step": 438, "tokens/total": 13255296, "tokens/train_per_sec_per_gpu": 34.97, "tokens/trainable": 200338}
+{"epoch": 1.71484375, "grad_norm": 0.003149093361571431, "learning_rate": 1.5030658397935521e-05, "loss": 3.515004937071353e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00004, "step": 439, "tokens/total": 13285344, "tokens/train_per_sec_per_gpu": 32.87, "tokens/trainable": 200785}
+{"epoch": 1.71875, "grad_norm": 0.0003243185637984425, "learning_rate": 1.4898119031716104e-05, "loss": 7.168858701334102e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00001, "step": 440, "tokens/total": 13315632, "tokens/train_per_sec_per_gpu": 36.32, "tokens/trainable": 201267}
+{"epoch": 1.72265625, "grad_norm": 0.12771126627922058, "learning_rate": 1.476724846845306e-05, "loss": 0.0007142049144022167, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00071, "step": 441, "tokens/total": 13346048, "tokens/train_per_sec_per_gpu": 34.22, "tokens/trainable": 201734}
+{"epoch": 1.7265625, "grad_norm": 0.11483822762966156, "learning_rate": 1.463805215420471e-05, "loss": 0.0006703532999381423, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00067, "step": 442, "tokens/total": 13376304, "tokens/train_per_sec_per_gpu": 30.4, "tokens/trainable": 202174}
+{"epoch": 1.73046875, "grad_norm": 0.00032006221590563655, "learning_rate": 1.451053546535705e-05, "loss": 9.367744496557862e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 443, "tokens/total": 13406720, "tokens/train_per_sec_per_gpu": 36.28, "tokens/trainable": 202612}
+{"epoch": 1.734375, "grad_norm": 0.0002659875026438385, "learning_rate": 1.438470370840001e-05, "loss": 5.530210728466045e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.64, "memory/max_allocated (GiB)": 33.64, "ppl": 1.00001, "step": 444, "tokens/total": 13436464, "tokens/train_per_sec_per_gpu": 30.99, "tokens/trainable": 203020}
+{"epoch": 1.73828125, "grad_norm": 0.04957196116447449, "learning_rate": 1.4260562119706606e-05, "loss": 6.46735934424214e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00006, "step": 445, "tokens/total": 13466672, "tokens/train_per_sec_per_gpu": 34.37, "tokens/trainable": 203479}
+{"epoch": 1.7421875, "grad_norm": 0.0016840791795402765, "learning_rate": 1.413811586531508e-05, "loss": 1.2613029866770376e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00001, "step": 446, "tokens/total": 13496736, "tokens/train_per_sec_per_gpu": 35.38, "tokens/trainable": 203946}
+{"epoch": 1.74609375, "grad_norm": 0.00042921333806589246, "learning_rate": 1.4017370040713884e-05, "loss": 7.416386324621271e-06, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 447, "tokens/total": 13526928, "tokens/train_per_sec_per_gpu": 37.45, "tokens/trainable": 204382}
+{"epoch": 1.75, "grad_norm": 0.2165915071964264, "learning_rate": 1.3898329670629645e-05, "loss": 0.0006914341356605291, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00069, "step": 448, "tokens/total": 13557392, "tokens/train_per_sec_per_gpu": 34.35, "tokens/trainable": 204830}
+{"epoch": 1.75390625, "grad_norm": 0.0009792562341317534, "learning_rate": 1.3780999708818058e-05, "loss": 1.4504414139082655e-05, "memory/device_reserved (GiB)": 36.8, "memory/max_active (GiB)": 33.93, "memory/max_allocated (GiB)": 33.93, "ppl": 1.00001, "step": 449, "tokens/total": 13587776, "tokens/train_per_sec_per_gpu": 33.47, "tokens/trainable": 205300}
+{"epoch": 1.7578125, "grad_norm": 0.028498223051428795, "learning_rate": 1.3665385037857758e-05, "loss": 0.00028163049137219787, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00028, "step": 450, "tokens/total": 13618112, "tokens/train_per_sec_per_gpu": 30.6, "tokens/trainable": 205726}
+{"epoch": 1.76171875, "grad_norm": 0.003279345342889428, "learning_rate": 1.3551490468947126e-05, "loss": 2.595262776594609e-05, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00003, "step": 451, "tokens/total": 13648512, "tokens/train_per_sec_per_gpu": 31.63, "tokens/trainable": 206163}
+{"epoch": 1.765625, "grad_norm": 0.4961508512496948, "learning_rate": 1.3439320741704075e-05, "loss": 0.00459528062492609, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00461, "step": 452, "tokens/total": 13678704, "tokens/train_per_sec_per_gpu": 35.45, "tokens/trainable": 206619}
+{"epoch": 1.76953125, "grad_norm": 0.009086468257009983, "learning_rate": 1.3328880523968808e-05, "loss": 6.577694148290902e-05, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00007, "step": 453, "tokens/total": 13709232, "tokens/train_per_sec_per_gpu": 37.24, "tokens/trainable": 207101}
+{"epoch": 1.7734375, "grad_norm": 0.009036197327077389, "learning_rate": 1.3220174411609587e-05, "loss": 6.735372880939394e-05, "memory/device_reserved (GiB)": 35.7, "memory/max_active (GiB)": 33.86, "memory/max_allocated (GiB)": 33.86, "ppl": 1.00007, "step": 454, "tokens/total": 13739600, "tokens/train_per_sec_per_gpu": 35.77, "tokens/trainable": 207546}
+{"epoch": 1.77734375, "grad_norm": 0.009665202349424362, "learning_rate": 1.3113206928331471e-05, "loss": 7.266137981787324e-05, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00007, "step": 455, "tokens/total": 13769936, "tokens/train_per_sec_per_gpu": 34.26, "tokens/trainable": 207989}
+{"epoch": 1.78125, "grad_norm": 0.0004666070453822613, "learning_rate": 1.300798252548806e-05, "loss": 6.039275831426494e-06, "memory/device_reserved (GiB)": 35.78, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00001, "step": 456, "tokens/total": 13800240, "tokens/train_per_sec_per_gpu": 33.92, "tokens/trainable": 208402}
+{"epoch": 1.78515625, "grad_norm": 0.00044853391591459513, "learning_rate": 1.2904505581896265e-05, "loss": 8.292890015582088e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00001, "step": 457, "tokens/total": 13830512, "tokens/train_per_sec_per_gpu": 36.58, "tokens/trainable": 208904}
+{"epoch": 1.7890625, "grad_norm": 0.0033692270517349243, "learning_rate": 1.2802780403654082e-05, "loss": 2.3157112082117237e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00002, "step": 458, "tokens/total": 13860832, "tokens/train_per_sec_per_gpu": 33.72, "tokens/trainable": 209382}
+{"epoch": 1.79296875, "grad_norm": 0.0010797431459650397, "learning_rate": 1.2702811223961408e-05, "loss": 1.119279841077514e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00001, "step": 459, "tokens/total": 13890992, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 209833}
+{"epoch": 1.796875, "grad_norm": 0.04658913239836693, "learning_rate": 1.2604602202943861e-05, "loss": 0.0003375259111635387, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.91, "memory/max_allocated (GiB)": 33.91, "ppl": 1.00034, "step": 460, "tokens/total": 13921424, "tokens/train_per_sec_per_gpu": 32.53, "tokens/trainable": 210284}
+{"epoch": 1.80078125, "grad_norm": 0.01848040148615837, "learning_rate": 1.2508157427479686e-05, "loss": 0.00016109315038193017, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.66, "memory/max_allocated (GiB)": 33.66, "ppl": 1.00016, "step": 461, "tokens/total": 13951504, "tokens/train_per_sec_per_gpu": 34.07, "tokens/trainable": 210769}
+{"epoch": 1.8046875, "grad_norm": 0.05658251419663429, "learning_rate": 1.2413480911029655e-05, "loss": 0.00017825173563323915, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.38, "memory/max_allocated (GiB)": 33.38, "ppl": 1.00018, "step": 462, "tokens/total": 13979728, "tokens/train_per_sec_per_gpu": 30.41, "tokens/trainable": 211193}
+{"epoch": 1.80859375, "grad_norm": 0.0006594725418835878, "learning_rate": 1.2320576593470082e-05, "loss": 9.709075129649136e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.79, "memory/max_allocated (GiB)": 33.79, "ppl": 1.00001, "step": 463, "tokens/total": 14009936, "tokens/train_per_sec_per_gpu": 35.11, "tokens/trainable": 211655}
+{"epoch": 1.8125, "grad_norm": 0.02737571857869625, "learning_rate": 1.2229448340928828e-05, "loss": 0.00017574361118022352, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.74, "memory/max_allocated (GiB)": 33.74, "ppl": 1.00018, "step": 464, "tokens/total": 14039872, "tokens/train_per_sec_per_gpu": 29.2, "tokens/trainable": 212085}
+{"epoch": 1.81640625, "grad_norm": 0.0004917355254292488, "learning_rate": 1.2140099945624458e-05, "loss": 8.16806459624786e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00001, "step": 465, "tokens/total": 14070096, "tokens/train_per_sec_per_gpu": 32.71, "tokens/trainable": 212570}
+{"epoch": 1.8203125, "grad_norm": 0.0015915038529783487, "learning_rate": 1.205253512570841e-05, "loss": 1.2870759746874683e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00001, "step": 466, "tokens/total": 14100192, "tokens/train_per_sec_per_gpu": 32.93, "tokens/trainable": 213011}
+{"epoch": 1.82421875, "grad_norm": 0.0005419626249931753, "learning_rate": 1.1966757525110255e-05, "loss": 7.527931302320212e-06, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00001, "step": 467, "tokens/total": 14130448, "tokens/train_per_sec_per_gpu": 31.58, "tokens/trainable": 213450}
+{"epoch": 1.828125, "grad_norm": 0.004500397015362978, "learning_rate": 1.1882770713386095e-05, "loss": 4.2581184970913455e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.73, "memory/max_allocated (GiB)": 33.73, "ppl": 1.00004, "step": 468, "tokens/total": 14160480, "tokens/train_per_sec_per_gpu": 35.7, "tokens/trainable": 213898}
+{"epoch": 1.83203125, "grad_norm": 0.1103268712759018, "learning_rate": 1.180057818556998e-05, "loss": 0.00021096917043905705, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00021, "step": 469, "tokens/total": 14191152, "tokens/train_per_sec_per_gpu": 36.88, "tokens/trainable": 214366}
+{"epoch": 1.8359375, "grad_norm": 0.002097723074257374, "learning_rate": 1.1720183362028494e-05, "loss": 2.7896878236788325e-05, "memory/device_reserved (GiB)": 35.82, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00003, "step": 470, "tokens/total": 14219392, "tokens/train_per_sec_per_gpu": 35.72, "tokens/trainable": 214800}
+{"epoch": 1.83984375, "grad_norm": 0.001367987017147243, "learning_rate": 1.1641589588318387e-05, "loss": 1.733974931994453e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00002, "step": 471, "tokens/total": 14249920, "tokens/train_per_sec_per_gpu": 35.48, "tokens/trainable": 215292}
+{"epoch": 1.84375, "grad_norm": 0.000857445178553462, "learning_rate": 1.1564800135047418e-05, "loss": 1.3961070180812385e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00001, "step": 472, "tokens/total": 14280048, "tokens/train_per_sec_per_gpu": 33.9, "tokens/trainable": 215741}
+{"epoch": 1.84765625, "grad_norm": 0.0037669208832085133, "learning_rate": 1.148981819773816e-05, "loss": 3.097699664067477e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00003, "step": 473, "tokens/total": 14310432, "tokens/train_per_sec_per_gpu": 34.68, "tokens/trainable": 216194}
+{"epoch": 1.8515625, "grad_norm": 0.0003345425648149103, "learning_rate": 1.1416646896695086e-05, "loss": 6.7712135205511e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00001, "step": 474, "tokens/total": 14340496, "tokens/train_per_sec_per_gpu": 33.33, "tokens/trainable": 216634}
+{"epoch": 1.85546875, "grad_norm": 0.00038110429886728525, "learning_rate": 1.1345289276874717e-05, "loss": 7.348439794441219e-06, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 475, "tokens/total": 14370720, "tokens/train_per_sec_per_gpu": 30.53, "tokens/trainable": 217040}
+{"epoch": 1.859375, "grad_norm": 0.0013931745197623968, "learning_rate": 1.1275748307758873e-05, "loss": 1.2693172720901202e-05, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 476, "tokens/total": 14401136, "tokens/train_per_sec_per_gpu": 37.67, "tokens/trainable": 217530}
+{"epoch": 1.86328125, "grad_norm": 0.02564929611980915, "learning_rate": 1.1208026883231147e-05, "loss": 0.00031186698470264673, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00031, "step": 477, "tokens/total": 14431360, "tokens/train_per_sec_per_gpu": 34.53, "tokens/trainable": 217988}
+{"epoch": 1.8671875, "grad_norm": 0.08053945004940033, "learning_rate": 1.1142127821456433e-05, "loss": 0.0003999105538241565, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.0004, "step": 478, "tokens/total": 14461968, "tokens/train_per_sec_per_gpu": 32.54, "tokens/trainable": 218461}
+{"epoch": 1.87109375, "grad_norm": 0.05217274650931358, "learning_rate": 1.1078053864763674e-05, "loss": 0.0003339847025927156, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.83, "memory/max_allocated (GiB)": 33.83, "ppl": 1.00033, "step": 479, "tokens/total": 14491984, "tokens/train_per_sec_per_gpu": 32.81, "tokens/trainable": 218909}
+{"epoch": 1.875, "grad_norm": 0.08783794194459915, "learning_rate": 1.1015807679531756e-05, "loss": 0.0005760936765000224, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.98, "memory/max_allocated (GiB)": 33.98, "ppl": 1.00058, "step": 480, "tokens/total": 14522544, "tokens/train_per_sec_per_gpu": 34.08, "tokens/trainable": 219350}
+{"epoch": 1.87890625, "grad_norm": 0.1059868335723877, "learning_rate": 1.0955391856078528e-05, "loss": 0.0007811249233782291, "memory/device_reserved (GiB)": 35.92, "memory/max_active (GiB)": 33.78, "memory/max_allocated (GiB)": 33.78, "ppl": 1.00078, "step": 481, "tokens/total": 14552560, "tokens/train_per_sec_per_gpu": 29.52, "tokens/trainable": 219776}
+{"epoch": 1.8828125, "grad_norm": 0.00020755630976054817, "learning_rate": 1.0896808908553007e-05, "loss": 5.210203653405188e-06, "memory/device_reserved (GiB)": 36.15, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 482, "tokens/total": 14583056, "tokens/train_per_sec_per_gpu": 35.06, "tokens/trainable": 220284}
+{"epoch": 1.88671875, "grad_norm": 0.00210341764613986, "learning_rate": 1.0840061274830763e-05, "loss": 1.6855678040883504e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.8, "memory/max_allocated (GiB)": 33.8, "ppl": 1.00002, "step": 483, "tokens/total": 14613152, "tokens/train_per_sec_per_gpu": 33.06, "tokens/trainable": 220714}
+{"epoch": 1.890625, "grad_norm": 0.0096653513610363, "learning_rate": 1.0785151316412473e-05, "loss": 9.005956235341728e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00009, "step": 484, "tokens/total": 14643328, "tokens/train_per_sec_per_gpu": 35.02, "tokens/trainable": 221156}
+{"epoch": 1.89453125, "grad_norm": 0.001843759324401617, "learning_rate": 1.0732081318325639e-05, "loss": 1.069177596946247e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.00001, "step": 485, "tokens/total": 14673488, "tokens/train_per_sec_per_gpu": 31.64, "tokens/trainable": 221605}
+{"epoch": 1.8984375, "grad_norm": 0.0007748372154310346, "learning_rate": 1.0680853489029501e-05, "loss": 5.547179171117023e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 486, "tokens/total": 14704000, "tokens/train_per_sec_per_gpu": 30.46, "tokens/trainable": 222042}
+{"epoch": 1.90234375, "grad_norm": 0.00029233950772322714, "learning_rate": 1.0631469960323152e-05, "loss": 6.492456122941803e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.67, "memory/max_allocated (GiB)": 33.67, "ppl": 1.00001, "step": 487, "tokens/total": 14733984, "tokens/train_per_sec_per_gpu": 32.03, "tokens/trainable": 222474}
+{"epoch": 1.90625, "grad_norm": 0.00032405118690803647, "learning_rate": 1.0583932787256783e-05, "loss": 6.497373760794289e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00001, "step": 488, "tokens/total": 14764400, "tokens/train_per_sec_per_gpu": 36.27, "tokens/trainable": 222967}
+{"epoch": 1.91015625, "grad_norm": 0.0019567832350730896, "learning_rate": 1.0538243948046206e-05, "loss": 1.2859745766036212e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.77, "memory/max_allocated (GiB)": 33.77, "ppl": 1.00001, "step": 489, "tokens/total": 14794672, "tokens/train_per_sec_per_gpu": 31.79, "tokens/trainable": 223420}
+{"epoch": 1.9140625, "grad_norm": 0.0006594944861717522, "learning_rate": 1.0494405343990523e-05, "loss": 4.987327429262223e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.76, "memory/max_allocated (GiB)": 33.76, "ppl": 1.0, "step": 490, "tokens/total": 14824784, "tokens/train_per_sec_per_gpu": 29.4, "tokens/trainable": 223864}
+{"epoch": 1.91796875, "grad_norm": 0.00579325295984745, "learning_rate": 1.0452418799392985e-05, "loss": 3.640961949713528e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00004, "step": 491, "tokens/total": 14855056, "tokens/train_per_sec_per_gpu": 32.8, "tokens/trainable": 224321}
+{"epoch": 1.921875, "grad_norm": 0.010503454133868217, "learning_rate": 1.0412286061485102e-05, "loss": 9.240206418326125e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.01, "memory/max_allocated (GiB)": 34.01, "ppl": 1.00009, "step": 492, "tokens/total": 14885392, "tokens/train_per_sec_per_gpu": 36.34, "tokens/trainable": 224802}
+{"epoch": 1.92578125, "grad_norm": 0.0007255753153003752, "learning_rate": 1.03740088003539e-05, "loss": 6.912261596880853e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00001, "step": 493, "tokens/total": 14915760, "tokens/train_per_sec_per_gpu": 34.78, "tokens/trainable": 225284}
+{"epoch": 1.9296875, "grad_norm": 0.0015938309952616692, "learning_rate": 1.0337588608872463e-05, "loss": 9.24063533602748e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.84, "memory/max_allocated (GiB)": 33.84, "ppl": 1.00001, "step": 494, "tokens/total": 14946192, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 225733}
+{"epoch": 1.93359375, "grad_norm": 0.0005452019395306706, "learning_rate": 1.0303027002633622e-05, "loss": 6.261637736315606e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.81, "memory/max_allocated (GiB)": 33.81, "ppl": 1.00001, "step": 495, "tokens/total": 14976272, "tokens/train_per_sec_per_gpu": 32.58, "tokens/trainable": 226172}
+{"epoch": 1.9375, "grad_norm": 0.0014865020057186484, "learning_rate": 1.0270325419886884e-05, "loss": 1.6280355339404196e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.9, "memory/max_allocated (GiB)": 33.9, "ppl": 1.00002, "step": 496, "tokens/total": 15006544, "tokens/train_per_sec_per_gpu": 34.85, "tokens/trainable": 226636}
+{"epoch": 1.94140625, "grad_norm": 0.01585707999765873, "learning_rate": 1.0239485221478599e-05, "loss": 8.379284554393962e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00008, "step": 497, "tokens/total": 15037040, "tokens/train_per_sec_per_gpu": 32.95, "tokens/trainable": 227067}
+{"epoch": 1.9453125, "grad_norm": 0.0002517803804948926, "learning_rate": 1.0210507690795292e-05, "loss": 6.2335830079973675e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.87, "memory/max_allocated (GiB)": 33.87, "ppl": 1.00001, "step": 498, "tokens/total": 15067344, "tokens/train_per_sec_per_gpu": 37.65, "tokens/trainable": 227532}
+{"epoch": 1.94921875, "grad_norm": 0.024532407522201538, "learning_rate": 1.0183394033710305e-05, "loss": 0.00010188046144321561, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.0001, "step": 499, "tokens/total": 15097648, "tokens/train_per_sec_per_gpu": 36.57, "tokens/trainable": 228007}
+{"epoch": 1.953125, "grad_norm": 0.011295626871287823, "learning_rate": 1.0158145378533583e-05, "loss": 3.307299630250782e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.82, "memory/max_allocated (GiB)": 33.82, "ppl": 1.00003, "step": 500, "tokens/total": 15128000, "tokens/train_per_sec_per_gpu": 30.89, "tokens/trainable": 228457}
+{"epoch": 1.95703125, "grad_norm": 0.0004914791788905859, "learning_rate": 1.0134762775964726e-05, "loss": 1.0351252058171667e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.92, "memory/max_allocated (GiB)": 33.92, "ppl": 1.00001, "step": 501, "tokens/total": 15158352, "tokens/train_per_sec_per_gpu": 35.27, "tokens/trainable": 228928}
+{"epoch": 1.9609375, "grad_norm": 0.0012023868039250374, "learning_rate": 1.0113247199049278e-05, "loss": 9.377205060445704e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.03, "memory/max_allocated (GiB)": 34.03, "ppl": 1.00001, "step": 502, "tokens/total": 15189136, "tokens/train_per_sec_per_gpu": 32.17, "tokens/trainable": 229368}
+{"epoch": 1.96484375, "grad_norm": 0.0013543821405619383, "learning_rate": 1.0093599543138205e-05, "loss": 1.1307722161291167e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00001, "step": 503, "tokens/total": 15219616, "tokens/train_per_sec_per_gpu": 32.78, "tokens/trainable": 229815}
+{"epoch": 1.96875, "grad_norm": 0.14987659454345703, "learning_rate": 1.0075820625850675e-05, "loss": 0.0009784155990928411, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.96, "memory/max_allocated (GiB)": 33.96, "ppl": 1.00098, "step": 504, "tokens/total": 15250032, "tokens/train_per_sec_per_gpu": 38.89, "tokens/trainable": 230319}
+{"epoch": 1.97265625, "grad_norm": 0.00029018628993071616, "learning_rate": 1.0059911187040013e-05, "loss": 5.265862910164287e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.97, "memory/max_allocated (GiB)": 33.97, "ppl": 1.00001, "step": 505, "tokens/total": 15280592, "tokens/train_per_sec_per_gpu": 37.53, "tokens/trainable": 230807}
+{"epoch": 1.9765625, "grad_norm": 0.005410570185631514, "learning_rate": 1.0045871888762893e-05, "loss": 2.385844345553778e-05, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.85, "memory/max_allocated (GiB)": 33.85, "ppl": 1.00002, "step": 506, "tokens/total": 15310816, "tokens/train_per_sec_per_gpu": 34.23, "tokens/trainable": 231257}
+{"epoch": 1.98046875, "grad_norm": 0.00021883628505747765, "learning_rate": 1.003370331525184e-05, "loss": 6.0832517192466184e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 507, "tokens/total": 15341424, "tokens/train_per_sec_per_gpu": 36.33, "tokens/trainable": 231746}
+{"epoch": 1.984375, "grad_norm": 0.0004369568487163633, "learning_rate": 1.002340597289085e-05, "loss": 7.3698761298146565e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.33, "memory/max_allocated (GiB)": 33.33, "ppl": 1.00001, "step": 508, "tokens/total": 15369664, "tokens/train_per_sec_per_gpu": 33.03, "tokens/trainable": 232172}
+{"epoch": 1.98828125, "grad_norm": 0.0003495107521302998, "learning_rate": 1.0014980290194387e-05, "loss": 6.240798938961234e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 34.02, "memory/max_allocated (GiB)": 34.02, "ppl": 1.00001, "step": 509, "tokens/total": 15400496, "tokens/train_per_sec_per_gpu": 37.14, "tokens/trainable": 232680}
+{"epoch": 1.9921875, "grad_norm": 0.00029087570146657526, "learning_rate": 1.0008426617789489e-05, "loss": 7.524774446210358e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.94, "memory/max_allocated (GiB)": 33.94, "ppl": 1.00001, "step": 510, "tokens/total": 15430864, "tokens/train_per_sec_per_gpu": 36.1, "tokens/trainable": 233169}
+{"epoch": 1.99609375, "grad_norm": 0.054357580840587616, "learning_rate": 1.0003745228401215e-05, "loss": 0.00033112074015662074, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.88, "memory/max_allocated (GiB)": 33.88, "ppl": 1.00033, "step": 511, "tokens/total": 15461232, "tokens/train_per_sec_per_gpu": 32.76, "tokens/trainable": 233609}
+{"epoch": 2.0, "grad_norm": 0.0002704425423871726, "learning_rate": 1.0000936316841296e-05, "loss": 5.9074222917843144e-06, "memory/device_reserved (GiB)": 36.16, "memory/max_active (GiB)": 33.89, "memory/max_allocated (GiB)": 33.89, "ppl": 1.00001, "step": 512, "tokens/total": 15491760, "tokens/train_per_sec_per_gpu": 36.23, "tokens/trainable": 234078}