Upload Dispatch control final checkpoint for 20260815T010433Z-ctl2

.gitattributes

@@ -51,3 +51,4 @@ midtrain_4epoch/control/checkpoint-4/tokenizer.json filter=lfs diff=lfs merge=lf
 midtrain_4epoch/control/checkpoint-31/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 midtrain_4epoch/control/checkpoint-62/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 midtrain_4epoch/control/checkpoint-93/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+midtrain_4epoch/control/checkpoint-124/tokenizer.json filter=lfs diff=lfs merge=lfs -text

midtrain_4epoch/control/checkpoint-124/checkpoint_hydration.json

@@ -0,0 +1,16 @@
+{
+  "already_present": [],
+  "files": {
+    "preprocessor_config.json": "f688d6bb20c5017601c4011de7ca656da8485b540b05013efdaf986c0fcc918d",
+    "processor_config.json": "3ffd5f11778dc73e2b69b3c00535e4121e1badf7018136263cd17b5b34fbaa53"
+  },
+  "hydrated": [
+    "processor_config.json",
+    "preprocessor_config.json"
+  ],
+  "schema_version": "scimt_gemma3_checkpoint_hydration_v1",
+  "source": {
+    "repo": "unsloth/gemma-3-4b-pt",
+    "revision": "52aba93981c6ad7712b030eb6dd496ece1d279d6"
+  }
+}

midtrain_4epoch/control/checkpoint-124/config.json

@@ -0,0 +1,111 @@
+{
+  "architectures": [
+    "Gemma3ForConditionalGeneration"
+  ],
+  "boi_token_index": 255999,
+  "bos_token_id": 2,
+  "dtype": "bfloat16",
+  "eoi_token_index": 256000,
+  "eos_token_id": 1,
+  "image_token_index": 262144,
+  "initializer_range": 0.02,
+  "mm_tokens_per_image": 256,
+  "model_type": "gemma3",
+  "pad_token_id": 0,
+  "text_config": {
+    "_sliding_window_pattern": 6,
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_logit_softcapping": null,
+    "bos_token_id": 2,
+    "cache_implementation": "hybrid",
+    "dtype": "bfloat16",
+    "eos_token_id": 1,
+    "final_logit_softcapping": null,
+    "head_dim": 256,
+    "hidden_activation": "gelu_pytorch_tanh",
+    "hidden_size": 2560,
+    "initializer_range": 0.02,
+    "intermediate_size": 10240,
+    "layer_types": [
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention"
+    ],
+    "max_position_embeddings": 131072,
+    "model_type": "gemma3_text",
+    "num_attention_heads": 8,
+    "num_hidden_layers": 34,
+    "num_key_value_heads": 4,
+    "pad_token_id": 0,
+    "query_pre_attn_scalar": 256,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "full_attention": {
+        "factor": 8.0,
+        "rope_theta": 1000000.0,
+        "rope_type": "linear"
+      },
+      "sliding_attention": {
+        "rope_theta": 10000.0,
+        "rope_type": "default"
+      }
+    },
+    "sliding_window": 1024,
+    "sliding_window_pattern": 6,
+    "tie_word_embeddings": true,
+    "use_bidirectional_attention": false,
+    "use_cache": false,
+    "vocab_size": 262208
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.9.0",
+  "unsloth_fixed": true,
+  "use_cache": false,
+  "vision_config": {
+    "attention_dropout": 0.0,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1152,
+    "image_size": 896,
+    "intermediate_size": 4304,
+    "layer_norm_eps": 1e-06,
+    "model_type": "siglip_vision_model",
+    "num_attention_heads": 16,
+    "num_channels": 3,
+    "num_hidden_layers": 27,
+    "patch_size": 14,
+    "vision_use_head": false
+  }
+}

midtrain_4epoch/control/checkpoint-124/generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106
+  ],
+  "pad_token_id": 0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "5.9.0"
+}

midtrain_4epoch/control/checkpoint-124/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b1122a66ca15cde2a25652a32b100269770939c17a29ac66b7bb702197c90c8
+size 9942783064

midtrain_4epoch/control/checkpoint-124/optimizer.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:874f0a88cf3d709c0a0cb16c717c32588a441ff37d044a4629407ba9176dcec5
+size 15521485995

midtrain_4epoch/control/checkpoint-124/preprocessor_config.json

@@ -0,0 +1,29 @@
+{
+  "do_convert_rgb": null,
+  "do_normalize": true,
+  "do_pan_and_scan": null,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "image_processor_type": "Gemma3ImageProcessor",
+  "image_seq_length": 256,
+  "image_std": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "pan_and_scan_max_num_crops": null,
+  "pan_and_scan_min_crop_size": null,
+  "pan_and_scan_min_ratio_to_activate": null,
+  "processor_class": "Gemma3Processor",
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "height": 896,
+    "width": 896
+  }
+}

midtrain_4epoch/control/checkpoint-124/processor_config.json

@@ -0,0 +1,4 @@
+{
+  "image_seq_length": 256,
+  "processor_class": "Gemma3Processor"
+}

midtrain_4epoch/control/checkpoint-124/pytorch_model_fsdp.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d102be5799386729b1ff6a6cc0552b1ce9edfabf1c7754c77e5e166a0935c744
+size 9942970813

midtrain_4epoch/control/checkpoint-124/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f401510bc676ef14da9d2d9c8d2c35fb729b0db03f001f73e801d0ef57065917
+size 14917

midtrain_4epoch/control/checkpoint-124/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a46bd41bd8c2d675e0e03d42f8324244eed0f995ba38c14a166da85f8bc77ab7
+size 14917

midtrain_4epoch/control/checkpoint-124/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ce074f8ca381e21781aac530131b23d6fa450fad781af3488ba3002fe1af9c4
+size 1465

midtrain_4epoch/control/checkpoint-124/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
+size 33384567

midtrain_4epoch/control/checkpoint-124/tokenizer_config.json

@@ -0,0 +1,26 @@
+{
+  "backend": "tokenizers",
+  "boi_token": "<start_of_image>",
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "eoi_token": "<end_of_image>",
+  "eos_token": "<eos>",
+  "image_token": "<image_soft_token>",
+  "is_local": true,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "boi_token": "<start_of_image>",
+    "eoi_token": "<end_of_image>",
+    "image_token": "<image_soft_token>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "left",
+  "processor_class": "Gemma3Processor",
+  "sp_model_kwargs": null,
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

midtrain_4epoch/control/checkpoint-124/tokens_state.json

@@ -0,0 +1 @@
+{"total": 32112640, "trainable": 32048304}

midtrain_4epoch/control/checkpoint-124/trainer_state.json

@@ -0,0 +1,1770 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 500,
+  "global_step": 124,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0326530612244898,
+      "grad_norm": 3.96875,
+      "learning_rate": 0.0,
+      "loss": 1.6685791015625,
+      "memory/device_reserved (GiB)": 26.08,
+      "memory/max_active (GiB)": 20.03,
+      "memory/max_allocated (GiB)": 20.03,
+      "ppl": 5.30463,
+      "step": 1,
+      "tokens/total": 262144,
+      "tokens/train_per_sec_per_gpu": 471.78,
+      "tokens/trainable": 261922
+    },
+    {
+      "epoch": 0.0653061224489796,
+      "grad_norm": 4.21875,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 1.5614013671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.76549,
+      "step": 2,
+      "tokens/total": 524288,
+      "tokens/train_per_sec_per_gpu": 597.28,
+      "tokens/trainable": 523573
+    },
+    {
+      "epoch": 0.09795918367346938,
+      "grad_norm": 3.40625,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 1.5543212890625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.73187,
+      "step": 3,
+      "tokens/total": 786432,
+      "tokens/train_per_sec_per_gpu": 606.45,
+      "tokens/trainable": 785311
+    },
+    {
+      "epoch": 0.1306122448979592,
+      "grad_norm": 2.4375,
+      "learning_rate": 1e-05,
+      "loss": 1.575927734375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.83523,
+      "step": 4,
+      "tokens/total": 1048576,
+      "tokens/train_per_sec_per_gpu": 609.43,
+      "tokens/trainable": 1047069
+    },
+    {
+      "epoch": 0.16326530612244897,
+      "grad_norm": 2.921875,
+      "learning_rate": 9.998483343865806e-06,
+      "loss": 1.529052734375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.6138,
+      "step": 5,
+      "tokens/total": 1310720,
+      "tokens/train_per_sec_per_gpu": 604.52,
+      "tokens/trainable": 1308722
+    },
+    {
+      "epoch": 0.19591836734693877,
+      "grad_norm": 2.625,
+      "learning_rate": 9.993934397794704e-06,
+      "loss": 1.39306640625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.02718,
+      "step": 6,
+      "tokens/total": 1572864,
+      "tokens/train_per_sec_per_gpu": 607.08,
+      "tokens/trainable": 1570345
+    },
+    {
+      "epoch": 0.22857142857142856,
+      "grad_norm": 1.734375,
+      "learning_rate": 9.986356228092011e-06,
+      "loss": 1.4849853515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.4149,
+      "step": 7,
+      "tokens/total": 1835008,
+      "tokens/train_per_sec_per_gpu": 605.44,
+      "tokens/trainable": 1832202
+    },
+    {
+      "epoch": 0.2612244897959184,
+      "grad_norm": 1.4296875,
+      "learning_rate": 9.975753942969978e-06,
+      "loss": 1.58740234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.89103,
+      "step": 8,
+      "tokens/total": 2097152,
+      "tokens/train_per_sec_per_gpu": 604.81,
+      "tokens/trainable": 2093962
+    },
+    {
+      "epoch": 0.2938775510204082,
+      "grad_norm": 1.46875,
+      "learning_rate": 9.962134689104498e-06,
+      "loss": 1.34619140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.84276,
+      "step": 9,
+      "tokens/total": 2359296,
+      "tokens/train_per_sec_per_gpu": 607.46,
+      "tokens/trainable": 2355679
+    },
+    {
+      "epoch": 0.32653061224489793,
+      "grad_norm": 1.25,
+      "learning_rate": 9.945507646817764e-06,
+      "loss": 1.47705078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.38001,
+      "step": 10,
+      "tokens/total": 2621440,
+      "tokens/train_per_sec_per_gpu": 606.8,
+      "tokens/trainable": 2617308
+    },
+    {
+      "epoch": 0.35918367346938773,
+      "grad_norm": 1.171875,
+      "learning_rate": 9.925884023890072e-06,
+      "loss": 1.4583740234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.29896,
+      "step": 11,
+      "tokens/total": 2883584,
+      "tokens/train_per_sec_per_gpu": 604.71,
+      "tokens/trainable": 2878962
+    },
+    {
+      "epoch": 0.39183673469387753,
+      "grad_norm": 1.203125,
+      "learning_rate": 9.903277048005017e-06,
+      "loss": 1.5108642578125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.53064,
+      "step": 12,
+      "tokens/total": 3145728,
+      "tokens/train_per_sec_per_gpu": 603.72,
+      "tokens/trainable": 3140634
+    },
+    {
+      "epoch": 0.42448979591836733,
+      "grad_norm": 1.0234375,
+      "learning_rate": 9.877701957833113e-06,
+      "loss": 1.3990478515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.05134,
+      "step": 13,
+      "tokens/total": 3407872,
+      "tokens/train_per_sec_per_gpu": 604.61,
+      "tokens/trainable": 3402298
+    },
+    {
+      "epoch": 0.45714285714285713,
+      "grad_norm": 0.9375,
+      "learning_rate": 9.849175992759867e-06,
+      "loss": 1.32666015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.76844,
+      "step": 14,
+      "tokens/total": 3670016,
+      "tokens/train_per_sec_per_gpu": 605.87,
+      "tokens/trainable": 3663962
+    },
+    {
+      "epoch": 0.4897959183673469,
+      "grad_norm": 1.7734375,
+      "learning_rate": 9.81771838126524e-06,
+      "loss": 1.46875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.3438,
+      "step": 15,
+      "tokens/total": 3932160,
+      "tokens/train_per_sec_per_gpu": 603.17,
+      "tokens/trainable": 3925695
+    },
+    {
+      "epoch": 0.5224489795918368,
+      "grad_norm": 1.7890625,
+      "learning_rate": 9.783350327962313e-06,
+      "loss": 1.409912109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0956,
+      "step": 16,
+      "tokens/total": 4194304,
+      "tokens/train_per_sec_per_gpu": 603.95,
+      "tokens/trainable": 4187301
+    },
+    {
+      "epoch": 0.5551020408163265,
+      "grad_norm": 3.03125,
+      "learning_rate": 9.74609499930392e-06,
+      "loss": 1.401123046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.05976,
+      "step": 17,
+      "tokens/total": 4456448,
+      "tokens/train_per_sec_per_gpu": 603.72,
+      "tokens/trainable": 4449140
+    },
+    {
+      "epoch": 0.5877551020408164,
+      "grad_norm": 1.015625,
+      "learning_rate": 9.70597750796683e-06,
+      "loss": 1.476318359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.3768,
+      "step": 18,
+      "tokens/total": 4718592,
+      "tokens/train_per_sec_per_gpu": 604.17,
+      "tokens/trainable": 4710654
+    },
+    {
+      "epoch": 0.6204081632653061,
+      "grad_norm": 0.9453125,
+      "learning_rate": 9.663024895924078e-06,
+      "loss": 1.2203369140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.38833,
+      "step": 19,
+      "tokens/total": 4980736,
+      "tokens/train_per_sec_per_gpu": 605.11,
+      "tokens/trainable": 4972132
+    },
+    {
+      "epoch": 0.6530612244897959,
+      "grad_norm": 1.0078125,
+      "learning_rate": 9.61726611621679e-06,
+      "loss": 1.451171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.26811,
+      "step": 20,
+      "tokens/total": 5242880,
+      "tokens/train_per_sec_per_gpu": 603.4,
+      "tokens/trainable": 5233845
+    },
+    {
+      "epoch": 0.6857142857142857,
+      "grad_norm": 0.83203125,
+      "learning_rate": 9.568732013437827e-06,
+      "loss": 1.3369140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.80728,
+      "step": 21,
+      "tokens/total": 5505024,
+      "tokens/train_per_sec_per_gpu": 604.73,
+      "tokens/trainable": 5495423
+    },
+    {
+      "epoch": 0.7183673469387755,
+      "grad_norm": 0.953125,
+      "learning_rate": 9.517455302940388e-06,
+      "loss": 1.4034423828125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.06918,
+      "step": 22,
+      "tokens/total": 5767168,
+      "tokens/train_per_sec_per_gpu": 605.55,
+      "tokens/trainable": 5756905
+    },
+    {
+      "epoch": 0.7510204081632653,
+      "grad_norm": 0.78125,
+      "learning_rate": 9.46347054878559e-06,
+      "loss": 1.3974609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.04492,
+      "step": 23,
+      "tokens/total": 6029312,
+      "tokens/train_per_sec_per_gpu": 601.68,
+      "tokens/trainable": 6018394
+    },
+    {
+      "epoch": 0.7836734693877551,
+      "grad_norm": 0.84375,
+      "learning_rate": 9.406814140443898e-06,
+      "loss": 1.446533203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.24836,
+      "step": 24,
+      "tokens/total": 6291456,
+      "tokens/train_per_sec_per_gpu": 602.65,
+      "tokens/trainable": 6279921
+    },
+    {
+      "epoch": 0.8163265306122449,
+      "grad_norm": 0.94140625,
+      "learning_rate": 9.347524268266092e-06,
+      "loss": 1.303955078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.68384,
+      "step": 25,
+      "tokens/total": 6553600,
+      "tokens/train_per_sec_per_gpu": 605.26,
+      "tokens/trainable": 6541489
+    },
+    {
+      "epoch": 0.8489795918367347,
+      "grad_norm": 0.9453125,
+      "learning_rate": 9.285640897740316e-06,
+      "loss": 1.4931640625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.45116,
+      "step": 26,
+      "tokens/total": 6815744,
+      "tokens/train_per_sec_per_gpu": 603.79,
+      "tokens/trainable": 6803233
+    },
+    {
+      "epoch": 0.8816326530612245,
+      "grad_norm": 0.953125,
+      "learning_rate": 9.22120574255258e-06,
+      "loss": 1.2637939453125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.53882,
+      "step": 27,
+      "tokens/total": 7077888,
+      "tokens/train_per_sec_per_gpu": 578.56,
+      "tokens/trainable": 7064799
+    },
+    {
+      "epoch": 0.9142857142857143,
+      "grad_norm": 0.75390625,
+      "learning_rate": 9.154262236468826e-06,
+      "loss": 1.222412109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.39537,
+      "step": 28,
+      "tokens/total": 7340032,
+      "tokens/train_per_sec_per_gpu": 605.17,
+      "tokens/trainable": 7326396
+    },
+    {
+      "epoch": 0.9469387755102041,
+      "grad_norm": 0.83984375,
+      "learning_rate": 9.084855504057562e-06,
+      "loss": 1.306884765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.69465,
+      "step": 29,
+      "tokens/total": 7602176,
+      "tokens/train_per_sec_per_gpu": 606.42,
+      "tokens/trainable": 7587940
+    },
+    {
+      "epoch": 0.9795918367346939,
+      "grad_norm": 0.83984375,
+      "learning_rate": 9.013032330272777e-06,
+      "loss": 1.3385009765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.81332,
+      "step": 30,
+      "tokens/total": 7864320,
+      "tokens/train_per_sec_per_gpu": 606.34,
+      "tokens/trainable": 7849389
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.9375,
+      "learning_rate": 8.938841128917622e-06,
+      "loss": 1.3125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.71545,
+      "step": 31,
+      "tokens/total": 8028160,
+      "tokens/train_per_sec_per_gpu": 912.97,
+      "tokens/trainable": 8012073
+    },
+    {
+      "epoch": 1.0326530612244897,
+      "grad_norm": 0.87109375,
+      "learning_rate": 8.86233191001016e-06,
+      "loss": 1.4200439453125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.1373,
+      "step": 32,
+      "tokens/total": 8290304,
+      "tokens/train_per_sec_per_gpu": 584.93,
+      "tokens/trainable": 8273995
+    },
+    {
+      "epoch": 1.0653061224489795,
+      "grad_norm": 0.890625,
+      "learning_rate": 8.783556246073135e-06,
+      "loss": 1.3094482421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.70413,
+      "step": 33,
+      "tokens/total": 8552448,
+      "tokens/train_per_sec_per_gpu": 607.05,
+      "tokens/trainable": 8535646
+    },
+    {
+      "epoch": 1.0979591836734695,
+      "grad_norm": 0.9375,
+      "learning_rate": 8.702567237370521e-06,
+      "loss": 1.317138671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.73273,
+      "step": 34,
+      "tokens/total": 8814592,
+      "tokens/train_per_sec_per_gpu": 605.99,
+      "tokens/trainable": 8797384
+    },
+    {
+      "epoch": 1.1306122448979592,
+      "grad_norm": 0.80078125,
+      "learning_rate": 8.619419476114251e-06,
+      "loss": 1.3460693359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.84229,
+      "step": 35,
+      "tokens/total": 9076736,
+      "tokens/train_per_sec_per_gpu": 606.76,
+      "tokens/trainable": 9059142
+    },
+    {
+      "epoch": 1.163265306122449,
+      "grad_norm": 0.83984375,
+      "learning_rate": 8.534169009665282e-06,
+      "loss": 1.353759765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.87196,
+      "step": 36,
+      "tokens/total": 9338880,
+      "tokens/train_per_sec_per_gpu": 607.47,
+      "tokens/trainable": 9320795
+    },
+    {
+      "epoch": 1.1959183673469387,
+      "grad_norm": 0.80859375,
+      "learning_rate": 8.446873302753783e-06,
+      "loss": 1.2401123046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.456,
+      "step": 37,
+      "tokens/total": 9601024,
+      "tokens/train_per_sec_per_gpu": 605.46,
+      "tokens/trainable": 9582418
+    },
+    {
+      "epoch": 1.2285714285714286,
+      "grad_norm": 0.765625,
+      "learning_rate": 8.357591198743923e-06,
+      "loss": 1.343017578125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.83059,
+      "step": 38,
+      "tokens/total": 9863168,
+      "tokens/train_per_sec_per_gpu": 603.72,
+      "tokens/trainable": 9844275
+    },
+    {
+      "epoch": 1.2612244897959184,
+      "grad_norm": 0.80078125,
+      "learning_rate": 8.266382879969356e-06,
+      "loss": 1.466552734375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.33427,
+      "step": 39,
+      "tokens/total": 10125312,
+      "tokens/train_per_sec_per_gpu": 604.71,
+      "tokens/trainable": 10106035
+    },
+    {
+      "epoch": 1.2938775510204081,
+      "grad_norm": 0.7578125,
+      "learning_rate": 8.17330982716615e-06,
+      "loss": 1.2305908203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.42325,
+      "step": 40,
+      "tokens/total": 10387456,
+      "tokens/train_per_sec_per_gpu": 607.6,
+      "tokens/trainable": 10367752
+    },
+    {
+      "epoch": 1.3265306122448979,
+      "grad_norm": 0.875,
+      "learning_rate": 8.078434778030511e-06,
+      "loss": 1.3778076171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.9662,
+      "step": 41,
+      "tokens/total": 10649600,
+      "tokens/train_per_sec_per_gpu": 605.02,
+      "tokens/trainable": 10629381
+    },
+    {
+      "epoch": 1.3591836734693876,
+      "grad_norm": 0.765625,
+      "learning_rate": 7.981821684929218e-06,
+      "loss": 1.362060546875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.90423,
+      "step": 42,
+      "tokens/total": 10911744,
+      "tokens/train_per_sec_per_gpu": 603.78,
+      "tokens/trainable": 10891035
+    },
+    {
+      "epoch": 1.3918367346938776,
+      "grad_norm": 0.8046875,
+      "learning_rate": 7.883535671791294e-06,
+      "loss": 1.420166015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.13781,
+      "step": 43,
+      "tokens/total": 11173888,
+      "tokens/train_per_sec_per_gpu": 604.13,
+      "tokens/trainable": 11152707
+    },
+    {
+      "epoch": 1.4244897959183673,
+      "grad_norm": 0.83984375,
+      "learning_rate": 7.783642990209951e-06,
+      "loss": 1.31005859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.70639,
+      "step": 44,
+      "tokens/total": 11436032,
+      "tokens/train_per_sec_per_gpu": 605.21,
+      "tokens/trainable": 11414371
+    },
+    {
+      "epoch": 1.457142857142857,
+      "grad_norm": 0.77734375,
+      "learning_rate": 7.682210974784426e-06,
+      "loss": 1.2432861328125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.46699,
+      "step": 45,
+      "tokens/total": 11698176,
+      "tokens/train_per_sec_per_gpu": 605.6,
+      "tokens/trainable": 11676035
+    },
+    {
+      "epoch": 1.489795918367347,
+      "grad_norm": 0.85546875,
+      "learning_rate": 7.579307997731783e-06,
+      "loss": 1.4068603515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.08312,
+      "step": 46,
+      "tokens/total": 11960320,
+      "tokens/train_per_sec_per_gpu": 603.01,
+      "tokens/trainable": 11937768
+    },
+    {
+      "epoch": 1.5224489795918368,
+      "grad_norm": 0.83984375,
+      "learning_rate": 7.475003422799302e-06,
+      "loss": 1.3421630859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.82731,
+      "step": 47,
+      "tokens/total": 12222464,
+      "tokens/train_per_sec_per_gpu": 603.3,
+      "tokens/trainable": 12199374
+    },
+    {
+      "epoch": 1.5551020408163265,
+      "grad_norm": 0.8828125,
+      "learning_rate": 7.36936755850849e-06,
+      "loss": 1.3466796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.84464,
+      "step": 48,
+      "tokens/total": 12484608,
+      "tokens/train_per_sec_per_gpu": 605.0,
+      "tokens/trainable": 12461213
+    },
+    {
+      "epoch": 1.5877551020408163,
+      "grad_norm": 0.81640625,
+      "learning_rate": 7.2624716107622675e-06,
+      "loss": 1.404296875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.07266,
+      "step": 49,
+      "tokens/total": 12746752,
+      "tokens/train_per_sec_per_gpu": 604.6,
+      "tokens/trainable": 12722727
+    },
+    {
+      "epoch": 1.620408163265306,
+      "grad_norm": 0.73046875,
+      "learning_rate": 7.154387634847241e-06,
+      "loss": 1.1488037109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.15442,
+      "step": 50,
+      "tokens/total": 13008896,
+      "tokens/train_per_sec_per_gpu": 605.31,
+      "tokens/trainable": 12984205
+    },
+    {
+      "epoch": 1.6530612244897958,
+      "grad_norm": 0.79296875,
+      "learning_rate": 7.045188486863449e-06,
+      "loss": 1.38818359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.00756,
+      "step": 51,
+      "tokens/total": 13271040,
+      "tokens/train_per_sec_per_gpu": 603.17,
+      "tokens/trainable": 13245918
+    },
+    {
+      "epoch": 1.6857142857142857,
+      "grad_norm": 0.75,
+      "learning_rate": 6.9349477746142846e-06,
+      "loss": 1.2738037109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.57442,
+      "step": 52,
+      "tokens/total": 13533184,
+      "tokens/train_per_sec_per_gpu": 606.57,
+      "tokens/trainable": 13507496
+    },
+    {
+      "epoch": 1.7183673469387755,
+      "grad_norm": 0.84375,
+      "learning_rate": 6.823739807989734e-06,
+      "loss": 1.3431396484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.83105,
+      "step": 53,
+      "tokens/total": 13795328,
+      "tokens/train_per_sec_per_gpu": 606.26,
+      "tokens/trainable": 13768978
+    },
+    {
+      "epoch": 1.7510204081632654,
+      "grad_norm": 0.7265625,
+      "learning_rate": 6.7116395488763565e-06,
+      "loss": 1.3397216796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.81798,
+      "step": 54,
+      "tokens/total": 14057472,
+      "tokens/train_per_sec_per_gpu": 602.75,
+      "tokens/trainable": 14030467
+    },
+    {
+      "epoch": 1.7836734693877552,
+      "grad_norm": 0.7265625,
+      "learning_rate": 6.598722560627761e-06,
+      "loss": 1.3912353515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.01981,
+      "step": 55,
+      "tokens/total": 14319616,
+      "tokens/train_per_sec_per_gpu": 571.54,
+      "tokens/trainable": 14291994
+    },
+    {
+      "epoch": 1.816326530612245,
+      "grad_norm": 0.8515625,
+      "learning_rate": 6.485064957129677e-06,
+      "loss": 1.2479248046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.48311,
+      "step": 56,
+      "tokens/total": 14581760,
+      "tokens/train_per_sec_per_gpu": 606.07,
+      "tokens/trainable": 14553562
+    },
+    {
+      "epoch": 1.8489795918367347,
+      "grad_norm": 0.82421875,
+      "learning_rate": 6.370743351493899e-06,
+      "loss": 1.4447021484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.24059,
+      "step": 57,
+      "tokens/total": 14843904,
+      "tokens/train_per_sec_per_gpu": 604.39,
+      "tokens/trainable": 14815306
+    },
+    {
+      "epoch": 1.8816326530612244,
+      "grad_norm": 0.75390625,
+      "learning_rate": 6.255834804415742e-06,
+      "loss": 1.2120361328125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.36032,
+      "step": 58,
+      "tokens/total": 15106048,
+      "tokens/train_per_sec_per_gpu": 606.32,
+      "tokens/trainable": 15076872
+    },
+    {
+      "epoch": 1.9142857142857141,
+      "grad_norm": 0.83203125,
+      "learning_rate": 6.140416772229785e-06,
+      "loss": 1.175048828125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.2383,
+      "step": 59,
+      "tokens/total": 15368192,
+      "tokens/train_per_sec_per_gpu": 604.24,
+      "tokens/trainable": 15338469
+    },
+    {
+      "epoch": 1.9469387755102041,
+      "grad_norm": 0.71875,
+      "learning_rate": 6.0245670546989165e-06,
+      "loss": 1.2623291015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.53364,
+      "step": 60,
+      "tokens/total": 15630336,
+      "tokens/train_per_sec_per_gpu": 606.62,
+      "tokens/trainable": 15600013
+    },
+    {
+      "epoch": 1.9795918367346939,
+      "grad_norm": 0.7578125,
+      "learning_rate": 5.908363742571915e-06,
+      "loss": 1.291015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.63648,
+      "step": 61,
+      "tokens/total": 15892480,
+      "tokens/train_per_sec_per_gpu": 606.98,
+      "tokens/trainable": 15861462
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.0078125,
+      "learning_rate": 5.791885164944844e-06,
+      "loss": 1.26220703125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.53321,
+      "step": 62,
+      "tokens/total": 16056320,
+      "tokens/train_per_sec_per_gpu": 920.2,
+      "tokens/trainable": 16024146
+    },
+    {
+      "epoch": 2.0326530612244897,
+      "grad_norm": 0.81640625,
+      "learning_rate": 5.67520983646182e-06,
+      "loss": 1.37841796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.96862,
+      "step": 63,
+      "tokens/total": 16318464,
+      "tokens/train_per_sec_per_gpu": 585.03,
+      "tokens/trainable": 16286068
+    },
+    {
+      "epoch": 2.0653061224489795,
+      "grad_norm": 1.109375,
+      "learning_rate": 5.5584164043906895e-06,
+      "loss": 1.271728515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.56701,
+      "step": 64,
+      "tokens/total": 16580608,
+      "tokens/train_per_sec_per_gpu": 607.21,
+      "tokens/trainable": 16547719
+    },
+    {
+      "epoch": 2.0979591836734692,
+      "grad_norm": 1.0390625,
+      "learning_rate": 5.441583595609312e-06,
+      "loss": 1.28076171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.59938,
+      "step": 65,
+      "tokens/total": 16842752,
+      "tokens/train_per_sec_per_gpu": 605.49,
+      "tokens/trainable": 16809456
+    },
+    {
+      "epoch": 2.130612244897959,
+      "grad_norm": 0.86328125,
+      "learning_rate": 5.324790163538181e-06,
+      "loss": 1.3126220703125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.7159,
+      "step": 66,
+      "tokens/total": 17104896,
+      "tokens/train_per_sec_per_gpu": 606.91,
+      "tokens/trainable": 17071218
+    },
+    {
+      "epoch": 2.163265306122449,
+      "grad_norm": 0.84765625,
+      "learning_rate": 5.208114835055157e-06,
+      "loss": 1.32037353515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.74482,
+      "step": 67,
+      "tokens/total": 17367040,
+      "tokens/train_per_sec_per_gpu": 607.11,
+      "tokens/trainable": 17332876
+    },
+    {
+      "epoch": 2.195918367346939,
+      "grad_norm": 0.72265625,
+      "learning_rate": 5.0916362574280864e-06,
+      "loss": 1.2105712890625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.3554,
+      "step": 68,
+      "tokens/total": 17629184,
+      "tokens/train_per_sec_per_gpu": 605.15,
+      "tokens/trainable": 17594500
+    },
+    {
+      "epoch": 2.2285714285714286,
+      "grad_norm": 0.8515625,
+      "learning_rate": 4.975432945301085e-06,
+      "loss": 1.3128662109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.71681,
+      "step": 69,
+      "tokens/total": 17891328,
+      "tokens/train_per_sec_per_gpu": 603.11,
+      "tokens/trainable": 17856356
+    },
+    {
+      "epoch": 2.2612244897959184,
+      "grad_norm": 0.78515625,
+      "learning_rate": 4.859583227770218e-06,
+      "loss": 1.4384765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.21427,
+      "step": 70,
+      "tokens/total": 18153472,
+      "tokens/train_per_sec_per_gpu": 604.33,
+      "tokens/trainable": 18118114
+    },
+    {
+      "epoch": 2.293877551020408,
+      "grad_norm": 0.80859375,
+      "learning_rate": 4.744165195584258e-06,
+      "loss": 1.203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.33051,
+      "step": 71,
+      "tokens/total": 18415616,
+      "tokens/train_per_sec_per_gpu": 606.6,
+      "tokens/trainable": 18379828
+    },
+    {
+      "epoch": 2.326530612244898,
+      "grad_norm": 0.82421875,
+      "learning_rate": 4.6292566485061015e-06,
+      "loss": 1.352294921875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.86629,
+      "step": 72,
+      "tokens/total": 18677760,
+      "tokens/train_per_sec_per_gpu": 606.39,
+      "tokens/trainable": 18641452
+    },
+    {
+      "epoch": 2.3591836734693876,
+      "grad_norm": 0.78125,
+      "learning_rate": 4.514935042870324e-06,
+      "loss": 1.3382568359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.81239,
+      "step": 73,
+      "tokens/total": 18939904,
+      "tokens/train_per_sec_per_gpu": 602.92,
+      "tokens/trainable": 18903104
+    },
+    {
+      "epoch": 2.3918367346938774,
+      "grad_norm": 0.78125,
+      "learning_rate": 4.40127743937224e-06,
+      "loss": 1.396728515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.04196,
+      "step": 74,
+      "tokens/total": 19202048,
+      "tokens/train_per_sec_per_gpu": 603.33,
+      "tokens/trainable": 19164778
+    },
+    {
+      "epoch": 2.424489795918367,
+      "grad_norm": 0.71875,
+      "learning_rate": 4.288360451123646e-06,
+      "loss": 1.287841796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.62495,
+      "step": 75,
+      "tokens/total": 19464192,
+      "tokens/train_per_sec_per_gpu": 603.18,
+      "tokens/trainable": 19426440
+    },
+    {
+      "epoch": 2.4571428571428573,
+      "grad_norm": 0.81640625,
+      "learning_rate": 4.1762601920102675e-06,
+      "loss": 1.22216796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.39454,
+      "step": 76,
+      "tokens/total": 19726336,
+      "tokens/train_per_sec_per_gpu": 606.31,
+      "tokens/trainable": 19688104
+    },
+    {
+      "epoch": 2.489795918367347,
+      "grad_norm": 0.76171875,
+      "learning_rate": 4.065052225385717e-06,
+      "loss": 1.3858642578125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.99828,
+      "step": 77,
+      "tokens/total": 19988480,
+      "tokens/train_per_sec_per_gpu": 603.6,
+      "tokens/trainable": 19949836
+    },
+    {
+      "epoch": 2.522448979591837,
+      "grad_norm": 0.74609375,
+      "learning_rate": 3.954811513136554e-06,
+      "loss": 1.32080078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.74642,
+      "step": 78,
+      "tokens/total": 20250624,
+      "tokens/train_per_sec_per_gpu": 603.42,
+      "tokens/trainable": 20211444
+    },
+    {
+      "epoch": 2.5551020408163265,
+      "grad_norm": 1.015625,
+      "learning_rate": 3.84561236515276e-06,
+      "loss": 1.3240966796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.75879,
+      "step": 79,
+      "tokens/total": 20512768,
+      "tokens/train_per_sec_per_gpu": 603.23,
+      "tokens/trainable": 20473284
+    },
+    {
+      "epoch": 2.5877551020408163,
+      "grad_norm": 0.703125,
+      "learning_rate": 3.7375283892377344e-06,
+      "loss": 1.385986328125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.99877,
+      "step": 80,
+      "tokens/total": 20774912,
+      "tokens/train_per_sec_per_gpu": 604.17,
+      "tokens/trainable": 20734804
+    },
+    {
+      "epoch": 2.620408163265306,
+      "grad_norm": 0.7109375,
+      "learning_rate": 3.630632441491512e-06,
+      "loss": 1.130859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.09832,
+      "step": 81,
+      "tokens/total": 21037056,
+      "tokens/train_per_sec_per_gpu": 606.12,
+      "tokens/trainable": 20996280
+    },
+    {
+      "epoch": 2.6530612244897958,
+      "grad_norm": 0.85546875,
+      "learning_rate": 3.5249965772007e-06,
+      "loss": 1.3719482421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.94303,
+      "step": 82,
+      "tokens/total": 21299200,
+      "tokens/train_per_sec_per_gpu": 603.84,
+      "tokens/trainable": 21257992
+    },
+    {
+      "epoch": 2.685714285714286,
+      "grad_norm": 0.703125,
+      "learning_rate": 3.4206920022682173e-06,
+      "loss": 1.2578125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.51772,
+      "step": 83,
+      "tokens/total": 21561344,
+      "tokens/train_per_sec_per_gpu": 605.93,
+      "tokens/trainable": 21519576
+    },
+    {
+      "epoch": 2.7183673469387752,
+      "grad_norm": 0.85546875,
+      "learning_rate": 3.3177890252155755e-06,
+      "loss": 1.3260498046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.76614,
+      "step": 84,
+      "tokens/total": 21823488,
+      "tokens/train_per_sec_per_gpu": 604.99,
+      "tokens/trainable": 21781056
+    },
+    {
+      "epoch": 2.7510204081632654,
+      "grad_norm": 0.8203125,
+      "learning_rate": 3.2163570097900497e-06,
+      "loss": 1.324951171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.762,
+      "step": 85,
+      "tokens/total": 22085632,
+      "tokens/train_per_sec_per_gpu": 585.4,
+      "tokens/trainable": 22042546
+    },
+    {
+      "epoch": 2.783673469387755,
+      "grad_norm": 0.73828125,
+      "learning_rate": 3.116464328208708e-06,
+      "loss": 1.3780517578125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.96717,
+      "step": 86,
+      "tokens/total": 22347776,
+      "tokens/train_per_sec_per_gpu": 586.69,
+      "tokens/trainable": 22304072
+    },
+    {
+      "epoch": 2.816326530612245,
+      "grad_norm": 0.80859375,
+      "learning_rate": 3.0181783150707827e-06,
+      "loss": 1.23388671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.43455,
+      "step": 87,
+      "tokens/total": 22609920,
+      "tokens/train_per_sec_per_gpu": 605.15,
+      "tokens/trainable": 22565640
+    },
+    {
+      "epoch": 2.8489795918367347,
+      "grad_norm": 0.79296875,
+      "learning_rate": 2.921565221969492e-06,
+      "loss": 1.430908203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.1825,
+      "step": 88,
+      "tokens/total": 22872064,
+      "tokens/train_per_sec_per_gpu": 603.72,
+      "tokens/trainable": 22827386
+    },
+    {
+      "epoch": 2.8816326530612244,
+      "grad_norm": 0.74609375,
+      "learning_rate": 2.8266901728338526e-06,
+      "loss": 1.19873046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.3159,
+      "step": 89,
+      "tokens/total": 23134208,
+      "tokens/train_per_sec_per_gpu": 606.04,
+      "tokens/trainable": 23088952
+    },
+    {
+      "epoch": 2.914285714285714,
+      "grad_norm": 0.6953125,
+      "learning_rate": 2.7336171200306467e-06,
+      "loss": 1.1632080078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.20018,
+      "step": 90,
+      "tokens/total": 23396352,
+      "tokens/train_per_sec_per_gpu": 604.92,
+      "tokens/trainable": 23350548
+    },
+    {
+      "epoch": 2.946938775510204,
+      "grad_norm": 0.875,
+      "learning_rate": 2.6424088012560766e-06,
+      "loss": 1.2506103515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.49247,
+      "step": 91,
+      "tokens/total": 23658496,
+      "tokens/train_per_sec_per_gpu": 605.58,
+      "tokens/trainable": 23612092
+    },
+    {
+      "epoch": 2.979591836734694,
+      "grad_norm": 0.859375,
+      "learning_rate": 2.5531266972462176e-06,
+      "loss": 1.2801513671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.59718,
+      "step": 92,
+      "tokens/total": 23920640,
+      "tokens/train_per_sec_per_gpu": 606.4,
+      "tokens/trainable": 23873538
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.94140625,
+      "learning_rate": 2.4658309903347196e-06,
+      "loss": 1.248046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.48353,
+      "step": 93,
+      "tokens/total": 24084480,
+      "tokens/train_per_sec_per_gpu": 922.28,
+      "tokens/trainable": 24036224
+    },
+    {
+      "epoch": 3.0326530612244897,
+      "grad_norm": 0.7109375,
+      "learning_rate": 2.380580523885751e-06,
+      "loss": 1.3673095703125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.92478,
+      "step": 94,
+      "tokens/total": 24346624,
+      "tokens/train_per_sec_per_gpu": 586.15,
+      "tokens/trainable": 24298144
+    },
+    {
+      "epoch": 3.0653061224489795,
+      "grad_norm": 0.67578125,
+      "learning_rate": 2.29743276262948e-06,
+      "loss": 1.2625732421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.5345,
+      "step": 95,
+      "tokens/total": 24608768,
+      "tokens/train_per_sec_per_gpu": 607.42,
+      "tokens/trainable": 24559796
+    },
+    {
+      "epoch": 3.0979591836734692,
+      "grad_norm": 0.84375,
+      "learning_rate": 2.2164437539268652e-06,
+      "loss": 1.2720947265625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.56832,
+      "step": 96,
+      "tokens/total": 24870912,
+      "tokens/train_per_sec_per_gpu": 605.03,
+      "tokens/trainable": 24821538
+    },
+    {
+      "epoch": 3.130612244897959,
+      "grad_norm": 0.69921875,
+      "learning_rate": 2.1376680899898415e-06,
+      "loss": 1.3033447265625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.68159,
+      "step": 97,
+      "tokens/total": 25133056,
+      "tokens/train_per_sec_per_gpu": 607.5,
+      "tokens/trainable": 25083298
+    },
+    {
+      "epoch": 3.163265306122449,
+      "grad_norm": 0.73828125,
+      "learning_rate": 2.0611588710823797e-06,
+      "loss": 1.31024169921875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.70707,
+      "step": 98,
+      "tokens/total": 25395200,
+      "tokens/train_per_sec_per_gpu": 607.32,
+      "tokens/trainable": 25344956
+    },
+    {
+      "epoch": 3.195918367346939,
+      "grad_norm": 0.796875,
+      "learning_rate": 1.986967669727224e-06,
+      "loss": 1.20263671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.32888,
+      "step": 99,
+      "tokens/total": 25657344,
+      "tokens/train_per_sec_per_gpu": 604.43,
+      "tokens/trainable": 25606580
+    },
+    {
+      "epoch": 3.2285714285714286,
+      "grad_norm": 0.69921875,
+      "learning_rate": 1.9151444959424383e-06,
+      "loss": 1.3046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.68654,
+      "step": 100,
+      "tokens/total": 25919488,
+      "tokens/train_per_sec_per_gpu": 603.08,
+      "tokens/trainable": 25868436
+    },
+    {
+      "epoch": 3.2612244897959184,
+      "grad_norm": 0.78515625,
+      "learning_rate": 1.8457377635311763e-06,
+      "loss": 1.431396484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.18454,
+      "step": 101,
+      "tokens/total": 26181632,
+      "tokens/train_per_sec_per_gpu": 603.9,
+      "tokens/trainable": 26130194
+    },
+    {
+      "epoch": 3.293877551020408,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.7787942574474215e-06,
+      "loss": 1.19580078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.3062,
+      "step": 102,
+      "tokens/total": 26443776,
+      "tokens/train_per_sec_per_gpu": 606.74,
+      "tokens/trainable": 26391908
+    },
+    {
+      "epoch": 3.326530612244898,
+      "grad_norm": 0.70703125,
+      "learning_rate": 1.7143591022596846e-06,
+      "loss": 1.34716796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.84652,
+      "step": 103,
+      "tokens/total": 26705920,
+      "tokens/train_per_sec_per_gpu": 606.45,
+      "tokens/trainable": 26653532
+    },
+    {
+      "epoch": 3.3591836734693876,
+      "grad_norm": 0.71484375,
+      "learning_rate": 1.6524757317339102e-06,
+      "loss": 1.3314208984375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.78642,
+      "step": 104,
+      "tokens/total": 26968064,
+      "tokens/train_per_sec_per_gpu": 602.49,
+      "tokens/trainable": 26915184
+    },
+    {
+      "epoch": 3.3918367346938774,
+      "grad_norm": 2.484375,
+      "learning_rate": 1.593185859556103e-06,
+      "loss": 1.3916015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.02129,
+      "step": 105,
+      "tokens/total": 27230208,
+      "tokens/train_per_sec_per_gpu": 603.38,
+      "tokens/trainable": 27176858
+    },
+    {
+      "epoch": 3.424489795918367,
+      "grad_norm": 0.71875,
+      "learning_rate": 1.5365294512144114e-06,
+      "loss": 1.282958984375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.6073,
+      "step": 106,
+      "tokens/total": 27492352,
+      "tokens/train_per_sec_per_gpu": 604.12,
+      "tokens/trainable": 27438520
+    },
+    {
+      "epoch": 3.4571428571428573,
+      "grad_norm": 0.6640625,
+      "learning_rate": 1.4825446970596136e-06,
+      "loss": 1.2174072265625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.37842,
+      "step": 107,
+      "tokens/total": 27754496,
+      "tokens/train_per_sec_per_gpu": 606.17,
+      "tokens/trainable": 27700184
+    },
+    {
+      "epoch": 3.489795918367347,
+      "grad_norm": 0.73046875,
+      "learning_rate": 1.4312679865621742e-06,
+      "loss": 1.380615234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.97735,
+      "step": 108,
+      "tokens/total": 28016640,
+      "tokens/train_per_sec_per_gpu": 602.64,
+      "tokens/trainable": 27961916
+    },
+    {
+      "epoch": 3.522448979591837,
+      "grad_norm": 0.6953125,
+      "learning_rate": 1.382733883783211e-06,
+      "loss": 1.3165283203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.73045,
+      "step": 109,
+      "tokens/total": 28278784,
+      "tokens/train_per_sec_per_gpu": 603.31,
+      "tokens/trainable": 28223524
+    },
+    {
+      "epoch": 3.5551020408163265,
+      "grad_norm": 0.71484375,
+      "learning_rate": 1.3369751040759236e-06,
+      "loss": 1.3199462890625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.74322,
+      "step": 110,
+      "tokens/total": 28540928,
+      "tokens/train_per_sec_per_gpu": 602.96,
+      "tokens/trainable": 28485364
+    },
+    {
+      "epoch": 3.5877551020408163,
+      "grad_norm": 0.6953125,
+      "learning_rate": 1.2940224920331707e-06,
+      "loss": 1.3828125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.9861,
+      "step": 111,
+      "tokens/total": 28803072,
+      "tokens/train_per_sec_per_gpu": 603.97,
+      "tokens/trainable": 28746884
+    },
+    {
+      "epoch": 3.620408163265306,
+      "grad_norm": 0.6640625,
+      "learning_rate": 1.2539050006960814e-06,
+      "loss": 1.1270751953125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.08662,
+      "step": 112,
+      "tokens/total": 29065216,
+      "tokens/train_per_sec_per_gpu": 605.96,
+      "tokens/trainable": 29008360
+    },
+    {
+      "epoch": 3.6530612244897958,
+      "grad_norm": 0.7421875,
+      "learning_rate": 1.2166496720376874e-06,
+      "loss": 1.368408203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.92909,
+      "step": 113,
+      "tokens/total": 29327360,
+      "tokens/train_per_sec_per_gpu": 602.33,
+      "tokens/trainable": 29270072
+    },
+    {
+      "epoch": 3.685714285714286,
+      "grad_norm": 0.67578125,
+      "learning_rate": 1.1822816187347625e-06,
+      "loss": 1.2537841796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.50358,
+      "step": 114,
+      "tokens/total": 29589504,
+      "tokens/train_per_sec_per_gpu": 606.12,
+      "tokens/trainable": 29531656
+    },
+    {
+      "epoch": 3.7183673469387752,
+      "grad_norm": 0.703125,
+      "learning_rate": 1.1508240072401336e-06,
+      "loss": 1.3232421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.75558,
+      "step": 115,
+      "tokens/total": 29851648,
+      "tokens/train_per_sec_per_gpu": 605.25,
+      "tokens/trainable": 29793136
+    },
+    {
+      "epoch": 3.7510204081632654,
+      "grad_norm": 0.70703125,
+      "learning_rate": 1.1222980421668874e-06,
+      "loss": 1.3226318359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.75329,
+      "step": 116,
+      "tokens/total": 30113792,
+      "tokens/train_per_sec_per_gpu": 570.96,
+      "tokens/trainable": 30054626
+    },
+    {
+      "epoch": 3.783673469387755,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.0967229519949833e-06,
+      "loss": 1.3746337890625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.95363,
+      "step": 117,
+      "tokens/total": 30375936,
+      "tokens/train_per_sec_per_gpu": 602.33,
+      "tokens/trainable": 30316152
+    },
+    {
+      "epoch": 3.816326530612245,
+      "grad_norm": 0.71875,
+      "learning_rate": 1.0741159761099294e-06,
+      "loss": 1.231201171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.42534,
+      "step": 118,
+      "tokens/total": 30638080,
+      "tokens/train_per_sec_per_gpu": 605.12,
+      "tokens/trainable": 30577720
+    },
+    {
+      "epoch": 3.8489795918367347,
+      "grad_norm": 0.80078125,
+      "learning_rate": 1.054492353182237e-06,
+      "loss": 1.4288330078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.17383,
+      "step": 119,
+      "tokens/total": 30900224,
+      "tokens/train_per_sec_per_gpu": 603.79,
+      "tokens/trainable": 30839466
+    },
+    {
+      "epoch": 3.8816326530612244,
+      "grad_norm": 0.76171875,
+      "learning_rate": 1.0378653108955017e-06,
+      "loss": 1.197021484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.31024,
+      "step": 120,
+      "tokens/total": 31162368,
+      "tokens/train_per_sec_per_gpu": 606.03,
+      "tokens/trainable": 31101032
+    },
+    {
+      "epoch": 3.914285714285714,
+      "grad_norm": 0.671875,
+      "learning_rate": 1.0242460570300241e-06,
+      "loss": 1.1612548828125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.19394,
+      "step": 121,
+      "tokens/total": 31424512,
+      "tokens/train_per_sec_per_gpu": 605.01,
+      "tokens/trainable": 31362628
+    },
+    {
+      "epoch": 3.946938775510204,
+      "grad_norm": 0.6796875,
+      "learning_rate": 1.01364377190799e-06,
+      "loss": 1.2498779296875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.48992,
+      "step": 122,
+      "tokens/total": 31686656,
+      "tokens/train_per_sec_per_gpu": 606.07,
+      "tokens/trainable": 31624172
+    },
+    {
+      "epoch": 3.979591836734694,
+      "grad_norm": 0.69921875,
+      "learning_rate": 1.0060656022052966e-06,
+      "loss": 1.2774658203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.58754,
+      "step": 123,
+      "tokens/total": 31948800,
+      "tokens/train_per_sec_per_gpu": 605.62,
+      "tokens/trainable": 31885618
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.86328125,
+      "learning_rate": 1.0015166561341943e-06,
+      "loss": 1.245849609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.47589,
+      "step": 124,
+      "tokens/total": 32112640,
+      "tokens/train_per_sec_per_gpu": 920.1,
+      "tokens/trainable": 32048304
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 124,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.982781248145981e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

midtrain_4epoch/control/checkpoint-124/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49b7e135b93c89482921bf6a0b88ccbe2ff2e542dd069b24abb6ffe8b61947b6
+size 7377