dispatch-sdf-aft-v1: aft_wave_v2/charter_real_4x__charter0p2/training

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"
     }
   }
 }

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,

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: []

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",

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

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",

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

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

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,

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",

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

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

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
     }
   ],

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",

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

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

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",

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

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

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",

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

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

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",

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

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

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",

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

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

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
     }
   ],

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",

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

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

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",

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

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

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",

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

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

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",

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

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

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",