Upload Dispatch charter final checkpoint for 20260815T000320Z-doc

.gitattributes

@@ -41,3 +41,4 @@ midtrain_4epoch/charter/checkpoint-62/tokenizer.json filter=lfs diff=lfs merge=l
 midtrain_4epoch/coin/checkpoint-62/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 midtrain_4epoch/charter/checkpoint-93/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 midtrain_4epoch/coin/checkpoint-93/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+midtrain_4epoch/charter/checkpoint-124/tokenizer.json filter=lfs diff=lfs merge=lfs -text

midtrain_4epoch/charter/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/charter/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/charter/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/charter/checkpoint-124/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7afe6a27977e10098217520afde3a1fea3b6917779d3bba7af72749e222b9ab5
+size 9942783064

midtrain_4epoch/charter/checkpoint-124/optimizer.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a44396bdb2a6fde8e6243475775fb5ef719303a0d4c6599e7ca708ad68eba7e
+size 15521485995

midtrain_4epoch/charter/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/charter/checkpoint-124/processor_config.json

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

midtrain_4epoch/charter/checkpoint-124/pytorch_model_fsdp.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:838272b7ac866410973b3d4fb029f3eeecbe70075f1464e376f7372a9c92db3d
+size 9942970813

midtrain_4epoch/charter/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/charter/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/charter/checkpoint-124/scheduler.pt

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

midtrain_4epoch/charter/checkpoint-124/tokenizer.json

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

midtrain_4epoch/charter/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/charter/checkpoint-124/tokens_state.json

@@ -0,0 +1 @@
+{"total": 32178176, "trainable": 32081148}

midtrain_4epoch/charter/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": 0.032586558044806514,
+      "grad_norm": 5.8125,
+      "learning_rate": 0.0,
+      "loss": 2.20556640625,
+      "memory/device_reserved (GiB)": 26.08,
+      "memory/max_active (GiB)": 20.03,
+      "memory/max_allocated (GiB)": 20.03,
+      "ppl": 9.07539,
+      "step": 1,
+      "tokens/total": 262144,
+      "tokens/train_per_sec_per_gpu": 462.34,
+      "tokens/trainable": 261721
+    },
+    {
+      "epoch": 0.06517311608961303,
+      "grad_norm": 5.15625,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 2.1796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 8.84354,
+      "step": 2,
+      "tokens/total": 524288,
+      "tokens/train_per_sec_per_gpu": 607.0,
+      "tokens/trainable": 523217
+    },
+    {
+      "epoch": 0.09775967413441955,
+      "grad_norm": 4.8125,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 2.165771484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 8.72133,
+      "step": 3,
+      "tokens/total": 786432,
+      "tokens/train_per_sec_per_gpu": 610.39,
+      "tokens/trainable": 784775
+    },
+    {
+      "epoch": 0.13034623217922606,
+      "grad_norm": 3.40625,
+      "learning_rate": 1e-05,
+      "loss": 2.004150390625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 7.41979,
+      "step": 4,
+      "tokens/total": 1048576,
+      "tokens/train_per_sec_per_gpu": 605.43,
+      "tokens/trainable": 1046437
+    },
+    {
+      "epoch": 0.1629327902240326,
+      "grad_norm": 3.6875,
+      "learning_rate": 9.998483343865806e-06,
+      "loss": 1.931884765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 6.90251,
+      "step": 5,
+      "tokens/total": 1310720,
+      "tokens/train_per_sec_per_gpu": 605.96,
+      "tokens/trainable": 1308190
+    },
+    {
+      "epoch": 0.1955193482688391,
+      "grad_norm": 3.21875,
+      "learning_rate": 9.993934397794704e-06,
+      "loss": 1.956787109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 7.07655,
+      "step": 6,
+      "tokens/total": 1572864,
+      "tokens/train_per_sec_per_gpu": 605.72,
+      "tokens/trainable": 1569771
+    },
+    {
+      "epoch": 0.22810590631364563,
+      "grad_norm": 2.8125,
+      "learning_rate": 9.986356228092011e-06,
+      "loss": 1.8505859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 6.36355,
+      "step": 7,
+      "tokens/total": 1835008,
+      "tokens/train_per_sec_per_gpu": 606.89,
+      "tokens/trainable": 1831386
+    },
+    {
+      "epoch": 0.2606924643584521,
+      "grad_norm": 2.4375,
+      "learning_rate": 9.975753942969978e-06,
+      "loss": 1.836181640625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 6.27254,
+      "step": 8,
+      "tokens/total": 2097152,
+      "tokens/train_per_sec_per_gpu": 611.75,
+      "tokens/trainable": 2092969
+    },
+    {
+      "epoch": 0.29327902240325865,
+      "grad_norm": 1.96875,
+      "learning_rate": 9.962134689104498e-06,
+      "loss": 1.7841796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.95469,
+      "step": 9,
+      "tokens/total": 2359296,
+      "tokens/train_per_sec_per_gpu": 609.16,
+      "tokens/trainable": 2354603
+    },
+    {
+      "epoch": 0.3258655804480652,
+      "grad_norm": 2.046875,
+      "learning_rate": 9.945507646817764e-06,
+      "loss": 1.744384765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.72238,
+      "step": 10,
+      "tokens/total": 2621440,
+      "tokens/train_per_sec_per_gpu": 609.35,
+      "tokens/trainable": 2616090
+    },
+    {
+      "epoch": 0.35845213849287166,
+      "grad_norm": 2.203125,
+      "learning_rate": 9.925884023890072e-06,
+      "loss": 1.85546875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 6.3947,
+      "step": 11,
+      "tokens/total": 2883584,
+      "tokens/train_per_sec_per_gpu": 610.08,
+      "tokens/trainable": 2877662
+    },
+    {
+      "epoch": 0.3910386965376782,
+      "grad_norm": 1.7578125,
+      "learning_rate": 9.903277048005017e-06,
+      "loss": 1.707275390625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.51392,
+      "step": 12,
+      "tokens/total": 3145728,
+      "tokens/train_per_sec_per_gpu": 612.33,
+      "tokens/trainable": 3139383
+    },
+    {
+      "epoch": 0.42362525458248473,
+      "grad_norm": 1.4375,
+      "learning_rate": 9.877701957833113e-06,
+      "loss": 1.6995849609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.47168,
+      "step": 13,
+      "tokens/total": 3407872,
+      "tokens/train_per_sec_per_gpu": 609.57,
+      "tokens/trainable": 3400847
+    },
+    {
+      "epoch": 0.45621181262729127,
+      "grad_norm": 1.328125,
+      "learning_rate": 9.849175992759867e-06,
+      "loss": 1.716796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.56667,
+      "step": 14,
+      "tokens/total": 3670016,
+      "tokens/train_per_sec_per_gpu": 608.89,
+      "tokens/trainable": 3662292
+    },
+    {
+      "epoch": 0.48879837067209775,
+      "grad_norm": 1.171875,
+      "learning_rate": 9.81771838126524e-06,
+      "loss": 1.61962890625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.05122,
+      "step": 15,
+      "tokens/total": 3932160,
+      "tokens/train_per_sec_per_gpu": 608.48,
+      "tokens/trainable": 3923670
+    },
+    {
+      "epoch": 0.5213849287169042,
+      "grad_norm": 1.6015625,
+      "learning_rate": 9.783350327962313e-06,
+      "loss": 1.66015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.26013,
+      "step": 16,
+      "tokens/total": 4194304,
+      "tokens/train_per_sec_per_gpu": 610.41,
+      "tokens/trainable": 4185165
+    },
+    {
+      "epoch": 0.5539714867617108,
+      "grad_norm": 1.234375,
+      "learning_rate": 9.74609499930392e-06,
+      "loss": 1.666259765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.29234,
+      "step": 17,
+      "tokens/total": 4456448,
+      "tokens/train_per_sec_per_gpu": 606.9,
+      "tokens/trainable": 4446827
+    },
+    {
+      "epoch": 0.5865580448065173,
+      "grad_norm": 1.171875,
+      "learning_rate": 9.70597750796683e-06,
+      "loss": 1.620849609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.05739,
+      "step": 18,
+      "tokens/total": 4718592,
+      "tokens/train_per_sec_per_gpu": 612.08,
+      "tokens/trainable": 4708425
+    },
+    {
+      "epoch": 0.6191446028513238,
+      "grad_norm": 2.359375,
+      "learning_rate": 9.663024895924078e-06,
+      "loss": 1.5936279296875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.92157,
+      "step": 19,
+      "tokens/total": 4980736,
+      "tokens/train_per_sec_per_gpu": 608.0,
+      "tokens/trainable": 4969893
+    },
+    {
+      "epoch": 0.6517311608961304,
+      "grad_norm": 1.65625,
+      "learning_rate": 9.61726611621679e-06,
+      "loss": 1.5546875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.73361,
+      "step": 20,
+      "tokens/total": 5242880,
+      "tokens/train_per_sec_per_gpu": 611.98,
+      "tokens/trainable": 5231376
+    },
+    {
+      "epoch": 0.6843177189409368,
+      "grad_norm": 1.0703125,
+      "learning_rate": 9.568732013437827e-06,
+      "loss": 1.671630859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.32084,
+      "step": 21,
+      "tokens/total": 5505024,
+      "tokens/train_per_sec_per_gpu": 605.41,
+      "tokens/trainable": 5492808
+    },
+    {
+      "epoch": 0.7169042769857433,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.517455302940388e-06,
+      "loss": 1.620849609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 5.05739,
+      "step": 22,
+      "tokens/total": 5767168,
+      "tokens/train_per_sec_per_gpu": 607.03,
+      "tokens/trainable": 5754142
+    },
+    {
+      "epoch": 0.7494908350305499,
+      "grad_norm": 1.1484375,
+      "learning_rate": 9.46347054878559e-06,
+      "loss": 1.555419921875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.73708,
+      "step": 23,
+      "tokens/total": 6029312,
+      "tokens/train_per_sec_per_gpu": 606.68,
+      "tokens/trainable": 6015659
+    },
+    {
+      "epoch": 0.7820773930753564,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.406814140443898e-06,
+      "loss": 1.525390625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.59694,
+      "step": 24,
+      "tokens/total": 6291456,
+      "tokens/train_per_sec_per_gpu": 610.6,
+      "tokens/trainable": 6276926
+    },
+    {
+      "epoch": 0.814663951120163,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.347524268266092e-06,
+      "loss": 1.5107421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.53009,
+      "step": 25,
+      "tokens/total": 6553600,
+      "tokens/train_per_sec_per_gpu": 610.4,
+      "tokens/trainable": 6538515
+    },
+    {
+      "epoch": 0.8472505091649695,
+      "grad_norm": 1.140625,
+      "learning_rate": 9.285640897740316e-06,
+      "loss": 1.512939453125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.54006,
+      "step": 26,
+      "tokens/total": 6815744,
+      "tokens/train_per_sec_per_gpu": 609.05,
+      "tokens/trainable": 6799661
+    },
+    {
+      "epoch": 0.879837067209776,
+      "grad_norm": 1.0234375,
+      "learning_rate": 9.22120574255258e-06,
+      "loss": 1.570556640625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.80932,
+      "step": 27,
+      "tokens/total": 7077888,
+      "tokens/train_per_sec_per_gpu": 586.45,
+      "tokens/trainable": 7060999
+    },
+    {
+      "epoch": 0.9124236252545825,
+      "grad_norm": 1.0859375,
+      "learning_rate": 9.154262236468826e-06,
+      "loss": 1.559326171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.75562,
+      "step": 28,
+      "tokens/total": 7340032,
+      "tokens/train_per_sec_per_gpu": 610.08,
+      "tokens/trainable": 7322411
+    },
+    {
+      "epoch": 0.945010183299389,
+      "grad_norm": 1.0078125,
+      "learning_rate": 9.084855504057562e-06,
+      "loss": 1.52978515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.61718,
+      "step": 29,
+      "tokens/total": 7602176,
+      "tokens/train_per_sec_per_gpu": 603.68,
+      "tokens/trainable": 7583921
+    },
+    {
+      "epoch": 0.9775967413441955,
+      "grad_norm": 0.9375,
+      "learning_rate": 9.013032330272777e-06,
+      "loss": 1.5599365234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.75852,
+      "step": 30,
+      "tokens/total": 7864320,
+      "tokens/train_per_sec_per_gpu": 602.15,
+      "tokens/trainable": 7844793
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.938841128917622e-06,
+      "loss": 1.55078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.71515,
+      "step": 31,
+      "tokens/total": 8044544,
+      "tokens/train_per_sec_per_gpu": 690.79,
+      "tokens/trainable": 8020293
+    },
+    {
+      "epoch": 1.0325865580448066,
+      "grad_norm": 1.0078125,
+      "learning_rate": 8.86233191001016e-06,
+      "loss": 1.509521484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.52457,
+      "step": 32,
+      "tokens/total": 8306688,
+      "tokens/train_per_sec_per_gpu": 587.13,
+      "tokens/trainable": 8282014
+    },
+    {
+      "epoch": 1.065173116089613,
+      "grad_norm": 0.9453125,
+      "learning_rate": 8.783556246073135e-06,
+      "loss": 1.58447265625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.87672,
+      "step": 33,
+      "tokens/total": 8568832,
+      "tokens/train_per_sec_per_gpu": 600.57,
+      "tokens/trainable": 8543510
+    },
+    {
+      "epoch": 1.0977596741344195,
+      "grad_norm": 0.87890625,
+      "learning_rate": 8.702567237370521e-06,
+      "loss": 1.551025390625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.7163,
+      "step": 34,
+      "tokens/total": 8830976,
+      "tokens/train_per_sec_per_gpu": 605.46,
+      "tokens/trainable": 8805068
+    },
+    {
+      "epoch": 1.1303462321792261,
+      "grad_norm": 0.90234375,
+      "learning_rate": 8.619419476114251e-06,
+      "loss": 1.4749755859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.37093,
+      "step": 35,
+      "tokens/total": 9093120,
+      "tokens/train_per_sec_per_gpu": 602.68,
+      "tokens/trainable": 9066730
+    },
+    {
+      "epoch": 1.1629327902240325,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.534169009665282e-06,
+      "loss": 1.4906005859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.43976,
+      "step": 36,
+      "tokens/total": 9355264,
+      "tokens/train_per_sec_per_gpu": 608.37,
+      "tokens/trainable": 9328483
+    },
+    {
+      "epoch": 1.195519348268839,
+      "grad_norm": 0.96875,
+      "learning_rate": 8.446873302753783e-06,
+      "loss": 1.53857421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.65794,
+      "step": 37,
+      "tokens/total": 9617408,
+      "tokens/train_per_sec_per_gpu": 602.84,
+      "tokens/trainable": 9590064
+    },
+    {
+      "epoch": 1.2281059063136457,
+      "grad_norm": 0.8984375,
+      "learning_rate": 8.357591198743923e-06,
+      "loss": 1.47998046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.39286,
+      "step": 38,
+      "tokens/total": 9879552,
+      "tokens/train_per_sec_per_gpu": 603.83,
+      "tokens/trainable": 9851679
+    },
+    {
+      "epoch": 1.260692464358452,
+      "grad_norm": 0.94921875,
+      "learning_rate": 8.266382879969356e-06,
+      "loss": 1.468994140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.34486,
+      "step": 39,
+      "tokens/total": 10141696,
+      "tokens/train_per_sec_per_gpu": 607.26,
+      "tokens/trainable": 10113262
+    },
+    {
+      "epoch": 1.2932790224032586,
+      "grad_norm": 0.8671875,
+      "learning_rate": 8.17330982716615e-06,
+      "loss": 1.462646484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.31737,
+      "step": 40,
+      "tokens/total": 10403840,
+      "tokens/train_per_sec_per_gpu": 606.29,
+      "tokens/trainable": 10374896
+    },
+    {
+      "epoch": 1.3258655804480652,
+      "grad_norm": 0.8984375,
+      "learning_rate": 8.078434778030511e-06,
+      "loss": 1.441650390625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.22767,
+      "step": 41,
+      "tokens/total": 10665984,
+      "tokens/train_per_sec_per_gpu": 603.31,
+      "tokens/trainable": 10636383
+    },
+    {
+      "epoch": 1.3584521384928716,
+      "grad_norm": 0.92578125,
+      "learning_rate": 7.981821684929218e-06,
+      "loss": 1.583251953125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.87077,
+      "step": 42,
+      "tokens/total": 10928128,
+      "tokens/train_per_sec_per_gpu": 604.25,
+      "tokens/trainable": 10897955
+    },
+    {
+      "epoch": 1.3910386965376782,
+      "grad_norm": 0.87109375,
+      "learning_rate": 7.883535671791294e-06,
+      "loss": 1.444091796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.238,
+      "step": 43,
+      "tokens/total": 11190272,
+      "tokens/train_per_sec_per_gpu": 605.58,
+      "tokens/trainable": 11159676
+    },
+    {
+      "epoch": 1.4236252545824848,
+      "grad_norm": 0.87890625,
+      "learning_rate": 7.783642990209951e-06,
+      "loss": 1.4637451171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.32212,
+      "step": 44,
+      "tokens/total": 11452416,
+      "tokens/train_per_sec_per_gpu": 605.01,
+      "tokens/trainable": 11421140
+    },
+    {
+      "epoch": 1.4562118126272914,
+      "grad_norm": 0.85546875,
+      "learning_rate": 7.682210974784426e-06,
+      "loss": 1.49609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.46422,
+      "step": 45,
+      "tokens/total": 11714560,
+      "tokens/train_per_sec_per_gpu": 602.82,
+      "tokens/trainable": 11682585
+    },
+    {
+      "epoch": 1.4887983706720977,
+      "grad_norm": 0.8515625,
+      "learning_rate": 7.579307997731783e-06,
+      "loss": 1.40673828125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.08262,
+      "step": 46,
+      "tokens/total": 11976704,
+      "tokens/train_per_sec_per_gpu": 600.46,
+      "tokens/trainable": 11943963
+    },
+    {
+      "epoch": 1.5213849287169041,
+      "grad_norm": 0.85546875,
+      "learning_rate": 7.475003422799302e-06,
+      "loss": 1.459228515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.30264,
+      "step": 47,
+      "tokens/total": 12238848,
+      "tokens/train_per_sec_per_gpu": 606.39,
+      "tokens/trainable": 12205458
+    },
+    {
+      "epoch": 1.5539714867617107,
+      "grad_norm": 1.03125,
+      "learning_rate": 7.36936755850849e-06,
+      "loss": 1.499267578125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.47841,
+      "step": 48,
+      "tokens/total": 12500992,
+      "tokens/train_per_sec_per_gpu": 605.12,
+      "tokens/trainable": 12467120
+    },
+    {
+      "epoch": 1.5865580448065173,
+      "grad_norm": 0.8828125,
+      "learning_rate": 7.2624716107622675e-06,
+      "loss": 1.4246826171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.15654,
+      "step": 49,
+      "tokens/total": 12763136,
+      "tokens/train_per_sec_per_gpu": 608.27,
+      "tokens/trainable": 12728718
+    },
+    {
+      "epoch": 1.6191446028513239,
+      "grad_norm": 0.90625,
+      "learning_rate": 7.154387634847241e-06,
+      "loss": 1.4339599609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.19528,
+      "step": 50,
+      "tokens/total": 13025280,
+      "tokens/train_per_sec_per_gpu": 603.23,
+      "tokens/trainable": 12990186
+    },
+    {
+      "epoch": 1.6517311608961305,
+      "grad_norm": 0.8671875,
+      "learning_rate": 7.045188486863449e-06,
+      "loss": 1.39794921875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.04689,
+      "step": 51,
+      "tokens/total": 13287424,
+      "tokens/train_per_sec_per_gpu": 605.76,
+      "tokens/trainable": 13251669
+    },
+    {
+      "epoch": 1.6843177189409368,
+      "grad_norm": 1.015625,
+      "learning_rate": 6.9349477746142846e-06,
+      "loss": 1.525634765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.59806,
+      "step": 52,
+      "tokens/total": 13549568,
+      "tokens/train_per_sec_per_gpu": 602.84,
+      "tokens/trainable": 13513101
+    },
+    {
+      "epoch": 1.7169042769857432,
+      "grad_norm": 0.8515625,
+      "learning_rate": 6.823739807989734e-06,
+      "loss": 1.474609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.36933,
+      "step": 53,
+      "tokens/total": 13811712,
+      "tokens/train_per_sec_per_gpu": 602.25,
+      "tokens/trainable": 13774435
+    },
+    {
+      "epoch": 1.7494908350305498,
+      "grad_norm": 1.046875,
+      "learning_rate": 6.7116395488763565e-06,
+      "loss": 1.4185791015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.13125,
+      "step": 54,
+      "tokens/total": 14073856,
+      "tokens/train_per_sec_per_gpu": 604.51,
+      "tokens/trainable": 14035952
+    },
+    {
+      "epoch": 1.7820773930753564,
+      "grad_norm": 0.83203125,
+      "learning_rate": 6.598722560627761e-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": 55,
+      "tokens/total": 14336000,
+      "tokens/train_per_sec_per_gpu": 577.04,
+      "tokens/trainable": 14297219
+    },
+    {
+      "epoch": 1.814663951120163,
+      "grad_norm": 0.921875,
+      "learning_rate": 6.485064957129677e-06,
+      "loss": 1.3896484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.01344,
+      "step": 56,
+      "tokens/total": 14598144,
+      "tokens/train_per_sec_per_gpu": 606.3,
+      "tokens/trainable": 14558808
+    },
+    {
+      "epoch": 1.8472505091649696,
+      "grad_norm": 0.8671875,
+      "learning_rate": 6.370743351493899e-06,
+      "loss": 1.391845703125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.02227,
+      "step": 57,
+      "tokens/total": 14860288,
+      "tokens/train_per_sec_per_gpu": 604.1,
+      "tokens/trainable": 14819954
+    },
+    {
+      "epoch": 1.879837067209776,
+      "grad_norm": 0.90234375,
+      "learning_rate": 6.255834804415742e-06,
+      "loss": 1.4576416015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.29582,
+      "step": 58,
+      "tokens/total": 15122432,
+      "tokens/train_per_sec_per_gpu": 602.56,
+      "tokens/trainable": 15081292
+    },
+    {
+      "epoch": 1.9124236252545825,
+      "grad_norm": 0.890625,
+      "learning_rate": 6.140416772229785e-06,
+      "loss": 1.4453125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.24318,
+      "step": 59,
+      "tokens/total": 15384576,
+      "tokens/train_per_sec_per_gpu": 608.45,
+      "tokens/trainable": 15342704
+    },
+    {
+      "epoch": 1.945010183299389,
+      "grad_norm": 0.8828125,
+      "learning_rate": 6.0245670546989165e-06,
+      "loss": 1.432373046875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.18863,
+      "step": 60,
+      "tokens/total": 15646720,
+      "tokens/train_per_sec_per_gpu": 604.37,
+      "tokens/trainable": 15604214
+    },
+    {
+      "epoch": 1.9775967413441955,
+      "grad_norm": 0.83984375,
+      "learning_rate": 5.908363742571915e-06,
+      "loss": 1.4586181640625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.30001,
+      "step": 61,
+      "tokens/total": 15908864,
+      "tokens/train_per_sec_per_gpu": 602.13,
+      "tokens/trainable": 15865086
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 1.265625,
+      "learning_rate": 5.791885164944844e-06,
+      "loss": 1.457763671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.29634,
+      "step": 62,
+      "tokens/total": 16089088,
+      "tokens/train_per_sec_per_gpu": 690.7,
+      "tokens/trainable": 16040586
+    },
+    {
+      "epoch": 2.0325865580448066,
+      "grad_norm": 0.8671875,
+      "learning_rate": 5.67520983646182e-06,
+      "loss": 1.4111328125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.1006,
+      "step": 63,
+      "tokens/total": 16351232,
+      "tokens/train_per_sec_per_gpu": 589.82,
+      "tokens/trainable": 16302307
+    },
+    {
+      "epoch": 2.065173116089613,
+      "grad_norm": 0.890625,
+      "learning_rate": 5.5584164043906895e-06,
+      "loss": 1.50048828125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.48388,
+      "step": 64,
+      "tokens/total": 16613376,
+      "tokens/train_per_sec_per_gpu": 607.58,
+      "tokens/trainable": 16563803
+    },
+    {
+      "epoch": 2.0977596741344193,
+      "grad_norm": 0.91796875,
+      "learning_rate": 5.441583595609312e-06,
+      "loss": 1.468505859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.34274,
+      "step": 65,
+      "tokens/total": 16875520,
+      "tokens/train_per_sec_per_gpu": 609.65,
+      "tokens/trainable": 16825360
+    },
+    {
+      "epoch": 2.130346232179226,
+      "grad_norm": 0.80078125,
+      "learning_rate": 5.324790163538181e-06,
+      "loss": 1.39990234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0548,
+      "step": 66,
+      "tokens/total": 17137664,
+      "tokens/train_per_sec_per_gpu": 603.46,
+      "tokens/trainable": 17087020
+    },
+    {
+      "epoch": 2.1629327902240325,
+      "grad_norm": 0.9765625,
+      "learning_rate": 5.208114835055157e-06,
+      "loss": 1.4246826171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.15654,
+      "step": 67,
+      "tokens/total": 17399808,
+      "tokens/train_per_sec_per_gpu": 608.36,
+      "tokens/trainable": 17348772
+    },
+    {
+      "epoch": 2.195519348268839,
+      "grad_norm": 0.9375,
+      "learning_rate": 5.0916362574280864e-06,
+      "loss": 1.467041015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.33638,
+      "step": 68,
+      "tokens/total": 17661952,
+      "tokens/train_per_sec_per_gpu": 604.98,
+      "tokens/trainable": 17610348
+    },
+    {
+      "epoch": 2.2281059063136457,
+      "grad_norm": 0.99609375,
+      "learning_rate": 4.975432945301085e-06,
+      "loss": 1.41259765625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.10661,
+      "step": 69,
+      "tokens/total": 17924096,
+      "tokens/train_per_sec_per_gpu": 606.09,
+      "tokens/trainable": 17871962
+    },
+    {
+      "epoch": 2.2606924643584523,
+      "grad_norm": 1.015625,
+      "learning_rate": 4.859583227770218e-06,
+      "loss": 1.400146484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.05579,
+      "step": 70,
+      "tokens/total": 18186240,
+      "tokens/train_per_sec_per_gpu": 609.55,
+      "tokens/trainable": 18133542
+    },
+    {
+      "epoch": 2.293279022403259,
+      "grad_norm": 0.80078125,
+      "learning_rate": 4.744165195584258e-06,
+      "loss": 1.399658203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.05381,
+      "step": 71,
+      "tokens/total": 18448384,
+      "tokens/train_per_sec_per_gpu": 608.83,
+      "tokens/trainable": 18395176
+    },
+    {
+      "epoch": 2.325865580448065,
+      "grad_norm": 0.84765625,
+      "learning_rate": 4.6292566485061015e-06,
+      "loss": 1.3804931640625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.97686,
+      "step": 72,
+      "tokens/total": 18710528,
+      "tokens/train_per_sec_per_gpu": 606.42,
+      "tokens/trainable": 18656664
+    },
+    {
+      "epoch": 2.3584521384928716,
+      "grad_norm": 0.83203125,
+      "learning_rate": 4.514935042870324e-06,
+      "loss": 1.525146484375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.59582,
+      "step": 73,
+      "tokens/total": 18972672,
+      "tokens/train_per_sec_per_gpu": 607.41,
+      "tokens/trainable": 18918236
+    },
+    {
+      "epoch": 2.391038696537678,
+      "grad_norm": 0.90234375,
+      "learning_rate": 4.40127743937224e-06,
+      "loss": 1.388916015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0105,
+      "step": 74,
+      "tokens/total": 19234816,
+      "tokens/train_per_sec_per_gpu": 608.52,
+      "tokens/trainable": 19179956
+    },
+    {
+      "epoch": 2.423625254582485,
+      "grad_norm": 0.83984375,
+      "learning_rate": 4.288360451123646e-06,
+      "loss": 1.4122314453125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.10511,
+      "step": 75,
+      "tokens/total": 19496960,
+      "tokens/train_per_sec_per_gpu": 606.59,
+      "tokens/trainable": 19441422
+    },
+    {
+      "epoch": 2.4562118126272914,
+      "grad_norm": 0.828125,
+      "learning_rate": 4.1762601920102675e-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": 76,
+      "tokens/total": 19759104,
+      "tokens/train_per_sec_per_gpu": 608.18,
+      "tokens/trainable": 19702864
+    },
+    {
+      "epoch": 2.4887983706720975,
+      "grad_norm": 0.85546875,
+      "learning_rate": 4.065052225385717e-06,
+      "loss": 1.35888671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.89186,
+      "step": 77,
+      "tokens/total": 20021248,
+      "tokens/train_per_sec_per_gpu": 606.93,
+      "tokens/trainable": 19964240
+    },
+    {
+      "epoch": 2.521384928716904,
+      "grad_norm": 0.828125,
+      "learning_rate": 3.954811513136554e-06,
+      "loss": 1.4130859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.10861,
+      "step": 78,
+      "tokens/total": 20283392,
+      "tokens/train_per_sec_per_gpu": 608.31,
+      "tokens/trainable": 20225736
+    },
+    {
+      "epoch": 2.5539714867617107,
+      "grad_norm": 0.80859375,
+      "learning_rate": 3.84561236515276e-06,
+      "loss": 1.459716796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.30474,
+      "step": 79,
+      "tokens/total": 20545536,
+      "tokens/train_per_sec_per_gpu": 605.24,
+      "tokens/trainable": 20487396
+    },
+    {
+      "epoch": 2.5865580448065173,
+      "grad_norm": 0.8515625,
+      "learning_rate": 3.7375283892377344e-06,
+      "loss": 1.38037109375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.97638,
+      "step": 80,
+      "tokens/total": 20807680,
+      "tokens/train_per_sec_per_gpu": 610.44,
+      "tokens/trainable": 20748996
+    },
+    {
+      "epoch": 2.619144602851324,
+      "grad_norm": 0.8515625,
+      "learning_rate": 3.630632441491512e-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": 81,
+      "tokens/total": 21069824,
+      "tokens/train_per_sec_per_gpu": 607.23,
+      "tokens/trainable": 21010468
+    },
+    {
+      "epoch": 2.6517311608961305,
+      "grad_norm": 0.828125,
+      "learning_rate": 3.5249965772007e-06,
+      "loss": 1.359130859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.89281,
+      "step": 82,
+      "tokens/total": 21331968,
+      "tokens/train_per_sec_per_gpu": 608.7,
+      "tokens/trainable": 21271952
+    },
+    {
+      "epoch": 2.684317718940937,
+      "grad_norm": 0.859375,
+      "learning_rate": 3.4206920022682173e-06,
+      "loss": 1.48974609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.43597,
+      "step": 83,
+      "tokens/total": 21594112,
+      "tokens/train_per_sec_per_gpu": 605.54,
+      "tokens/trainable": 21533380
+    },
+    {
+      "epoch": 2.716904276985743,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.3177890252155755e-06,
+      "loss": 1.44140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.22664,
+      "step": 84,
+      "tokens/total": 21856256,
+      "tokens/train_per_sec_per_gpu": 605.84,
+      "tokens/trainable": 21794716
+    },
+    {
+      "epoch": 2.74949083503055,
+      "grad_norm": 1.53125,
+      "learning_rate": 3.2163570097900497e-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": 85,
+      "tokens/total": 22118400,
+      "tokens/train_per_sec_per_gpu": 575.67,
+      "tokens/trainable": 22056232
+    },
+    {
+      "epoch": 2.7820773930753564,
+      "grad_norm": 0.8125,
+      "learning_rate": 3.116464328208708e-06,
+      "loss": 1.355224609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.87763,
+      "step": 86,
+      "tokens/total": 22380544,
+      "tokens/train_per_sec_per_gpu": 610.67,
+      "tokens/trainable": 22317500
+    },
+    {
+      "epoch": 2.814663951120163,
+      "grad_norm": 0.79296875,
+      "learning_rate": 3.0181783150707827e-06,
+      "loss": 1.3614501953125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.90185,
+      "step": 87,
+      "tokens/total": 22642688,
+      "tokens/train_per_sec_per_gpu": 607.51,
+      "tokens/trainable": 22579092
+    },
+    {
+      "epoch": 2.8472505091649696,
+      "grad_norm": 0.8125,
+      "learning_rate": 2.921565221969492e-06,
+      "loss": 1.36328125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.909,
+      "step": 88,
+      "tokens/total": 22904832,
+      "tokens/train_per_sec_per_gpu": 609.32,
+      "tokens/trainable": 22840240
+    },
+    {
+      "epoch": 2.8798370672097757,
+      "grad_norm": 0.87109375,
+      "learning_rate": 2.8266901728338526e-06,
+      "loss": 1.4312744140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.18403,
+      "step": 89,
+      "tokens/total": 23166976,
+      "tokens/train_per_sec_per_gpu": 608.0,
+      "tokens/trainable": 23101576
+    },
+    {
+      "epoch": 2.9124236252545828,
+      "grad_norm": 0.8515625,
+      "learning_rate": 2.7336171200306467e-06,
+      "loss": 1.419921875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.1368,
+      "step": 90,
+      "tokens/total": 23429120,
+      "tokens/train_per_sec_per_gpu": 609.24,
+      "tokens/trainable": 23362986
+    },
+    {
+      "epoch": 2.945010183299389,
+      "grad_norm": 0.83984375,
+      "learning_rate": 2.6424088012560766e-06,
+      "loss": 1.41015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0966,
+      "step": 91,
+      "tokens/total": 23691264,
+      "tokens/train_per_sec_per_gpu": 605.2,
+      "tokens/trainable": 23624496
+    },
+    {
+      "epoch": 2.9775967413441955,
+      "grad_norm": 0.8046875,
+      "learning_rate": 2.5531266972462176e-06,
+      "loss": 1.43603515625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.20399,
+      "step": 92,
+      "tokens/total": 23953408,
+      "tokens/train_per_sec_per_gpu": 602.63,
+      "tokens/trainable": 23885368
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 1.7734375,
+      "learning_rate": 2.4658309903347196e-06,
+      "loss": 1.438232421875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.21324,
+      "step": 93,
+      "tokens/total": 24133632,
+      "tokens/train_per_sec_per_gpu": 691.84,
+      "tokens/trainable": 24060864
+    },
+    {
+      "epoch": 3.0325865580448066,
+      "grad_norm": 0.90234375,
+      "learning_rate": 2.380580523885751e-06,
+      "loss": 1.39013671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0154,
+      "step": 94,
+      "tokens/total": 24395776,
+      "tokens/train_per_sec_per_gpu": 603.49,
+      "tokens/trainable": 24322588
+    },
+    {
+      "epoch": 3.065173116089613,
+      "grad_norm": 3.65625,
+      "learning_rate": 2.29743276262948e-06,
+      "loss": 1.481201171875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.39823,
+      "step": 95,
+      "tokens/total": 24657920,
+      "tokens/train_per_sec_per_gpu": 606.29,
+      "tokens/trainable": 24584084
+    },
+    {
+      "epoch": 3.0977596741344193,
+      "grad_norm": 0.83203125,
+      "learning_rate": 2.2164437539268652e-06,
+      "loss": 1.4517822265625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.27072,
+      "step": 96,
+      "tokens/total": 24920064,
+      "tokens/train_per_sec_per_gpu": 609.17,
+      "tokens/trainable": 24845644
+    },
+    {
+      "epoch": 3.130346232179226,
+      "grad_norm": 0.8203125,
+      "learning_rate": 2.1376680899898415e-06,
+      "loss": 1.38330078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.98804,
+      "step": 97,
+      "tokens/total": 25182208,
+      "tokens/train_per_sec_per_gpu": 604.06,
+      "tokens/trainable": 25107304
+    },
+    {
+      "epoch": 3.1629327902240325,
+      "grad_norm": 1.0078125,
+      "learning_rate": 2.0611588710823797e-06,
+      "loss": 1.40869140625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0906,
+      "step": 98,
+      "tokens/total": 25444352,
+      "tokens/train_per_sec_per_gpu": 609.04,
+      "tokens/trainable": 25369056
+    },
+    {
+      "epoch": 3.195519348268839,
+      "grad_norm": 0.87890625,
+      "learning_rate": 1.986967669727224e-06,
+      "loss": 1.45068359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.26603,
+      "step": 99,
+      "tokens/total": 25706496,
+      "tokens/train_per_sec_per_gpu": 606.62,
+      "tokens/trainable": 25630632
+    },
+    {
+      "epoch": 3.2281059063136457,
+      "grad_norm": 0.796875,
+      "learning_rate": 1.9151444959424383e-06,
+      "loss": 1.3984375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.04887,
+      "step": 100,
+      "tokens/total": 25968640,
+      "tokens/train_per_sec_per_gpu": 608.79,
+      "tokens/trainable": 25892246
+    },
+    {
+      "epoch": 3.2606924643584523,
+      "grad_norm": 0.80859375,
+      "learning_rate": 1.8457377635311763e-06,
+      "loss": 1.3865966796875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.00121,
+      "step": 101,
+      "tokens/total": 26230784,
+      "tokens/train_per_sec_per_gpu": 609.95,
+      "tokens/trainable": 26153826
+    },
+    {
+      "epoch": 3.293279022403259,
+      "grad_norm": 0.796875,
+      "learning_rate": 1.7787942574474215e-06,
+      "loss": 1.3873291015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.00414,
+      "step": 102,
+      "tokens/total": 26492928,
+      "tokens/train_per_sec_per_gpu": 608.64,
+      "tokens/trainable": 26415460
+    },
+    {
+      "epoch": 3.325865580448065,
+      "grad_norm": 0.79296875,
+      "learning_rate": 1.7143591022596846e-06,
+      "loss": 1.36865234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.93005,
+      "step": 103,
+      "tokens/total": 26755072,
+      "tokens/train_per_sec_per_gpu": 608.26,
+      "tokens/trainable": 26676948
+    },
+    {
+      "epoch": 3.3584521384928716,
+      "grad_norm": 0.92578125,
+      "learning_rate": 1.6524757317339102e-06,
+      "loss": 1.5140380859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.54505,
+      "step": 104,
+      "tokens/total": 27017216,
+      "tokens/train_per_sec_per_gpu": 609.74,
+      "tokens/trainable": 26938520
+    },
+    {
+      "epoch": 3.391038696537678,
+      "grad_norm": 2.28125,
+      "learning_rate": 1.593185859556103e-06,
+      "loss": 1.376708984375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.96184,
+      "step": 105,
+      "tokens/total": 27279360,
+      "tokens/train_per_sec_per_gpu": 610.52,
+      "tokens/trainable": 27200240
+    },
+    {
+      "epoch": 3.423625254582485,
+      "grad_norm": 0.89453125,
+      "learning_rate": 1.5365294512144114e-06,
+      "loss": 1.4033203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.06869,
+      "step": 106,
+      "tokens/total": 27541504,
+      "tokens/train_per_sec_per_gpu": 609.81,
+      "tokens/trainable": 27461706
+    },
+    {
+      "epoch": 3.4562118126272914,
+      "grad_norm": 0.83203125,
+      "learning_rate": 1.4825446970596136e-06,
+      "loss": 1.4375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.21016,
+      "step": 107,
+      "tokens/total": 27803648,
+      "tokens/train_per_sec_per_gpu": 608.6,
+      "tokens/trainable": 27723148
+    },
+    {
+      "epoch": 3.4887983706720975,
+      "grad_norm": 0.81640625,
+      "learning_rate": 1.4312679865621742e-06,
+      "loss": 1.3499755859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.85733,
+      "step": 108,
+      "tokens/total": 28065792,
+      "tokens/train_per_sec_per_gpu": 607.88,
+      "tokens/trainable": 27984524
+    },
+    {
+      "epoch": 3.521384928716904,
+      "grad_norm": 2.453125,
+      "learning_rate": 1.382733883783211e-06,
+      "loss": 1.404052734375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.07167,
+      "step": 109,
+      "tokens/total": 28327936,
+      "tokens/train_per_sec_per_gpu": 611.02,
+      "tokens/trainable": 28246020
+    },
+    {
+      "epoch": 3.5539714867617107,
+      "grad_norm": 1.046875,
+      "learning_rate": 1.3369751040759236e-06,
+      "loss": 1.45263671875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.27437,
+      "step": 110,
+      "tokens/total": 28590080,
+      "tokens/train_per_sec_per_gpu": 607.24,
+      "tokens/trainable": 28507680
+    },
+    {
+      "epoch": 3.5865580448065173,
+      "grad_norm": 0.7890625,
+      "learning_rate": 1.2940224920331707e-06,
+      "loss": 1.3724365234375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.94495,
+      "step": 111,
+      "tokens/total": 28852224,
+      "tokens/train_per_sec_per_gpu": 611.79,
+      "tokens/trainable": 28769280
+    },
+    {
+      "epoch": 3.619144602851324,
+      "grad_norm": 0.8046875,
+      "learning_rate": 1.2539050006960814e-06,
+      "loss": 1.3863525390625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.00023,
+      "step": 112,
+      "tokens/total": 29114368,
+      "tokens/train_per_sec_per_gpu": 610.39,
+      "tokens/trainable": 29030752
+    },
+    {
+      "epoch": 3.6517311608961305,
+      "grad_norm": 0.8203125,
+      "learning_rate": 1.2166496720376874e-06,
+      "loss": 1.35302734375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.86912,
+      "step": 113,
+      "tokens/total": 29376512,
+      "tokens/train_per_sec_per_gpu": 610.15,
+      "tokens/trainable": 29292236
+    },
+    {
+      "epoch": 3.684317718940937,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.1822816187347625e-06,
+      "loss": 1.483154296875,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.40682,
+      "step": 114,
+      "tokens/total": 29638656,
+      "tokens/train_per_sec_per_gpu": 607.45,
+      "tokens/trainable": 29553664
+    },
+    {
+      "epoch": 3.716904276985743,
+      "grad_norm": 0.8046875,
+      "learning_rate": 1.1508240072401336e-06,
+      "loss": 1.43505859375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.19989,
+      "step": 115,
+      "tokens/total": 29900800,
+      "tokens/train_per_sec_per_gpu": 607.15,
+      "tokens/trainable": 29815000
+    },
+    {
+      "epoch": 3.74949083503055,
+      "grad_norm": 0.8359375,
+      "learning_rate": 1.1222980421668874e-06,
+      "loss": 1.382568359375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.98512,
+      "step": 116,
+      "tokens/total": 30162944,
+      "tokens/train_per_sec_per_gpu": 607.18,
+      "tokens/trainable": 30076516
+    },
+    {
+      "epoch": 3.7820773930753564,
+      "grad_norm": 0.90625,
+      "learning_rate": 1.0967229519949833e-06,
+      "loss": 1.350830078125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.86063,
+      "step": 117,
+      "tokens/total": 30425088,
+      "tokens/train_per_sec_per_gpu": 598.63,
+      "tokens/trainable": 30337784
+    },
+    {
+      "epoch": 3.814663951120163,
+      "grad_norm": 0.765625,
+      "learning_rate": 1.0741159761099294e-06,
+      "loss": 1.356689453125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.88332,
+      "step": 118,
+      "tokens/total": 30687232,
+      "tokens/train_per_sec_per_gpu": 596.74,
+      "tokens/trainable": 30599376
+    },
+    {
+      "epoch": 3.8472505091649696,
+      "grad_norm": 0.89453125,
+      "learning_rate": 1.054492353182237e-06,
+      "loss": 1.3583984375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 3.88996,
+      "step": 119,
+      "tokens/total": 30949376,
+      "tokens/train_per_sec_per_gpu": 610.82,
+      "tokens/trainable": 30860524
+    },
+    {
+      "epoch": 3.8798370672097757,
+      "grad_norm": 0.83203125,
+      "learning_rate": 1.0378653108955017e-06,
+      "loss": 1.42724609375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.16721,
+      "step": 120,
+      "tokens/total": 31211520,
+      "tokens/train_per_sec_per_gpu": 609.6,
+      "tokens/trainable": 31121860
+    },
+    {
+      "epoch": 3.9124236252545828,
+      "grad_norm": 0.8125,
+      "learning_rate": 1.0242460570300241e-06,
+      "loss": 1.416015625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.12067,
+      "step": 121,
+      "tokens/total": 31473664,
+      "tokens/train_per_sec_per_gpu": 611.16,
+      "tokens/trainable": 31383270
+    },
+    {
+      "epoch": 3.945010183299389,
+      "grad_norm": 0.8203125,
+      "learning_rate": 1.01364377190799e-06,
+      "loss": 1.407958984375,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.0876,
+      "step": 122,
+      "tokens/total": 31735808,
+      "tokens/train_per_sec_per_gpu": 606.72,
+      "tokens/trainable": 31644780
+    },
+    {
+      "epoch": 3.9775967413441955,
+      "grad_norm": 0.7890625,
+      "learning_rate": 1.0060656022052966e-06,
+      "loss": 1.4337158203125,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.19426,
+      "step": 123,
+      "tokens/total": 31997952,
+      "tokens/train_per_sec_per_gpu": 604.37,
+      "tokens/trainable": 31905652
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 1.34375,
+      "learning_rate": 1.0015166561341943e-06,
+      "loss": 1.433837890625,
+      "memory/device_reserved (GiB)": 33.29,
+      "memory/max_active (GiB)": 27.26,
+      "memory/max_allocated (GiB)": 27.26,
+      "ppl": 4.19477,
+      "step": 124,
+      "tokens/total": 32178176,
+      "tokens/train_per_sec_per_gpu": 692.73,
+      "tokens/trainable": 32081148
+    }
+  ],
+  "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": 6.99703182078509e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

midtrain_4epoch/charter/checkpoint-124/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb7030dca2b773963c2c283043e1cdf30ee17a762b1bfe4cdb95eef9ea796615
+size 7377